Filtering invalid utf8












46















I have a text file in an unknown or mixed encoding. I want to see the lines that contain a byte sequence that is not valid UTF-8 (by piping the text file into some program). Equivalently, I want to filter out the lines that are valid UTF-8. In other words, I'm looking for grep [notutf8].



An ideal solution would be portable, short and generalizable to other encodings, but if you feel the best way is to bake in the definition of UTF-8, go ahead.






command-line text-processing character-encoding unicode







share|improve this question























  • See also keithdevens.com/weblog/archive/2004/Jun/29/UTF-8.regex for a possible regex.

    – Mikel
    Jan 27 '11 at 6:54











  • @Mikel: or unix.stackexchange.com/questions/6460/… … I was hoping for something less clumsly.

    – Gilles
    Jan 27 '11 at 8:18
















46















I have a text file in an unknown or mixed encoding. I want to see the lines that contain a byte sequence that is not valid UTF-8 (by piping the text file into some program). Equivalently, I want to filter out the lines that are valid UTF-8. In other words, I'm looking for grep [notutf8].



An ideal solution would be portable, short and generalizable to other encodings, but if you feel the best way is to bake in the definition of UTF-8, go ahead.






command-line text-processing character-encoding unicode







share|improve this question























  • See also keithdevens.com/weblog/archive/2004/Jun/29/UTF-8.regex for a possible regex.

    – Mikel
    Jan 27 '11 at 6:54











  • @Mikel: or unix.stackexchange.com/questions/6460/… … I was hoping for something less clumsly.

    – Gilles
    Jan 27 '11 at 8:18














46












46








46


19






I have a text file in an unknown or mixed encoding. I want to see the lines that contain a byte sequence that is not valid UTF-8 (by piping the text file into some program). Equivalently, I want to filter out the lines that are valid UTF-8. In other words, I'm looking for grep [notutf8].



An ideal solution would be portable, short and generalizable to other encodings, but if you feel the best way is to bake in the definition of UTF-8, go ahead.






command-line text-processing character-encoding unicode







share|improve this question














I have a text file in an unknown or mixed encoding. I want to see the lines that contain a byte sequence that is not valid UTF-8 (by piping the text file into some program). Equivalently, I want to filter out the lines that are valid UTF-8. In other words, I'm looking for grep [notutf8].



An ideal solution would be portable, short and generalizable to other encodings, but if you feel the best way is to bake in the definition of UTF-8, go ahead.






command-line text-processing character-encoding unicode




command-line text-processing character-encoding unicode






share|improve this question













share|improve this question











share|improve this question




share|improve this question










asked Jan 27 '11 at 0:13









GillesGilles

533k12810721594




533k12810721594













  • See also keithdevens.com/weblog/archive/2004/Jun/29/UTF-8.regex for a possible regex.

    – Mikel
    Jan 27 '11 at 6:54











  • @Mikel: or unix.stackexchange.com/questions/6460/… … I was hoping for something less clumsly.

    – Gilles
    Jan 27 '11 at 8:18



















  • See also keithdevens.com/weblog/archive/2004/Jun/29/UTF-8.regex for a possible regex.

    – Mikel
    Jan 27 '11 at 6:54











  • @Mikel: or unix.stackexchange.com/questions/6460/… … I was hoping for something less clumsly.

    – Gilles
    Jan 27 '11 at 8:18

















See also keithdevens.com/weblog/archive/2004/Jun/29/UTF-8.regex for a possible regex.

– Mikel
Jan 27 '11 at 6:54





See also keithdevens.com/weblog/archive/2004/Jun/29/UTF-8.regex for a possible regex.

– Mikel
Jan 27 '11 at 6:54













@Mikel: or unix.stackexchange.com/questions/6460/… … I was hoping for something less clumsly.

– Gilles
Jan 27 '11 at 8:18





@Mikel: or unix.stackexchange.com/questions/6460/… … I was hoping for something less clumsly.

– Gilles
Jan 27 '11 at 8:18










6 Answers
6






active

oldest

votes


















30














If you want to use grep, you can do:



grep -axv '.*' file


in UTF-8 locales to get the lines that have at least an invalid UTF-8 sequence (this works with GNU Grep at least).






share|improve this answer


























  • Except for -a, that's required to work by POSIX. However GNU grep at least fails to spot the UTF-8 encoded UTF-16 surrogate non-characters or codepoints above 0x10FFFF.

    – Stéphane Chazelas
    Dec 19 '14 at 23:10






  • 1





    @StéphaneChazelas Conversely, the -a is needed by GNU grep (which isn't POSIX compliant, I assume). Concerning, the surrogate area and the codepoints above 0x10FFFF, this is a bug then (which could explain that). For this, adding -P should work with GNU grep 2.21 (but is slow); it is buggy at least in Debian grep/2.20-4.

    – vinc17
    Dec 19 '14 at 23:37











  • Sorry, my bad, the behaviour is unspecified in POSIX since grep is a text utility (only expected to work on text input), so I suppose GNU grep's behaviour is as valid as any here.

    – Stéphane Chazelas
    Dec 19 '14 at 23:50













  • @StéphaneChazelas I confirm that POSIX says: "The input files shall be text files." (though not in the description part, which is a bit misleading). This also means that in case of invalid sequences, the behavior is undefined by POSIX. Hence the need to know the implementation, such as GNU grep (whose intent is to regard invalid sequences as non-matching), and possible bugs.

    – vinc17
    Dec 20 '14 at 0:12






  • 1





    I'm switching the accepted answer to this one (sorry, Peter.O because it's simple and works well for my primary use case, which is a heuristic to distinguish UTF-8 from other common encodings (especially 8-bit encodings). Stéphane Chazelas and Peter.O provide more accurate answers in terms of UTF-8 compliance.

    – Gilles
    Dec 23 '14 at 2:11



















29














I think you probably want iconv. It's for converting between codesets and supports an absurd number of formats. For example, to strip anything not valid in UTF-8 you could use:



iconv -c -t UTF-8 < input.txt > output.txt



Without the -c option it'll report problems in converting to stderr, so with process direction could you save a list of these. Another way would be to strip the non-UTF8 stuff and then



diff input.txt output.txt



for a list of where changes were made.






share|improve this answer
























  • Ok, that's iconv -c -t UTF-8 <input.txt | diff input.txt - | sed -ne 's/^< //p'. It won't work as a pipeline, though, since you need to read the input twice (no, tee won't do, it might block depending on how much buffering iconv and diff do).

    – Gilles
    Jan 27 '11 at 8:22






  • 2





    Random note: input & output may not be the same file or you will end up with an empty file

    – drahnr
    Aug 24 '13 at 8:02






  • 1





    Or use process substitution if your shell supports it diff <(iconv -c -t UTF-8 <input.txt) input.txt

    – Karl
    Nov 11 '15 at 23:41













  • How do you do this and make the output to the same file as the input. I just did this and got a blank file iconv -c -t UTF-8 < input.txt > input.txt

    – Costas Vrahimis
    Nov 8 '16 at 0:24






  • 1





    Thanks.. This allow restoring broken utf-8 postgresql dump , but not discarding valid utf-8

    – Superbiji
    Dec 16 '16 at 10:29



















19














Edit: I've fixed a typo-bug in the regex.. It needed a 'x80` not 80.



The regex to filter out invalid UTF-8 forms, for strict adherance to UTF-8, is as follows 



perl -l -ne '/

 ^( ([x00-x7F])              # 1-byte pattern

   |([xC2-xDF][x80-xBF])   # 2-byte pattern

   |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF])) # 3-byte pattern

   |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))       # 4-byte pattern

  )*$ /x or print'


Output (of key lines.from Test 1):



Codepoint

=========  

00001000  Test=1 mode=strict               valid,invalid,fail=(1000,0,0)          

0000E000  Test=1 mode=strict               valid,invalid,fail=(D800,800,0)          

0010FFFF  mode=strict  test-return=(0,0)   valid,invalid,fail=(10F800,800,0)



Q. How does one create test data to test a regex which filters invalid Unicode ?

A. Create your own UTF-8 test algorithm, and break it's rules...

Catch-22.. But then, how do you then test your test algorithm?



The regex, above, has been tested (using iconv as the reference) for every integer value from 0x00000 to 0x10FFFF.. This upper value being the maximum integer value of a Unicode Codepoint





  • In November 2003 UTF-8 was restricted by RFC 3629 to four bytes covering only the range U+0000 to U+10FFFF, in order to match the constraints of the UTF-16 character encoding.`


According to this wikipedia UTF-8 page,.




  • UTF-8 encodes each of the 1,112,064 code points in the Unicode character set, using one to four 8-bit bytes


This numeber (1,112,064) equates to a range 0x000000 to 0x10F7FF, which is 0x0800 shy of the actual maximum integer-value for the highest Unicode Codepoint: 0x10FFFF



This block of integers is missing from the Unicode Codepoints spectrum, because of the need for the UTF-16 encoding to step beyond its original design intent via a system called surrogate pairs. A block of 0x0800 integers has been reserved to be used by UTF-16.. This block spans the range0x00D800 to 0x00DFFF. None of these inteters are legal Unicode values, and are therefore invalid UTF-8 values.



In Test 1, the regex has been tested against every number in the range of Unicode Codepoints, and it matches exectly the results of iconv .. ie. 0x010F7FF valid values, and 0x000800 invalid values.



However, the issue now arises of, *How does the regex handle Out-Of-Range UTF-8 Value; above 0x010FFFF (UTF-8 can extend to 6 bytes, with a maximum integer value of 0x7FFFFFFF?

To generate the necessary *non-unicode UTF-8 byte values, I've used the following command: 



  perl -C -e 'print chr 0x'$hexUTF32BE


To test their validity (in some fashion), I've used Gilles' UTF-8 regex... 



  perl -l -ne '/

   ^( [00-177]                 # 1-byte pattern

     |[300-337][200-277]      # 2-byte pattern

     |[340-357][200-277]{2}   # 3-byte pattern

     |[360-367][200-277]{3}   # 4-byte pattern

     |[370-373][200-277]{4}   # 5-byte pattern

     |[374-375][200-277]{5}   # 6-byte pattern

    )*$ /x or print'


The output of 'perl's print chr' matches the filtering of Gilles' regex.. One reinforces the validity of the other..
I can't use iconv because it only handles the valid-Unicode Standard subset of the broader (original) UTF-8 standard...



The nunbers involved are rather large, so I've tested top-of-range, bottom-of-range, and several scans stepping by increments such as, 11111, 13579, 33333, 53441... The results all match, so now all that remains is to test the regex against these out-of-range UTF-8-style values (invalid for Unicode, and therefore also invalid for strict UTF-8 itself) ..




Here are the test modules: 



[[ "$(locale charmap)" != "UTF-8" ]] && { echo "ERROR: locale must be UTF-8, but it is $(locale charmap)"; exit 1; }



# Testing the UTF-8 regex

#

# Tests to check that the observed byte-ranges (above) have

#  been  accurately observed and included in the test code and final regex. 

# =========================================================================

: 2 bytes; B2=0 #  run-test=1   do-not-test=0

: 3 bytes; B3=0 #  run-test=1   do-not-test=0

: 4 bytes; B4=0 #  run-test=1   do-not-test=0 



:   regex; Rx=1 #  run-test=1   do-not-test=0



           ((strict=16)); mode[$strict]=strict # iconv -f UTF-16BE  then iconv -f UTF-32BE beyond 0xFFFF)

           ((   lax=32)); mode[$lax]=lax       # iconv -f UTF-32BE  only)



          # modebits=$strict

                  # UTF-8, in relation to UTF-16 has invalid values

                  # modebits=$strict automatically shifts to modebits=$lax

                  # when the tested integer exceeds 0xFFFF

          # modebits=$lax 

                  # UTF-8, in relation to UTF-32, has no restrictione





           # Test 1 Sequentially tests a range of Big-Endian integers

           #      * Unicode Codepoints are a subset ofBig-Endian integers            

           #        ( based on 'iconv' -f UTF-32BE -f UTF-8 )    

           # Note: strict UTF-8 has a few quirks because of UTF-16

                    #    Set modebits=16 to "strictly" test the low range



             Test=1; modebits=$strict

           # Test=2; modebits=$lax

           # Test=3

              mode3wlo=$(( 1*4)) # minimum chars * 4 ( '4' is for UTF-32BE )

              mode3whi=$((10*4)) # minimum chars * 4 ( '4' is for UTF-32BE )





#########################################################################  



# 1 byte  UTF-8 values: Nothing to do; no complexities.



#########################################################################



#  2 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B2==1)) ; then  

  echo "# Test 2 bytes for Valid UTF-8 values: ie. values which are in range"

  # =========================================================================

  time 

  for d1 in {194..223} ;do

      #     bin       oct  hex  dec

      # lo  11000010  302   C2  194

      # hi  11011111  337   DF  223

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {128..191} ;do

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r  |

              iconv -f UTF-8 >/dev/null || { 

                echo "ERROR: Invalid UTF-8 found: ${B2b1}${B2b2}"; exit 20; }

          #

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 2 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  time 

  for d1 in {128..193} {224..255} ;do 

 #for d1 in {128..194} {224..255} ;do # force a valid UTF-8 (needs $B2b2) 

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {0..127} {192..255} ;do

     #for d2 in {0..128} {192..255} ;do # force a valid UTF-8 (needs $B2b1)

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r |

              iconv -f UTF-8 2>/dev/null && { 

                echo "ERROR: VALID UTF-8 found: ${B2b1}${B2b2}"; exit 21; }

          #

      done

  done

  echo

fi



#########################################################################



#  3 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B3==1)) ; then  

  echo "# Test 3 bytes for Valid UTF-8 values: ie. values which are in range"

  # ========================================================================

  time 

  for d1 in {224..239} ;do

      #     bin       oct  hex  dec

      # lo  11100000  340   E0  224

      # hi  11101111  357   EF  239

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {160..191})"

          #     bin       oct  hex  dec  

          # lo  10100000  240   A0  160  

          # hi  10111111  277   BF  191

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {128..159})"

          #     bin       oct  hex  dec  

          # lo  10000000  200   80  128  

          # hi  10011111  237   9F  159

      else

          B3b2range="$(echo {128..191})"

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

      fi

      # 

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r  |

                  iconv -f UTF-8 >/dev/null || { 

                    echo "ERROR: Invalid UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 30; }

              #

          done

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 3 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  #

  # real     26m28.462s  

  # user     27m12.526s  | stepping by 2

  # sys      13m11.193s /

  #

  # real    239m00.836s 

  # user    225m11.108s  | stepping by 1

  # sys     120m00.538s /

  #

  time 

  for d1 in {128..223..1} {240..255..1} ;do 

 #for d1 in {128..224..1} {239..255..1} ;do # force a valid UTF-8 (needs $B2b2,$B3b3) 

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {0..159..1} {192..255..1})"

         #B3b2range="$(> {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {0..127..1} {160..255..1})"

         #B3b2range="$(echo {0..128..1} {160..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      else

          B3b2range="$(echo {0..127..1} {192..255..1})"

         #B3b2range="$(echo {0..128..1} {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      fi

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {0..127..1} {192..255..1} ;do

         #for d3 in {0..128..1} {192..255..1} ;do # force a valid UTF-8 (needs $B2b1)

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r |

                  iconv -f UTF-8 2>/dev/null && { 

                    echo "ERROR: VALID UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 31; }

              #

          done

      done

  done

  echo



fi



#########################################################################



#  Brute force testing in the Astral Plane will take a VERY LONG time..

#  Perhaps selective testing is more appropriate, now that the previous tests 

#     have panned out okay... 

#  

#  4 Byte  UTF-8 values:

if ((B4==1)) ; then  

  echo "# Test 4 bytes for Valid UTF-8 values: ie. values which are in range"

  # ==================================================================

  # real    58m18.531s 

  # user    56m44.317s  | 

  # sys     27m29.867s /

  time 

  for d1 in {240..244} ;do

      #     bin       oct  hex  dec

      # lo  11110000  360   F0  240

      # hi  11110100  364   F4  244  -- F4 encodes some values greater than 0x10FFFF;

      #                                    such a sequence is invalid.

      B4b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B4b1 == "F0" ]] ; then

        B4b2range="$(echo {144..191})" ## f0 90 80 80  to  f0 bf bf bf

        #     bin       oct  hex  dec          010000  --  03FFFF 

        # lo  10010000  220   90  144  

        # hi  10111111  277   BF  191

        #                            

      elif [[ $B4b1 == "F4" ]] ; then

        B4b2range="$(echo {128..143})" ## f4 80 80 80  to  f4 8f bf bf

        #     bin       oct  hex  dec          100000  --  10FFFF 

        # lo  10000000  200   80  128  

        # hi  10001111  217   8F  143  -- F4 encodes some values greater than 0x10FFFF;

        #                                    such a sequence is invalid.

      else

        B4b2range="$(echo {128..191})" ## fx 80 80 80  to  f3 bf bf bf

        #     bin       oct  hex  dec          0C0000  --  0FFFFF

        # lo  10000000  200   80  128          0A0000

        # hi  10111111  277   BF  191

      fi

      #

      for d2 in $B4b2range ;do

          B4b2=$(printf "%0.2X" $d2)

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B4b3=$(printf "%0.2X" $d3)

              echo "${B4b1} ${B4b2} ${B4b3} xx"

              #

              for d4 in {128..191} ;do

                  #     bin       oct  hex  dec

                  # lo  10000000  200   80  128

                  # hi  10111111  277   BF  191

                  B4b4=$(printf "%0.2X" $d4)

                  #

                  echo -n "${B4b1}${B4b2}${B4b3}${B4b4}" |

                    xxd -p -u -r  |

                      iconv -f UTF-8 >/dev/null || { 

                        echo "ERROR: Invalid UTF-8 found: ${B4b1}${B4b2}${B4b3}${B4b4}"; exit 40; }

                  #

              done

          done

      done

  done

  echo "# Test 4 bytes for Valid UTF-8 values: END"

  echo

fi



########################################################################

# There is no test (yet) for negated range values in the astral plane. #  

#                           (all negated range values must be invalid) #

#  I won't bother; This was mainly for me to ge the general feel of    #     

#   the tests, and the final test below should flush anything out..    #

# Traversing the intire UTF-8 range takes quite a while...             #

#   so no need to do it twice (albeit in a slightly different manner)  #

########################################################################



################################

### The construction of:    ####

###  The Regular Expression ####

###      (de-construction?) ####

################################



#     BYTE 1                BYTE 2       BYTE 3      BYTE 4 

# 1: [x00-x7F]

#    ===========

#    ([x00-x7F])

#

# 2: [xC2-xDF]           [x80-xBF]

#    =================================

#    ([xC2-xDF][x80-xBF])

# 

# 3: [xE0]                [xA0-xBF]  [x80-xBF]   

#    [xED]                [x80-x9F]  [x80-xBF]

#    [xE1-xECxEE-xEF]  [x80-xBF]  [x80-xBF]

#    ==============================================

#    ((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))

#

# 4  [xF0]                [x90-xBF]  [x80-xBF]  [x80-xBF]    

#    [xF1-xF3]           [x80-xBF]  [x80-xBF]  [x80-xBF]

#    [xF4]                [x80-x8F]  [x80-xBF]  [x80-xBF]

#    ===========================================================

#    ((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))

#

# The final regex

# ===============

# 1-4:  (([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))

# 4-1:  (((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|([xC2-xDF][x80-xBF])|([x00-x7F]))





#######################################################################

#  The final Test; for a single character (multi chars to follow)     #  

#   Compare the return code of 'iconv' against the 'regex'            #

#   for the full range of 0x000000 to 0x10FFFF                        #

#                                                                     #     

#  Note; this script has 3 modes:                                     #

#        Run this test TWICE, set each mode Manually!                 #     

#                                                                     #     

#     1. Sequentially test every value from 0x000000 to 0x10FFFF      #     

#     2. Throw a spanner into the works! Force random byte patterns   #     

#     2. Throw a spanner into the works! Force random longer strings  #     

#        ==============================                               #     

#                                                                     #     

#  Note: The purpose of this routine is to determine if there is any  #

#        difference how 'iconv' and 'regex' handle the same data      #  

#                                                                     #     

#######################################################################

if ((Rx==1)) ; then

  # real    191m34.826s

  # user    158m24.114s

  # sys      83m10.676s

  time { 

  invalCt=0

  validCt=0

   failCt=0

  decBeg=$((0x00110000)) # incement by decimal integer

  decMax=$((0x7FFFFFFF)) # incement by decimal integer

  # 

  for ((CPDec=decBeg;CPDec<=decMax;CPDec+=13247)) ;do

      ((D==1)) && echo "=========================================================="

      #

      # Convert decimal integer '$CPDec' to Hex-digits; 6-long  (dec2hex)

      hexUTF32BE=$(printf '%0.8Xn' $CPDec)  # hexUTF32BE



      # progress count  

      if (((CPDec%$((0x1000)))==0)) ;then

          ((Test>2)) && echo

          echo "$hexUTF32BE  Test=$Test mode=${mode[$modebits]}            "

      fi

      if   ((Test==1 || Test==2 ))

      then # Test 1. Sequentially test every value from 0x000000 to 0x10FFFF

          #

          if   ((Test==2)) ; then

              bits=32

              UTF8="$( perl -C -e 'print chr 0x'$hexUTF32BE |

                perl -l -ne '/^(  [00-177]

                                | [300-337][200-277]

                                | [340-357][200-277]{2}

                                | [360-367][200-277]{3}

                                | [370-373][200-277]{4}

                                | [374-375][200-277]{5}

                               )*$/x and print' |xxd -p )"

              UTF8="${UTF8%0a}"

              [[ -n "$UTF8" ]] 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=



          elif ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              bits=16

              UTF8="$( echo -n "${hexUTF32BE:4}" |

                xxd -p -u -r |

                  iconv -f UTF-16BE -t UTF-8 2>/dev/null)" 

                    && rcIco16=0 || rcIco16=1  

                       rcIco32=

          else

              bits=32

              UTF8="$( echo -n "$hexUTF32BE" |

                xxd -p -u -r |

                  iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=

          fi

          # echo "1 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((${rcIco16}${rcIco32}!=0)) ;then

              # 'iconv -f UTF-16BE' failed produce a reliable UTF-8

              if ((bits==16)) ;then

                  ((D==1)) &&           echo "bits-$bits rcIconv: error    $hexUTF32BE .. 'strict' failed, now trying 'lax'"

                  #  iconv failed to create a  'srict' UTF-8 so   

                  #      try UTF-32BE to get a   'lax' UTF-8 pattern    

                  UTF8="$( echo -n "$hexUTF32BE" |

                    xxd -p -u -r |

                      iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                        && rcIco32=0 || rcIco32=1

                  #echo "2 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

                  if ((rcIco32!=0)) ;then

                      ((D==1)) &&               echo -n "bits-$bits rcIconv: Cannot gen UTF-8 for: $hexUTF32BE"

                      rcIco32=1

                  fi

              fi

          fi

          # echo "3 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((rcIco16==0 || rcIco32==0)) ;then

              # 'strict(16)' OR 'lax(32)'... 'iconv' managed to generate a UTF-8 pattern  

                  ((D==1)) &&       echo -n "bits-$bits rcIconv: pattern* $hexUTF32BE"

                  ((D==1)) &&       if [[ $bits == "16" && $rcIco32 == "0" ]] ;then 

                  echo " .. 'lax' UTF-8 produced a pattern"

              else

                  echo

              fi

               # regex test

              if ((modebits==strict)) ;then

                 #rxOut="$(echo -n "$UTF8" |perl -l -ne '/^(([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))*$/ or print' )"

                                     rxOut="$(echo -n "$UTF8" |

                  perl -l -ne '/^( ([x00-x7F])             # 1-byte pattern

                                  |([xC2-xDF][x80-xBF])  # 2-byte pattern

                                  |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))  # 3-byte pattern

                                  |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))        # 4-byte pattern

                                 )*$ /x or print' )"

               else

                  if ((Test==2)) ;then

                      rx="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ and print')"

                      [[ "$UTF8" != "$rx" ]] && rxOut="$UTF8" || rxOut=

                      rx="$(echo -n "$rx" |sed -e "s/(..)/1 /g")"  

                  else 

                      rxOut="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ or print' )"

                  fi

              fi

              if [[ "$rxOut" == "" ]] ;then

                ((D==1)) &&           echo "        rcRegex: ok"

                  rcRegex=0

              else

                  ((D==1)) &&           echo -n "bits-$bits rcRegex: error    $hexUTF32BE .. 'strict' failed,"

                  ((D==1)) &&           if [[  "12" == *$Test* ]] ;then 

                                            echo # "  (codepoint) Test $Test" 

                                        else

                                            echo

                                        fi

                  rcRegex=1

              fi

          fi

          #

      elif [[ $Test == 2 ]]

      then # Test 2. Throw a randomizing spanner into the works! 

          #          Then test the  arbitary bytes ASIS

          #

          hexLineRand="$(echo -n "$hexUTF32BE" |

            sed -re "s/(.)(.)(.)(.)(.)(.)(.)(.)/1n2n3n4n5n6n7n8/" |

              sort -R |

                tr -d 'n')"

          # 

      elif [[ $Test == 3 ]]

      then # Test 3. Test single UTF-16BE bytes in the range 0x00000000 to 0x7FFFFFFF

          #

          echo "Test 3 is not properly implemented yet.. Exiting"

          exit 99 

      else

          echo "ERROR: Invalid mode"

          exit

      fi

      #

      #

      if ((Test==1 || Test=2)) ;then

          if ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              ((rcIconv=rcIco16))

          else

              ((rcIconv=rcIco32))

          fi

          if ((rcRegex!=rcIconv)) ;then

              [[ $Test != 1 ]] && echo

              if ((rcRegex==1)) ;then

                  echo "ERROR: 'regex' ok, but NOT 'iconv': ${hexUTF32BE} "

              else

                  echo "ERROR: 'iconv' ok, but NOT 'regex': ${hexUTF32BE} "

              fi

              ((failCt++));

          elif ((rcRegex!=0)) ;then

            # ((invalCt++)); echo -ne "$hexUTF32BE  exit-codes $${rcIco16}${rcIco32}=,$rcRegext: $(printf "%0.8Xn" $invalCt)t$hexLine$(printf "%$(((mode3whi*2)-${#hexLine}))s")r"

              ((invalCt++)) 

          else

              ((validCt++)) 

          fi

          if   ((Test==1)) ;then

              echo -ne "$hexUTF32BE "    "mode=${mode[$modebits]}  test-return=($rcIconv,$rcRegex)   valid,invalid,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))          r"

          else 

              echo -ne "$hexUTF32BE $rx mode=${mode[$modebits]} test-return=($rcIconv,$rcRegex)  val,inval,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))r"

          fi

      fi

  done

  } # End time

fi

exit





share|improve this answer


























  • The main problem with my regexp is that it allowed some forbidden sequences such as 300200 (really bad: that's code point 0 not expressed with a null byte!). I think your regexp rejects them correctly.

    – Gilles
    May 8 '11 at 12:38



















6





+100









I find uconv (in icu-devtools package in Debian) useful to inspect UTF-8 data:



$ print '\xE9 xe9 u20ac ud800udc00 U110000' |

    uconv --callback escape-c -t us

xE9 xE9 u20ac xEDxA0x80xEDxB0x80 xF4x90x80x80


(The xs help spotting the invalid characters (except for the false positive voluntarily introduced with a literal xE9 above)).



(plenty of other nice usages).






share|improve this answer


























  • I think recode can be used similarly - except that I think it should fail if asked to translate an invalid multibyte sequence. I'm not sure though; it won't fail for print...|recode u8..u8/x4 for example (which just does a hexdump as you do above) because it doesn't do anything but iconv data data, but it does fail like recode u8..u2..u8/x4 because it translates then prints. But I don't know enough about it to be sure - and there are a lot of possibilities.

    – mikeserv
    Dec 20 '14 at 19:29











  • If I have a file, say, test.txt. How should I suppose to find the invalid character using your solution? What does us in your code mean?

    – jdhao
    Dec 26 '17 at 7:51











  • @Hao, us means United States, that is short for ASCII. It converts the input into a ASCII one where the non-ASCII characters are converted to uXXXX notation and the non-characters to xXX.

    – Stéphane Chazelas
    Dec 26 '17 at 14:19











  • Where should I put my file to use your script? Is the last line in the code block the output of your code? It is a little confusing to me.

    – jdhao
    Dec 26 '17 at 14:27



















3














Python has had a built-in unicode function since version 2.0.



#!/usr/bin/env python2

import sys

for line in sys.stdin:

    try:

        unicode(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.write(line)


In Python 3, unicode has been folded into str. It needs to be passed a bytes-like object, here the underlying buffer objects for the standard descriptors.



#!/usr/bin/env python3

import sys

for line in sys.stdin.buffer:

    try:

        str(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.buffer.write(line)





share|improve this answer
























  • The python 2 one fails to flag UTF-8 encoded UTF-16 surrogate non-characters (at least with 2.7.6).

    – Stéphane Chazelas
    Dec 19 '14 at 23:12











  • @StéphaneChazelas Dammit. Thanks. I've only run nominal tests so far, I'll run Peter's test battery later.

    – Gilles
    Dec 19 '14 at 23:14



















0














I came across similar problem (detail in "Context" section) and arrived with following ftfy_line_by_line.py solution:



#!/usr/bin/env python3

import ftfy, sys

with open(sys.argv[1], mode='rt', encoding='utf8', errors='replace') as f:

  for line in f:

    sys.stdout.buffer.write(ftfy.fix_text(line).encode('utf8', 'replace'))

    #print(ftfy.fix_text(line).rstrip().decode(encoding="utf-8", errors="replace"))


Using encode+replace + ftfy to auto-fix Mojibake and other corrections.



Context



I've collected >10GiB CSV of basic filesystem metadata using following gen_basic_files_metadata.csv.sh script, running essentially:



find "${path}" -type f -exec stat --format="%i,%Y,%s,${hostname},%m,%n" "{}" ;


The trouble I had was with inconsistent encoding of filenames across file systems, causing UnicodeDecodeError when processing further with python applications (csvsql to be more specific).



Therefore I applied above ftfy script, and it took



Please note ftfy is pretty slow, processing those >10GiB took:



real    147m35.182s

user    146m14.329s

sys     2m8.713s


while sha256sum for comparison:



real    6m28.897s

user    1m9.273s

sys     0m6.210s


on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz + 16GiB RAM (and data on external drive)






share|improve this answer


























  • And yes, I know that this find command will not properly encode filenames containing quotes according to csv standard

    – Grzegorz Wierzowiecki
    May 25 '17 at 14:12











Your Answer








StackExchange.ready(function() {
var channelOptions = {
tags: "".split(" "),
id: "106"
};
initTagRenderer("".split(" "), "".split(" "), channelOptions);

StackExchange.using("externalEditor", function() {
// Have to fire editor after snippets, if snippets enabled
if (StackExchange.settings.snippets.snippetsEnabled) {
StackExchange.using("snippets", function() {
createEditor();
});
}
else {
createEditor();
}
});

function createEditor() {
StackExchange.prepareEditor({
heartbeatType: 'answer',
autoActivateHeartbeat: false,
convertImagesToLinks: false,
noModals: true,
showLowRepImageUploadWarning: true,
reputationToPostImages: null,
bindNavPrevention: true,
postfix: "",
imageUploader: {
brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
allowUrls: true
},
onDemand: true,
discardSelector: ".discard-answer"
,immediatelyShowMarkdownHelp:true
});


}
});














draft saved

draft discarded


















StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2funix.stackexchange.com%2fquestions%2f6516%2ffiltering-invalid-utf8%23new-answer', 'question_page');
}
);

Post as a guest















Required, but never shown

























6 Answers
6






active

oldest

votes








6 Answers
6






active

oldest

votes









active

oldest

votes






active

oldest

votes









30














If you want to use grep, you can do:



grep -axv '.*' file


in UTF-8 locales to get the lines that have at least an invalid UTF-8 sequence (this works with GNU Grep at least).






share|improve this answer


























  • Except for -a, that's required to work by POSIX. However GNU grep at least fails to spot the UTF-8 encoded UTF-16 surrogate non-characters or codepoints above 0x10FFFF.

    – Stéphane Chazelas
    Dec 19 '14 at 23:10






  • 1





    @StéphaneChazelas Conversely, the -a is needed by GNU grep (which isn't POSIX compliant, I assume). Concerning, the surrogate area and the codepoints above 0x10FFFF, this is a bug then (which could explain that). For this, adding -P should work with GNU grep 2.21 (but is slow); it is buggy at least in Debian grep/2.20-4.

    – vinc17
    Dec 19 '14 at 23:37











  • Sorry, my bad, the behaviour is unspecified in POSIX since grep is a text utility (only expected to work on text input), so I suppose GNU grep's behaviour is as valid as any here.

    – Stéphane Chazelas
    Dec 19 '14 at 23:50













  • @StéphaneChazelas I confirm that POSIX says: "The input files shall be text files." (though not in the description part, which is a bit misleading). This also means that in case of invalid sequences, the behavior is undefined by POSIX. Hence the need to know the implementation, such as GNU grep (whose intent is to regard invalid sequences as non-matching), and possible bugs.

    – vinc17
    Dec 20 '14 at 0:12






  • 1





    I'm switching the accepted answer to this one (sorry, Peter.O because it's simple and works well for my primary use case, which is a heuristic to distinguish UTF-8 from other common encodings (especially 8-bit encodings). Stéphane Chazelas and Peter.O provide more accurate answers in terms of UTF-8 compliance.

    – Gilles
    Dec 23 '14 at 2:11
















30














If you want to use grep, you can do:



grep -axv '.*' file


in UTF-8 locales to get the lines that have at least an invalid UTF-8 sequence (this works with GNU Grep at least).






share|improve this answer


























  • Except for -a, that's required to work by POSIX. However GNU grep at least fails to spot the UTF-8 encoded UTF-16 surrogate non-characters or codepoints above 0x10FFFF.

    – Stéphane Chazelas
    Dec 19 '14 at 23:10






  • 1





    @StéphaneChazelas Conversely, the -a is needed by GNU grep (which isn't POSIX compliant, I assume). Concerning, the surrogate area and the codepoints above 0x10FFFF, this is a bug then (which could explain that). For this, adding -P should work with GNU grep 2.21 (but is slow); it is buggy at least in Debian grep/2.20-4.

    – vinc17
    Dec 19 '14 at 23:37











  • Sorry, my bad, the behaviour is unspecified in POSIX since grep is a text utility (only expected to work on text input), so I suppose GNU grep's behaviour is as valid as any here.

    – Stéphane Chazelas
    Dec 19 '14 at 23:50













  • @StéphaneChazelas I confirm that POSIX says: "The input files shall be text files." (though not in the description part, which is a bit misleading). This also means that in case of invalid sequences, the behavior is undefined by POSIX. Hence the need to know the implementation, such as GNU grep (whose intent is to regard invalid sequences as non-matching), and possible bugs.

    – vinc17
    Dec 20 '14 at 0:12






  • 1





    I'm switching the accepted answer to this one (sorry, Peter.O because it's simple and works well for my primary use case, which is a heuristic to distinguish UTF-8 from other common encodings (especially 8-bit encodings). Stéphane Chazelas and Peter.O provide more accurate answers in terms of UTF-8 compliance.

    – Gilles
    Dec 23 '14 at 2:11














30












30








30







If you want to use grep, you can do:



grep -axv '.*' file


in UTF-8 locales to get the lines that have at least an invalid UTF-8 sequence (this works with GNU Grep at least).






share|improve this answer















If you want to use grep, you can do:



grep -axv '.*' file


in UTF-8 locales to get the lines that have at least an invalid UTF-8 sequence (this works with GNU Grep at least).







share|improve this answer














share|improve this answer



share|improve this answer








edited Jun 30 '15 at 20:27









Stéphane Chazelas

302k56567919




302k56567919










answered Dec 19 '14 at 22:14









vinc17vinc17

8,9091736




8,9091736













  • Except for -a, that's required to work by POSIX. However GNU grep at least fails to spot the UTF-8 encoded UTF-16 surrogate non-characters or codepoints above 0x10FFFF.

    – Stéphane Chazelas
    Dec 19 '14 at 23:10






  • 1





    @StéphaneChazelas Conversely, the -a is needed by GNU grep (which isn't POSIX compliant, I assume). Concerning, the surrogate area and the codepoints above 0x10FFFF, this is a bug then (which could explain that). For this, adding -P should work with GNU grep 2.21 (but is slow); it is buggy at least in Debian grep/2.20-4.

    – vinc17
    Dec 19 '14 at 23:37











  • Sorry, my bad, the behaviour is unspecified in POSIX since grep is a text utility (only expected to work on text input), so I suppose GNU grep's behaviour is as valid as any here.

    – Stéphane Chazelas
    Dec 19 '14 at 23:50













  • @StéphaneChazelas I confirm that POSIX says: "The input files shall be text files." (though not in the description part, which is a bit misleading). This also means that in case of invalid sequences, the behavior is undefined by POSIX. Hence the need to know the implementation, such as GNU grep (whose intent is to regard invalid sequences as non-matching), and possible bugs.

    – vinc17
    Dec 20 '14 at 0:12






  • 1





    I'm switching the accepted answer to this one (sorry, Peter.O because it's simple and works well for my primary use case, which is a heuristic to distinguish UTF-8 from other common encodings (especially 8-bit encodings). Stéphane Chazelas and Peter.O provide more accurate answers in terms of UTF-8 compliance.

    – Gilles
    Dec 23 '14 at 2:11



















  • Except for -a, that's required to work by POSIX. However GNU grep at least fails to spot the UTF-8 encoded UTF-16 surrogate non-characters or codepoints above 0x10FFFF.

    – Stéphane Chazelas
    Dec 19 '14 at 23:10






  • 1





    @StéphaneChazelas Conversely, the -a is needed by GNU grep (which isn't POSIX compliant, I assume). Concerning, the surrogate area and the codepoints above 0x10FFFF, this is a bug then (which could explain that). For this, adding -P should work with GNU grep 2.21 (but is slow); it is buggy at least in Debian grep/2.20-4.

    – vinc17
    Dec 19 '14 at 23:37











  • Sorry, my bad, the behaviour is unspecified in POSIX since grep is a text utility (only expected to work on text input), so I suppose GNU grep's behaviour is as valid as any here.

    – Stéphane Chazelas
    Dec 19 '14 at 23:50













  • @StéphaneChazelas I confirm that POSIX says: "The input files shall be text files." (though not in the description part, which is a bit misleading). This also means that in case of invalid sequences, the behavior is undefined by POSIX. Hence the need to know the implementation, such as GNU grep (whose intent is to regard invalid sequences as non-matching), and possible bugs.

    – vinc17
    Dec 20 '14 at 0:12






  • 1





    I'm switching the accepted answer to this one (sorry, Peter.O because it's simple and works well for my primary use case, which is a heuristic to distinguish UTF-8 from other common encodings (especially 8-bit encodings). Stéphane Chazelas and Peter.O provide more accurate answers in terms of UTF-8 compliance.

    – Gilles
    Dec 23 '14 at 2:11

















Except for -a, that's required to work by POSIX. However GNU grep at least fails to spot the UTF-8 encoded UTF-16 surrogate non-characters or codepoints above 0x10FFFF.

– Stéphane Chazelas
Dec 19 '14 at 23:10





Except for -a, that's required to work by POSIX. However GNU grep at least fails to spot the UTF-8 encoded UTF-16 surrogate non-characters or codepoints above 0x10FFFF.

– Stéphane Chazelas
Dec 19 '14 at 23:10




1




1





@StéphaneChazelas Conversely, the -a is needed by GNU grep (which isn't POSIX compliant, I assume). Concerning, the surrogate area and the codepoints above 0x10FFFF, this is a bug then (which could explain that). For this, adding -P should work with GNU grep 2.21 (but is slow); it is buggy at least in Debian grep/2.20-4.

– vinc17
Dec 19 '14 at 23:37





@StéphaneChazelas Conversely, the -a is needed by GNU grep (which isn't POSIX compliant, I assume). Concerning, the surrogate area and the codepoints above 0x10FFFF, this is a bug then (which could explain that). For this, adding -P should work with GNU grep 2.21 (but is slow); it is buggy at least in Debian grep/2.20-4.

– vinc17
Dec 19 '14 at 23:37













Sorry, my bad, the behaviour is unspecified in POSIX since grep is a text utility (only expected to work on text input), so I suppose GNU grep's behaviour is as valid as any here.

– Stéphane Chazelas
Dec 19 '14 at 23:50







Sorry, my bad, the behaviour is unspecified in POSIX since grep is a text utility (only expected to work on text input), so I suppose GNU grep's behaviour is as valid as any here.

– Stéphane Chazelas
Dec 19 '14 at 23:50















@StéphaneChazelas I confirm that POSIX says: "The input files shall be text files." (though not in the description part, which is a bit misleading). This also means that in case of invalid sequences, the behavior is undefined by POSIX. Hence the need to know the implementation, such as GNU grep (whose intent is to regard invalid sequences as non-matching), and possible bugs.

– vinc17
Dec 20 '14 at 0:12





@StéphaneChazelas I confirm that POSIX says: "The input files shall be text files." (though not in the description part, which is a bit misleading). This also means that in case of invalid sequences, the behavior is undefined by POSIX. Hence the need to know the implementation, such as GNU grep (whose intent is to regard invalid sequences as non-matching), and possible bugs.

– vinc17
Dec 20 '14 at 0:12




1




1





I'm switching the accepted answer to this one (sorry, Peter.O because it's simple and works well for my primary use case, which is a heuristic to distinguish UTF-8 from other common encodings (especially 8-bit encodings). Stéphane Chazelas and Peter.O provide more accurate answers in terms of UTF-8 compliance.

– Gilles
Dec 23 '14 at 2:11





I'm switching the accepted answer to this one (sorry, Peter.O because it's simple and works well for my primary use case, which is a heuristic to distinguish UTF-8 from other common encodings (especially 8-bit encodings). Stéphane Chazelas and Peter.O provide more accurate answers in terms of UTF-8 compliance.

– Gilles
Dec 23 '14 at 2:11













29














I think you probably want iconv. It's for converting between codesets and supports an absurd number of formats. For example, to strip anything not valid in UTF-8 you could use:



iconv -c -t UTF-8 < input.txt > output.txt



Without the -c option it'll report problems in converting to stderr, so with process direction could you save a list of these. Another way would be to strip the non-UTF8 stuff and then



diff input.txt output.txt



for a list of where changes were made.






share|improve this answer
























  • Ok, that's iconv -c -t UTF-8 <input.txt | diff input.txt - | sed -ne 's/^< //p'. It won't work as a pipeline, though, since you need to read the input twice (no, tee won't do, it might block depending on how much buffering iconv and diff do).

    – Gilles
    Jan 27 '11 at 8:22






  • 2





    Random note: input & output may not be the same file or you will end up with an empty file

    – drahnr
    Aug 24 '13 at 8:02






  • 1





    Or use process substitution if your shell supports it diff <(iconv -c -t UTF-8 <input.txt) input.txt

    – Karl
    Nov 11 '15 at 23:41













  • How do you do this and make the output to the same file as the input. I just did this and got a blank file iconv -c -t UTF-8 < input.txt > input.txt

    – Costas Vrahimis
    Nov 8 '16 at 0:24






  • 1





    Thanks.. This allow restoring broken utf-8 postgresql dump , but not discarding valid utf-8

    – Superbiji
    Dec 16 '16 at 10:29
















29














I think you probably want iconv. It's for converting between codesets and supports an absurd number of formats. For example, to strip anything not valid in UTF-8 you could use:



iconv -c -t UTF-8 < input.txt > output.txt



Without the -c option it'll report problems in converting to stderr, so with process direction could you save a list of these. Another way would be to strip the non-UTF8 stuff and then



diff input.txt output.txt



for a list of where changes were made.






share|improve this answer
























  • Ok, that's iconv -c -t UTF-8 <input.txt | diff input.txt - | sed -ne 's/^< //p'. It won't work as a pipeline, though, since you need to read the input twice (no, tee won't do, it might block depending on how much buffering iconv and diff do).

    – Gilles
    Jan 27 '11 at 8:22






  • 2





    Random note: input & output may not be the same file or you will end up with an empty file

    – drahnr
    Aug 24 '13 at 8:02






  • 1





    Or use process substitution if your shell supports it diff <(iconv -c -t UTF-8 <input.txt) input.txt

    – Karl
    Nov 11 '15 at 23:41













  • How do you do this and make the output to the same file as the input. I just did this and got a blank file iconv -c -t UTF-8 < input.txt > input.txt

    – Costas Vrahimis
    Nov 8 '16 at 0:24






  • 1





    Thanks.. This allow restoring broken utf-8 postgresql dump , but not discarding valid utf-8

    – Superbiji
    Dec 16 '16 at 10:29














29












29








29







I think you probably want iconv. It's for converting between codesets and supports an absurd number of formats. For example, to strip anything not valid in UTF-8 you could use:



iconv -c -t UTF-8 < input.txt > output.txt



Without the -c option it'll report problems in converting to stderr, so with process direction could you save a list of these. Another way would be to strip the non-UTF8 stuff and then



diff input.txt output.txt



for a list of where changes were made.






share|improve this answer













I think you probably want iconv. It's for converting between codesets and supports an absurd number of formats. For example, to strip anything not valid in UTF-8 you could use:



iconv -c -t UTF-8 < input.txt > output.txt



Without the -c option it'll report problems in converting to stderr, so with process direction could you save a list of these. Another way would be to strip the non-UTF8 stuff and then



diff input.txt output.txt



for a list of where changes were made.







share|improve this answer












share|improve this answer



share|improve this answer










answered Jan 27 '11 at 3:43









frabjousfrabjous

4,3271825




4,3271825













  • Ok, that's iconv -c -t UTF-8 <input.txt | diff input.txt - | sed -ne 's/^< //p'. It won't work as a pipeline, though, since you need to read the input twice (no, tee won't do, it might block depending on how much buffering iconv and diff do).

    – Gilles
    Jan 27 '11 at 8:22






  • 2





    Random note: input & output may not be the same file or you will end up with an empty file

    – drahnr
    Aug 24 '13 at 8:02






  • 1





    Or use process substitution if your shell supports it diff <(iconv -c -t UTF-8 <input.txt) input.txt

    – Karl
    Nov 11 '15 at 23:41













  • How do you do this and make the output to the same file as the input. I just did this and got a blank file iconv -c -t UTF-8 < input.txt > input.txt

    – Costas Vrahimis
    Nov 8 '16 at 0:24






  • 1





    Thanks.. This allow restoring broken utf-8 postgresql dump , but not discarding valid utf-8

    – Superbiji
    Dec 16 '16 at 10:29



















  • Ok, that's iconv -c -t UTF-8 <input.txt | diff input.txt - | sed -ne 's/^< //p'. It won't work as a pipeline, though, since you need to read the input twice (no, tee won't do, it might block depending on how much buffering iconv and diff do).

    – Gilles
    Jan 27 '11 at 8:22






  • 2





    Random note: input & output may not be the same file or you will end up with an empty file

    – drahnr
    Aug 24 '13 at 8:02






  • 1





    Or use process substitution if your shell supports it diff <(iconv -c -t UTF-8 <input.txt) input.txt

    – Karl
    Nov 11 '15 at 23:41













  • How do you do this and make the output to the same file as the input. I just did this and got a blank file iconv -c -t UTF-8 < input.txt > input.txt

    – Costas Vrahimis
    Nov 8 '16 at 0:24






  • 1





    Thanks.. This allow restoring broken utf-8 postgresql dump , but not discarding valid utf-8

    – Superbiji
    Dec 16 '16 at 10:29

















Ok, that's iconv -c -t UTF-8 <input.txt | diff input.txt - | sed -ne 's/^< //p'. It won't work as a pipeline, though, since you need to read the input twice (no, tee won't do, it might block depending on how much buffering iconv and diff do).

– Gilles
Jan 27 '11 at 8:22





Ok, that's iconv -c -t UTF-8 <input.txt | diff input.txt - | sed -ne 's/^< //p'. It won't work as a pipeline, though, since you need to read the input twice (no, tee won't do, it might block depending on how much buffering iconv and diff do).

– Gilles
Jan 27 '11 at 8:22




2




2





Random note: input & output may not be the same file or you will end up with an empty file

– drahnr
Aug 24 '13 at 8:02





Random note: input & output may not be the same file or you will end up with an empty file

– drahnr
Aug 24 '13 at 8:02




1




1





Or use process substitution if your shell supports it diff <(iconv -c -t UTF-8 <input.txt) input.txt

– Karl
Nov 11 '15 at 23:41







Or use process substitution if your shell supports it diff <(iconv -c -t UTF-8 <input.txt) input.txt

– Karl
Nov 11 '15 at 23:41















How do you do this and make the output to the same file as the input. I just did this and got a blank file iconv -c -t UTF-8 < input.txt > input.txt

– Costas Vrahimis
Nov 8 '16 at 0:24





How do you do this and make the output to the same file as the input. I just did this and got a blank file iconv -c -t UTF-8 < input.txt > input.txt

– Costas Vrahimis
Nov 8 '16 at 0:24




1




1





Thanks.. This allow restoring broken utf-8 postgresql dump , but not discarding valid utf-8

– Superbiji
Dec 16 '16 at 10:29





Thanks.. This allow restoring broken utf-8 postgresql dump , but not discarding valid utf-8

– Superbiji
Dec 16 '16 at 10:29











19














Edit: I've fixed a typo-bug in the regex.. It needed a 'x80` not 80.



The regex to filter out invalid UTF-8 forms, for strict adherance to UTF-8, is as follows 



perl -l -ne '/

 ^( ([x00-x7F])              # 1-byte pattern

   |([xC2-xDF][x80-xBF])   # 2-byte pattern

   |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF])) # 3-byte pattern

   |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))       # 4-byte pattern

  )*$ /x or print'


Output (of key lines.from Test 1):



Codepoint

=========  

00001000  Test=1 mode=strict               valid,invalid,fail=(1000,0,0)          

0000E000  Test=1 mode=strict               valid,invalid,fail=(D800,800,0)          

0010FFFF  mode=strict  test-return=(0,0)   valid,invalid,fail=(10F800,800,0)



Q. How does one create test data to test a regex which filters invalid Unicode ?

A. Create your own UTF-8 test algorithm, and break it's rules...

Catch-22.. But then, how do you then test your test algorithm?



The regex, above, has been tested (using iconv as the reference) for every integer value from 0x00000 to 0x10FFFF.. This upper value being the maximum integer value of a Unicode Codepoint





  • In November 2003 UTF-8 was restricted by RFC 3629 to four bytes covering only the range U+0000 to U+10FFFF, in order to match the constraints of the UTF-16 character encoding.`


According to this wikipedia UTF-8 page,.




  • UTF-8 encodes each of the 1,112,064 code points in the Unicode character set, using one to four 8-bit bytes


This numeber (1,112,064) equates to a range 0x000000 to 0x10F7FF, which is 0x0800 shy of the actual maximum integer-value for the highest Unicode Codepoint: 0x10FFFF



This block of integers is missing from the Unicode Codepoints spectrum, because of the need for the UTF-16 encoding to step beyond its original design intent via a system called surrogate pairs. A block of 0x0800 integers has been reserved to be used by UTF-16.. This block spans the range0x00D800 to 0x00DFFF. None of these inteters are legal Unicode values, and are therefore invalid UTF-8 values.



In Test 1, the regex has been tested against every number in the range of Unicode Codepoints, and it matches exectly the results of iconv .. ie. 0x010F7FF valid values, and 0x000800 invalid values.



However, the issue now arises of, *How does the regex handle Out-Of-Range UTF-8 Value; above 0x010FFFF (UTF-8 can extend to 6 bytes, with a maximum integer value of 0x7FFFFFFF?

To generate the necessary *non-unicode UTF-8 byte values, I've used the following command: 



  perl -C -e 'print chr 0x'$hexUTF32BE


To test their validity (in some fashion), I've used Gilles' UTF-8 regex... 



  perl -l -ne '/

   ^( [00-177]                 # 1-byte pattern

     |[300-337][200-277]      # 2-byte pattern

     |[340-357][200-277]{2}   # 3-byte pattern

     |[360-367][200-277]{3}   # 4-byte pattern

     |[370-373][200-277]{4}   # 5-byte pattern

     |[374-375][200-277]{5}   # 6-byte pattern

    )*$ /x or print'


The output of 'perl's print chr' matches the filtering of Gilles' regex.. One reinforces the validity of the other..
I can't use iconv because it only handles the valid-Unicode Standard subset of the broader (original) UTF-8 standard...



The nunbers involved are rather large, so I've tested top-of-range, bottom-of-range, and several scans stepping by increments such as, 11111, 13579, 33333, 53441... The results all match, so now all that remains is to test the regex against these out-of-range UTF-8-style values (invalid for Unicode, and therefore also invalid for strict UTF-8 itself) ..




Here are the test modules: 



[[ "$(locale charmap)" != "UTF-8" ]] && { echo "ERROR: locale must be UTF-8, but it is $(locale charmap)"; exit 1; }



# Testing the UTF-8 regex

#

# Tests to check that the observed byte-ranges (above) have

#  been  accurately observed and included in the test code and final regex. 

# =========================================================================

: 2 bytes; B2=0 #  run-test=1   do-not-test=0

: 3 bytes; B3=0 #  run-test=1   do-not-test=0

: 4 bytes; B4=0 #  run-test=1   do-not-test=0 



:   regex; Rx=1 #  run-test=1   do-not-test=0



           ((strict=16)); mode[$strict]=strict # iconv -f UTF-16BE  then iconv -f UTF-32BE beyond 0xFFFF)

           ((   lax=32)); mode[$lax]=lax       # iconv -f UTF-32BE  only)



          # modebits=$strict

                  # UTF-8, in relation to UTF-16 has invalid values

                  # modebits=$strict automatically shifts to modebits=$lax

                  # when the tested integer exceeds 0xFFFF

          # modebits=$lax 

                  # UTF-8, in relation to UTF-32, has no restrictione





           # Test 1 Sequentially tests a range of Big-Endian integers

           #      * Unicode Codepoints are a subset ofBig-Endian integers            

           #        ( based on 'iconv' -f UTF-32BE -f UTF-8 )    

           # Note: strict UTF-8 has a few quirks because of UTF-16

                    #    Set modebits=16 to "strictly" test the low range



             Test=1; modebits=$strict

           # Test=2; modebits=$lax

           # Test=3

              mode3wlo=$(( 1*4)) # minimum chars * 4 ( '4' is for UTF-32BE )

              mode3whi=$((10*4)) # minimum chars * 4 ( '4' is for UTF-32BE )





#########################################################################  



# 1 byte  UTF-8 values: Nothing to do; no complexities.



#########################################################################



#  2 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B2==1)) ; then  

  echo "# Test 2 bytes for Valid UTF-8 values: ie. values which are in range"

  # =========================================================================

  time 

  for d1 in {194..223} ;do

      #     bin       oct  hex  dec

      # lo  11000010  302   C2  194

      # hi  11011111  337   DF  223

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {128..191} ;do

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r  |

              iconv -f UTF-8 >/dev/null || { 

                echo "ERROR: Invalid UTF-8 found: ${B2b1}${B2b2}"; exit 20; }

          #

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 2 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  time 

  for d1 in {128..193} {224..255} ;do 

 #for d1 in {128..194} {224..255} ;do # force a valid UTF-8 (needs $B2b2) 

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {0..127} {192..255} ;do

     #for d2 in {0..128} {192..255} ;do # force a valid UTF-8 (needs $B2b1)

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r |

              iconv -f UTF-8 2>/dev/null && { 

                echo "ERROR: VALID UTF-8 found: ${B2b1}${B2b2}"; exit 21; }

          #

      done

  done

  echo

fi



#########################################################################



#  3 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B3==1)) ; then  

  echo "# Test 3 bytes for Valid UTF-8 values: ie. values which are in range"

  # ========================================================================

  time 

  for d1 in {224..239} ;do

      #     bin       oct  hex  dec

      # lo  11100000  340   E0  224

      # hi  11101111  357   EF  239

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {160..191})"

          #     bin       oct  hex  dec  

          # lo  10100000  240   A0  160  

          # hi  10111111  277   BF  191

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {128..159})"

          #     bin       oct  hex  dec  

          # lo  10000000  200   80  128  

          # hi  10011111  237   9F  159

      else

          B3b2range="$(echo {128..191})"

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

      fi

      # 

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r  |

                  iconv -f UTF-8 >/dev/null || { 

                    echo "ERROR: Invalid UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 30; }

              #

          done

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 3 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  #

  # real     26m28.462s  

  # user     27m12.526s  | stepping by 2

  # sys      13m11.193s /

  #

  # real    239m00.836s 

  # user    225m11.108s  | stepping by 1

  # sys     120m00.538s /

  #

  time 

  for d1 in {128..223..1} {240..255..1} ;do 

 #for d1 in {128..224..1} {239..255..1} ;do # force a valid UTF-8 (needs $B2b2,$B3b3) 

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {0..159..1} {192..255..1})"

         #B3b2range="$(> {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {0..127..1} {160..255..1})"

         #B3b2range="$(echo {0..128..1} {160..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      else

          B3b2range="$(echo {0..127..1} {192..255..1})"

         #B3b2range="$(echo {0..128..1} {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      fi

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {0..127..1} {192..255..1} ;do

         #for d3 in {0..128..1} {192..255..1} ;do # force a valid UTF-8 (needs $B2b1)

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r |

                  iconv -f UTF-8 2>/dev/null && { 

                    echo "ERROR: VALID UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 31; }

              #

          done

      done

  done

  echo



fi



#########################################################################



#  Brute force testing in the Astral Plane will take a VERY LONG time..

#  Perhaps selective testing is more appropriate, now that the previous tests 

#     have panned out okay... 

#  

#  4 Byte  UTF-8 values:

if ((B4==1)) ; then  

  echo "# Test 4 bytes for Valid UTF-8 values: ie. values which are in range"

  # ==================================================================

  # real    58m18.531s 

  # user    56m44.317s  | 

  # sys     27m29.867s /

  time 

  for d1 in {240..244} ;do

      #     bin       oct  hex  dec

      # lo  11110000  360   F0  240

      # hi  11110100  364   F4  244  -- F4 encodes some values greater than 0x10FFFF;

      #                                    such a sequence is invalid.

      B4b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B4b1 == "F0" ]] ; then

        B4b2range="$(echo {144..191})" ## f0 90 80 80  to  f0 bf bf bf

        #     bin       oct  hex  dec          010000  --  03FFFF 

        # lo  10010000  220   90  144  

        # hi  10111111  277   BF  191

        #                            

      elif [[ $B4b1 == "F4" ]] ; then

        B4b2range="$(echo {128..143})" ## f4 80 80 80  to  f4 8f bf bf

        #     bin       oct  hex  dec          100000  --  10FFFF 

        # lo  10000000  200   80  128  

        # hi  10001111  217   8F  143  -- F4 encodes some values greater than 0x10FFFF;

        #                                    such a sequence is invalid.

      else

        B4b2range="$(echo {128..191})" ## fx 80 80 80  to  f3 bf bf bf

        #     bin       oct  hex  dec          0C0000  --  0FFFFF

        # lo  10000000  200   80  128          0A0000

        # hi  10111111  277   BF  191

      fi

      #

      for d2 in $B4b2range ;do

          B4b2=$(printf "%0.2X" $d2)

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B4b3=$(printf "%0.2X" $d3)

              echo "${B4b1} ${B4b2} ${B4b3} xx"

              #

              for d4 in {128..191} ;do

                  #     bin       oct  hex  dec

                  # lo  10000000  200   80  128

                  # hi  10111111  277   BF  191

                  B4b4=$(printf "%0.2X" $d4)

                  #

                  echo -n "${B4b1}${B4b2}${B4b3}${B4b4}" |

                    xxd -p -u -r  |

                      iconv -f UTF-8 >/dev/null || { 

                        echo "ERROR: Invalid UTF-8 found: ${B4b1}${B4b2}${B4b3}${B4b4}"; exit 40; }

                  #

              done

          done

      done

  done

  echo "# Test 4 bytes for Valid UTF-8 values: END"

  echo

fi



########################################################################

# There is no test (yet) for negated range values in the astral plane. #  

#                           (all negated range values must be invalid) #

#  I won't bother; This was mainly for me to ge the general feel of    #     

#   the tests, and the final test below should flush anything out..    #

# Traversing the intire UTF-8 range takes quite a while...             #

#   so no need to do it twice (albeit in a slightly different manner)  #

########################################################################



################################

### The construction of:    ####

###  The Regular Expression ####

###      (de-construction?) ####

################################



#     BYTE 1                BYTE 2       BYTE 3      BYTE 4 

# 1: [x00-x7F]

#    ===========

#    ([x00-x7F])

#

# 2: [xC2-xDF]           [x80-xBF]

#    =================================

#    ([xC2-xDF][x80-xBF])

# 

# 3: [xE0]                [xA0-xBF]  [x80-xBF]   

#    [xED]                [x80-x9F]  [x80-xBF]

#    [xE1-xECxEE-xEF]  [x80-xBF]  [x80-xBF]

#    ==============================================

#    ((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))

#

# 4  [xF0]                [x90-xBF]  [x80-xBF]  [x80-xBF]    

#    [xF1-xF3]           [x80-xBF]  [x80-xBF]  [x80-xBF]

#    [xF4]                [x80-x8F]  [x80-xBF]  [x80-xBF]

#    ===========================================================

#    ((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))

#

# The final regex

# ===============

# 1-4:  (([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))

# 4-1:  (((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|([xC2-xDF][x80-xBF])|([x00-x7F]))





#######################################################################

#  The final Test; for a single character (multi chars to follow)     #  

#   Compare the return code of 'iconv' against the 'regex'            #

#   for the full range of 0x000000 to 0x10FFFF                        #

#                                                                     #     

#  Note; this script has 3 modes:                                     #

#        Run this test TWICE, set each mode Manually!                 #     

#                                                                     #     

#     1. Sequentially test every value from 0x000000 to 0x10FFFF      #     

#     2. Throw a spanner into the works! Force random byte patterns   #     

#     2. Throw a spanner into the works! Force random longer strings  #     

#        ==============================                               #     

#                                                                     #     

#  Note: The purpose of this routine is to determine if there is any  #

#        difference how 'iconv' and 'regex' handle the same data      #  

#                                                                     #     

#######################################################################

if ((Rx==1)) ; then

  # real    191m34.826s

  # user    158m24.114s

  # sys      83m10.676s

  time { 

  invalCt=0

  validCt=0

   failCt=0

  decBeg=$((0x00110000)) # incement by decimal integer

  decMax=$((0x7FFFFFFF)) # incement by decimal integer

  # 

  for ((CPDec=decBeg;CPDec<=decMax;CPDec+=13247)) ;do

      ((D==1)) && echo "=========================================================="

      #

      # Convert decimal integer '$CPDec' to Hex-digits; 6-long  (dec2hex)

      hexUTF32BE=$(printf '%0.8Xn' $CPDec)  # hexUTF32BE



      # progress count  

      if (((CPDec%$((0x1000)))==0)) ;then

          ((Test>2)) && echo

          echo "$hexUTF32BE  Test=$Test mode=${mode[$modebits]}            "

      fi

      if   ((Test==1 || Test==2 ))

      then # Test 1. Sequentially test every value from 0x000000 to 0x10FFFF

          #

          if   ((Test==2)) ; then

              bits=32

              UTF8="$( perl -C -e 'print chr 0x'$hexUTF32BE |

                perl -l -ne '/^(  [00-177]

                                | [300-337][200-277]

                                | [340-357][200-277]{2}

                                | [360-367][200-277]{3}

                                | [370-373][200-277]{4}

                                | [374-375][200-277]{5}

                               )*$/x and print' |xxd -p )"

              UTF8="${UTF8%0a}"

              [[ -n "$UTF8" ]] 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=



          elif ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              bits=16

              UTF8="$( echo -n "${hexUTF32BE:4}" |

                xxd -p -u -r |

                  iconv -f UTF-16BE -t UTF-8 2>/dev/null)" 

                    && rcIco16=0 || rcIco16=1  

                       rcIco32=

          else

              bits=32

              UTF8="$( echo -n "$hexUTF32BE" |

                xxd -p -u -r |

                  iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=

          fi

          # echo "1 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((${rcIco16}${rcIco32}!=0)) ;then

              # 'iconv -f UTF-16BE' failed produce a reliable UTF-8

              if ((bits==16)) ;then

                  ((D==1)) &&           echo "bits-$bits rcIconv: error    $hexUTF32BE .. 'strict' failed, now trying 'lax'"

                  #  iconv failed to create a  'srict' UTF-8 so   

                  #      try UTF-32BE to get a   'lax' UTF-8 pattern    

                  UTF8="$( echo -n "$hexUTF32BE" |

                    xxd -p -u -r |

                      iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                        && rcIco32=0 || rcIco32=1

                  #echo "2 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

                  if ((rcIco32!=0)) ;then

                      ((D==1)) &&               echo -n "bits-$bits rcIconv: Cannot gen UTF-8 for: $hexUTF32BE"

                      rcIco32=1

                  fi

              fi

          fi

          # echo "3 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((rcIco16==0 || rcIco32==0)) ;then

              # 'strict(16)' OR 'lax(32)'... 'iconv' managed to generate a UTF-8 pattern  

                  ((D==1)) &&       echo -n "bits-$bits rcIconv: pattern* $hexUTF32BE"

                  ((D==1)) &&       if [[ $bits == "16" && $rcIco32 == "0" ]] ;then 

                  echo " .. 'lax' UTF-8 produced a pattern"

              else

                  echo

              fi

               # regex test

              if ((modebits==strict)) ;then

                 #rxOut="$(echo -n "$UTF8" |perl -l -ne '/^(([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))*$/ or print' )"

                                     rxOut="$(echo -n "$UTF8" |

                  perl -l -ne '/^( ([x00-x7F])             # 1-byte pattern

                                  |([xC2-xDF][x80-xBF])  # 2-byte pattern

                                  |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))  # 3-byte pattern

                                  |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))        # 4-byte pattern

                                 )*$ /x or print' )"

               else

                  if ((Test==2)) ;then

                      rx="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ and print')"

                      [[ "$UTF8" != "$rx" ]] && rxOut="$UTF8" || rxOut=

                      rx="$(echo -n "$rx" |sed -e "s/(..)/1 /g")"  

                  else 

                      rxOut="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ or print' )"

                  fi

              fi

              if [[ "$rxOut" == "" ]] ;then

                ((D==1)) &&           echo "        rcRegex: ok"

                  rcRegex=0

              else

                  ((D==1)) &&           echo -n "bits-$bits rcRegex: error    $hexUTF32BE .. 'strict' failed,"

                  ((D==1)) &&           if [[  "12" == *$Test* ]] ;then 

                                            echo # "  (codepoint) Test $Test" 

                                        else

                                            echo

                                        fi

                  rcRegex=1

              fi

          fi

          #

      elif [[ $Test == 2 ]]

      then # Test 2. Throw a randomizing spanner into the works! 

          #          Then test the  arbitary bytes ASIS

          #

          hexLineRand="$(echo -n "$hexUTF32BE" |

            sed -re "s/(.)(.)(.)(.)(.)(.)(.)(.)/1n2n3n4n5n6n7n8/" |

              sort -R |

                tr -d 'n')"

          # 

      elif [[ $Test == 3 ]]

      then # Test 3. Test single UTF-16BE bytes in the range 0x00000000 to 0x7FFFFFFF

          #

          echo "Test 3 is not properly implemented yet.. Exiting"

          exit 99 

      else

          echo "ERROR: Invalid mode"

          exit

      fi

      #

      #

      if ((Test==1 || Test=2)) ;then

          if ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              ((rcIconv=rcIco16))

          else

              ((rcIconv=rcIco32))

          fi

          if ((rcRegex!=rcIconv)) ;then

              [[ $Test != 1 ]] && echo

              if ((rcRegex==1)) ;then

                  echo "ERROR: 'regex' ok, but NOT 'iconv': ${hexUTF32BE} "

              else

                  echo "ERROR: 'iconv' ok, but NOT 'regex': ${hexUTF32BE} "

              fi

              ((failCt++));

          elif ((rcRegex!=0)) ;then

            # ((invalCt++)); echo -ne "$hexUTF32BE  exit-codes $${rcIco16}${rcIco32}=,$rcRegext: $(printf "%0.8Xn" $invalCt)t$hexLine$(printf "%$(((mode3whi*2)-${#hexLine}))s")r"

              ((invalCt++)) 

          else

              ((validCt++)) 

          fi

          if   ((Test==1)) ;then

              echo -ne "$hexUTF32BE "    "mode=${mode[$modebits]}  test-return=($rcIconv,$rcRegex)   valid,invalid,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))          r"

          else 

              echo -ne "$hexUTF32BE $rx mode=${mode[$modebits]} test-return=($rcIconv,$rcRegex)  val,inval,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))r"

          fi

      fi

  done

  } # End time

fi

exit





share|improve this answer


























  • The main problem with my regexp is that it allowed some forbidden sequences such as 300200 (really bad: that's code point 0 not expressed with a null byte!). I think your regexp rejects them correctly.

    – Gilles
    May 8 '11 at 12:38
















19














Edit: I've fixed a typo-bug in the regex.. It needed a 'x80` not 80.



The regex to filter out invalid UTF-8 forms, for strict adherance to UTF-8, is as follows 



perl -l -ne '/

 ^( ([x00-x7F])              # 1-byte pattern

   |([xC2-xDF][x80-xBF])   # 2-byte pattern

   |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF])) # 3-byte pattern

   |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))       # 4-byte pattern

  )*$ /x or print'


Output (of key lines.from Test 1):



Codepoint

=========  

00001000  Test=1 mode=strict               valid,invalid,fail=(1000,0,0)          

0000E000  Test=1 mode=strict               valid,invalid,fail=(D800,800,0)          

0010FFFF  mode=strict  test-return=(0,0)   valid,invalid,fail=(10F800,800,0)



Q. How does one create test data to test a regex which filters invalid Unicode ?

A. Create your own UTF-8 test algorithm, and break it's rules...

Catch-22.. But then, how do you then test your test algorithm?



The regex, above, has been tested (using iconv as the reference) for every integer value from 0x00000 to 0x10FFFF.. This upper value being the maximum integer value of a Unicode Codepoint





  • In November 2003 UTF-8 was restricted by RFC 3629 to four bytes covering only the range U+0000 to U+10FFFF, in order to match the constraints of the UTF-16 character encoding.`


According to this wikipedia UTF-8 page,.




  • UTF-8 encodes each of the 1,112,064 code points in the Unicode character set, using one to four 8-bit bytes


This numeber (1,112,064) equates to a range 0x000000 to 0x10F7FF, which is 0x0800 shy of the actual maximum integer-value for the highest Unicode Codepoint: 0x10FFFF



This block of integers is missing from the Unicode Codepoints spectrum, because of the need for the UTF-16 encoding to step beyond its original design intent via a system called surrogate pairs. A block of 0x0800 integers has been reserved to be used by UTF-16.. This block spans the range0x00D800 to 0x00DFFF. None of these inteters are legal Unicode values, and are therefore invalid UTF-8 values.



In Test 1, the regex has been tested against every number in the range of Unicode Codepoints, and it matches exectly the results of iconv .. ie. 0x010F7FF valid values, and 0x000800 invalid values.



However, the issue now arises of, *How does the regex handle Out-Of-Range UTF-8 Value; above 0x010FFFF (UTF-8 can extend to 6 bytes, with a maximum integer value of 0x7FFFFFFF?

To generate the necessary *non-unicode UTF-8 byte values, I've used the following command: 



  perl -C -e 'print chr 0x'$hexUTF32BE


To test their validity (in some fashion), I've used Gilles' UTF-8 regex... 



  perl -l -ne '/

   ^( [00-177]                 # 1-byte pattern

     |[300-337][200-277]      # 2-byte pattern

     |[340-357][200-277]{2}   # 3-byte pattern

     |[360-367][200-277]{3}   # 4-byte pattern

     |[370-373][200-277]{4}   # 5-byte pattern

     |[374-375][200-277]{5}   # 6-byte pattern

    )*$ /x or print'


The output of 'perl's print chr' matches the filtering of Gilles' regex.. One reinforces the validity of the other..
I can't use iconv because it only handles the valid-Unicode Standard subset of the broader (original) UTF-8 standard...



The nunbers involved are rather large, so I've tested top-of-range, bottom-of-range, and several scans stepping by increments such as, 11111, 13579, 33333, 53441... The results all match, so now all that remains is to test the regex against these out-of-range UTF-8-style values (invalid for Unicode, and therefore also invalid for strict UTF-8 itself) ..




Here are the test modules: 



[[ "$(locale charmap)" != "UTF-8" ]] && { echo "ERROR: locale must be UTF-8, but it is $(locale charmap)"; exit 1; }



# Testing the UTF-8 regex

#

# Tests to check that the observed byte-ranges (above) have

#  been  accurately observed and included in the test code and final regex. 

# =========================================================================

: 2 bytes; B2=0 #  run-test=1   do-not-test=0

: 3 bytes; B3=0 #  run-test=1   do-not-test=0

: 4 bytes; B4=0 #  run-test=1   do-not-test=0 



:   regex; Rx=1 #  run-test=1   do-not-test=0



           ((strict=16)); mode[$strict]=strict # iconv -f UTF-16BE  then iconv -f UTF-32BE beyond 0xFFFF)

           ((   lax=32)); mode[$lax]=lax       # iconv -f UTF-32BE  only)



          # modebits=$strict

                  # UTF-8, in relation to UTF-16 has invalid values

                  # modebits=$strict automatically shifts to modebits=$lax

                  # when the tested integer exceeds 0xFFFF

          # modebits=$lax 

                  # UTF-8, in relation to UTF-32, has no restrictione





           # Test 1 Sequentially tests a range of Big-Endian integers

           #      * Unicode Codepoints are a subset ofBig-Endian integers            

           #        ( based on 'iconv' -f UTF-32BE -f UTF-8 )    

           # Note: strict UTF-8 has a few quirks because of UTF-16

                    #    Set modebits=16 to "strictly" test the low range



             Test=1; modebits=$strict

           # Test=2; modebits=$lax

           # Test=3

              mode3wlo=$(( 1*4)) # minimum chars * 4 ( '4' is for UTF-32BE )

              mode3whi=$((10*4)) # minimum chars * 4 ( '4' is for UTF-32BE )





#########################################################################  



# 1 byte  UTF-8 values: Nothing to do; no complexities.



#########################################################################



#  2 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B2==1)) ; then  

  echo "# Test 2 bytes for Valid UTF-8 values: ie. values which are in range"

  # =========================================================================

  time 

  for d1 in {194..223} ;do

      #     bin       oct  hex  dec

      # lo  11000010  302   C2  194

      # hi  11011111  337   DF  223

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {128..191} ;do

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r  |

              iconv -f UTF-8 >/dev/null || { 

                echo "ERROR: Invalid UTF-8 found: ${B2b1}${B2b2}"; exit 20; }

          #

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 2 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  time 

  for d1 in {128..193} {224..255} ;do 

 #for d1 in {128..194} {224..255} ;do # force a valid UTF-8 (needs $B2b2) 

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {0..127} {192..255} ;do

     #for d2 in {0..128} {192..255} ;do # force a valid UTF-8 (needs $B2b1)

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r |

              iconv -f UTF-8 2>/dev/null && { 

                echo "ERROR: VALID UTF-8 found: ${B2b1}${B2b2}"; exit 21; }

          #

      done

  done

  echo

fi



#########################################################################



#  3 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B3==1)) ; then  

  echo "# Test 3 bytes for Valid UTF-8 values: ie. values which are in range"

  # ========================================================================

  time 

  for d1 in {224..239} ;do

      #     bin       oct  hex  dec

      # lo  11100000  340   E0  224

      # hi  11101111  357   EF  239

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {160..191})"

          #     bin       oct  hex  dec  

          # lo  10100000  240   A0  160  

          # hi  10111111  277   BF  191

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {128..159})"

          #     bin       oct  hex  dec  

          # lo  10000000  200   80  128  

          # hi  10011111  237   9F  159

      else

          B3b2range="$(echo {128..191})"

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

      fi

      # 

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r  |

                  iconv -f UTF-8 >/dev/null || { 

                    echo "ERROR: Invalid UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 30; }

              #

          done

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 3 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  #

  # real     26m28.462s  

  # user     27m12.526s  | stepping by 2

  # sys      13m11.193s /

  #

  # real    239m00.836s 

  # user    225m11.108s  | stepping by 1

  # sys     120m00.538s /

  #

  time 

  for d1 in {128..223..1} {240..255..1} ;do 

 #for d1 in {128..224..1} {239..255..1} ;do # force a valid UTF-8 (needs $B2b2,$B3b3) 

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {0..159..1} {192..255..1})"

         #B3b2range="$(> {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {0..127..1} {160..255..1})"

         #B3b2range="$(echo {0..128..1} {160..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      else

          B3b2range="$(echo {0..127..1} {192..255..1})"

         #B3b2range="$(echo {0..128..1} {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      fi

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {0..127..1} {192..255..1} ;do

         #for d3 in {0..128..1} {192..255..1} ;do # force a valid UTF-8 (needs $B2b1)

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r |

                  iconv -f UTF-8 2>/dev/null && { 

                    echo "ERROR: VALID UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 31; }

              #

          done

      done

  done

  echo



fi



#########################################################################



#  Brute force testing in the Astral Plane will take a VERY LONG time..

#  Perhaps selective testing is more appropriate, now that the previous tests 

#     have panned out okay... 

#  

#  4 Byte  UTF-8 values:

if ((B4==1)) ; then  

  echo "# Test 4 bytes for Valid UTF-8 values: ie. values which are in range"

  # ==================================================================

  # real    58m18.531s 

  # user    56m44.317s  | 

  # sys     27m29.867s /

  time 

  for d1 in {240..244} ;do

      #     bin       oct  hex  dec

      # lo  11110000  360   F0  240

      # hi  11110100  364   F4  244  -- F4 encodes some values greater than 0x10FFFF;

      #                                    such a sequence is invalid.

      B4b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B4b1 == "F0" ]] ; then

        B4b2range="$(echo {144..191})" ## f0 90 80 80  to  f0 bf bf bf

        #     bin       oct  hex  dec          010000  --  03FFFF 

        # lo  10010000  220   90  144  

        # hi  10111111  277   BF  191

        #                            

      elif [[ $B4b1 == "F4" ]] ; then

        B4b2range="$(echo {128..143})" ## f4 80 80 80  to  f4 8f bf bf

        #     bin       oct  hex  dec          100000  --  10FFFF 

        # lo  10000000  200   80  128  

        # hi  10001111  217   8F  143  -- F4 encodes some values greater than 0x10FFFF;

        #                                    such a sequence is invalid.

      else

        B4b2range="$(echo {128..191})" ## fx 80 80 80  to  f3 bf bf bf

        #     bin       oct  hex  dec          0C0000  --  0FFFFF

        # lo  10000000  200   80  128          0A0000

        # hi  10111111  277   BF  191

      fi

      #

      for d2 in $B4b2range ;do

          B4b2=$(printf "%0.2X" $d2)

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B4b3=$(printf "%0.2X" $d3)

              echo "${B4b1} ${B4b2} ${B4b3} xx"

              #

              for d4 in {128..191} ;do

                  #     bin       oct  hex  dec

                  # lo  10000000  200   80  128

                  # hi  10111111  277   BF  191

                  B4b4=$(printf "%0.2X" $d4)

                  #

                  echo -n "${B4b1}${B4b2}${B4b3}${B4b4}" |

                    xxd -p -u -r  |

                      iconv -f UTF-8 >/dev/null || { 

                        echo "ERROR: Invalid UTF-8 found: ${B4b1}${B4b2}${B4b3}${B4b4}"; exit 40; }

                  #

              done

          done

      done

  done

  echo "# Test 4 bytes for Valid UTF-8 values: END"

  echo

fi



########################################################################

# There is no test (yet) for negated range values in the astral plane. #  

#                           (all negated range values must be invalid) #

#  I won't bother; This was mainly for me to ge the general feel of    #     

#   the tests, and the final test below should flush anything out..    #

# Traversing the intire UTF-8 range takes quite a while...             #

#   so no need to do it twice (albeit in a slightly different manner)  #

########################################################################



################################

### The construction of:    ####

###  The Regular Expression ####

###      (de-construction?) ####

################################



#     BYTE 1                BYTE 2       BYTE 3      BYTE 4 

# 1: [x00-x7F]

#    ===========

#    ([x00-x7F])

#

# 2: [xC2-xDF]           [x80-xBF]

#    =================================

#    ([xC2-xDF][x80-xBF])

# 

# 3: [xE0]                [xA0-xBF]  [x80-xBF]   

#    [xED]                [x80-x9F]  [x80-xBF]

#    [xE1-xECxEE-xEF]  [x80-xBF]  [x80-xBF]

#    ==============================================

#    ((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))

#

# 4  [xF0]                [x90-xBF]  [x80-xBF]  [x80-xBF]    

#    [xF1-xF3]           [x80-xBF]  [x80-xBF]  [x80-xBF]

#    [xF4]                [x80-x8F]  [x80-xBF]  [x80-xBF]

#    ===========================================================

#    ((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))

#

# The final regex

# ===============

# 1-4:  (([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))

# 4-1:  (((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|([xC2-xDF][x80-xBF])|([x00-x7F]))





#######################################################################

#  The final Test; for a single character (multi chars to follow)     #  

#   Compare the return code of 'iconv' against the 'regex'            #

#   for the full range of 0x000000 to 0x10FFFF                        #

#                                                                     #     

#  Note; this script has 3 modes:                                     #

#        Run this test TWICE, set each mode Manually!                 #     

#                                                                     #     

#     1. Sequentially test every value from 0x000000 to 0x10FFFF      #     

#     2. Throw a spanner into the works! Force random byte patterns   #     

#     2. Throw a spanner into the works! Force random longer strings  #     

#        ==============================                               #     

#                                                                     #     

#  Note: The purpose of this routine is to determine if there is any  #

#        difference how 'iconv' and 'regex' handle the same data      #  

#                                                                     #     

#######################################################################

if ((Rx==1)) ; then

  # real    191m34.826s

  # user    158m24.114s

  # sys      83m10.676s

  time { 

  invalCt=0

  validCt=0

   failCt=0

  decBeg=$((0x00110000)) # incement by decimal integer

  decMax=$((0x7FFFFFFF)) # incement by decimal integer

  # 

  for ((CPDec=decBeg;CPDec<=decMax;CPDec+=13247)) ;do

      ((D==1)) && echo "=========================================================="

      #

      # Convert decimal integer '$CPDec' to Hex-digits; 6-long  (dec2hex)

      hexUTF32BE=$(printf '%0.8Xn' $CPDec)  # hexUTF32BE



      # progress count  

      if (((CPDec%$((0x1000)))==0)) ;then

          ((Test>2)) && echo

          echo "$hexUTF32BE  Test=$Test mode=${mode[$modebits]}            "

      fi

      if   ((Test==1 || Test==2 ))

      then # Test 1. Sequentially test every value from 0x000000 to 0x10FFFF

          #

          if   ((Test==2)) ; then

              bits=32

              UTF8="$( perl -C -e 'print chr 0x'$hexUTF32BE |

                perl -l -ne '/^(  [00-177]

                                | [300-337][200-277]

                                | [340-357][200-277]{2}

                                | [360-367][200-277]{3}

                                | [370-373][200-277]{4}

                                | [374-375][200-277]{5}

                               )*$/x and print' |xxd -p )"

              UTF8="${UTF8%0a}"

              [[ -n "$UTF8" ]] 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=



          elif ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              bits=16

              UTF8="$( echo -n "${hexUTF32BE:4}" |

                xxd -p -u -r |

                  iconv -f UTF-16BE -t UTF-8 2>/dev/null)" 

                    && rcIco16=0 || rcIco16=1  

                       rcIco32=

          else

              bits=32

              UTF8="$( echo -n "$hexUTF32BE" |

                xxd -p -u -r |

                  iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=

          fi

          # echo "1 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((${rcIco16}${rcIco32}!=0)) ;then

              # 'iconv -f UTF-16BE' failed produce a reliable UTF-8

              if ((bits==16)) ;then

                  ((D==1)) &&           echo "bits-$bits rcIconv: error    $hexUTF32BE .. 'strict' failed, now trying 'lax'"

                  #  iconv failed to create a  'srict' UTF-8 so   

                  #      try UTF-32BE to get a   'lax' UTF-8 pattern    

                  UTF8="$( echo -n "$hexUTF32BE" |

                    xxd -p -u -r |

                      iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                        && rcIco32=0 || rcIco32=1

                  #echo "2 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

                  if ((rcIco32!=0)) ;then

                      ((D==1)) &&               echo -n "bits-$bits rcIconv: Cannot gen UTF-8 for: $hexUTF32BE"

                      rcIco32=1

                  fi

              fi

          fi

          # echo "3 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((rcIco16==0 || rcIco32==0)) ;then

              # 'strict(16)' OR 'lax(32)'... 'iconv' managed to generate a UTF-8 pattern  

                  ((D==1)) &&       echo -n "bits-$bits rcIconv: pattern* $hexUTF32BE"

                  ((D==1)) &&       if [[ $bits == "16" && $rcIco32 == "0" ]] ;then 

                  echo " .. 'lax' UTF-8 produced a pattern"

              else

                  echo

              fi

               # regex test

              if ((modebits==strict)) ;then

                 #rxOut="$(echo -n "$UTF8" |perl -l -ne '/^(([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))*$/ or print' )"

                                     rxOut="$(echo -n "$UTF8" |

                  perl -l -ne '/^( ([x00-x7F])             # 1-byte pattern

                                  |([xC2-xDF][x80-xBF])  # 2-byte pattern

                                  |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))  # 3-byte pattern

                                  |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))        # 4-byte pattern

                                 )*$ /x or print' )"

               else

                  if ((Test==2)) ;then

                      rx="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ and print')"

                      [[ "$UTF8" != "$rx" ]] && rxOut="$UTF8" || rxOut=

                      rx="$(echo -n "$rx" |sed -e "s/(..)/1 /g")"  

                  else 

                      rxOut="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ or print' )"

                  fi

              fi

              if [[ "$rxOut" == "" ]] ;then

                ((D==1)) &&           echo "        rcRegex: ok"

                  rcRegex=0

              else

                  ((D==1)) &&           echo -n "bits-$bits rcRegex: error    $hexUTF32BE .. 'strict' failed,"

                  ((D==1)) &&           if [[  "12" == *$Test* ]] ;then 

                                            echo # "  (codepoint) Test $Test" 

                                        else

                                            echo

                                        fi

                  rcRegex=1

              fi

          fi

          #

      elif [[ $Test == 2 ]]

      then # Test 2. Throw a randomizing spanner into the works! 

          #          Then test the  arbitary bytes ASIS

          #

          hexLineRand="$(echo -n "$hexUTF32BE" |

            sed -re "s/(.)(.)(.)(.)(.)(.)(.)(.)/1n2n3n4n5n6n7n8/" |

              sort -R |

                tr -d 'n')"

          # 

      elif [[ $Test == 3 ]]

      then # Test 3. Test single UTF-16BE bytes in the range 0x00000000 to 0x7FFFFFFF

          #

          echo "Test 3 is not properly implemented yet.. Exiting"

          exit 99 

      else

          echo "ERROR: Invalid mode"

          exit

      fi

      #

      #

      if ((Test==1 || Test=2)) ;then

          if ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              ((rcIconv=rcIco16))

          else

              ((rcIconv=rcIco32))

          fi

          if ((rcRegex!=rcIconv)) ;then

              [[ $Test != 1 ]] && echo

              if ((rcRegex==1)) ;then

                  echo "ERROR: 'regex' ok, but NOT 'iconv': ${hexUTF32BE} "

              else

                  echo "ERROR: 'iconv' ok, but NOT 'regex': ${hexUTF32BE} "

              fi

              ((failCt++));

          elif ((rcRegex!=0)) ;then

            # ((invalCt++)); echo -ne "$hexUTF32BE  exit-codes $${rcIco16}${rcIco32}=,$rcRegext: $(printf "%0.8Xn" $invalCt)t$hexLine$(printf "%$(((mode3whi*2)-${#hexLine}))s")r"

              ((invalCt++)) 

          else

              ((validCt++)) 

          fi

          if   ((Test==1)) ;then

              echo -ne "$hexUTF32BE "    "mode=${mode[$modebits]}  test-return=($rcIconv,$rcRegex)   valid,invalid,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))          r"

          else 

              echo -ne "$hexUTF32BE $rx mode=${mode[$modebits]} test-return=($rcIconv,$rcRegex)  val,inval,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))r"

          fi

      fi

  done

  } # End time

fi

exit





share|improve this answer


























  • The main problem with my regexp is that it allowed some forbidden sequences such as 300200 (really bad: that's code point 0 not expressed with a null byte!). I think your regexp rejects them correctly.

    – Gilles
    May 8 '11 at 12:38














19












19








19







Edit: I've fixed a typo-bug in the regex.. It needed a 'x80` not 80.



The regex to filter out invalid UTF-8 forms, for strict adherance to UTF-8, is as follows 



perl -l -ne '/

 ^( ([x00-x7F])              # 1-byte pattern

   |([xC2-xDF][x80-xBF])   # 2-byte pattern

   |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF])) # 3-byte pattern

   |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))       # 4-byte pattern

  )*$ /x or print'


Output (of key lines.from Test 1):



Codepoint

=========  

00001000  Test=1 mode=strict               valid,invalid,fail=(1000,0,0)          

0000E000  Test=1 mode=strict               valid,invalid,fail=(D800,800,0)          

0010FFFF  mode=strict  test-return=(0,0)   valid,invalid,fail=(10F800,800,0)



Q. How does one create test data to test a regex which filters invalid Unicode ?

A. Create your own UTF-8 test algorithm, and break it's rules...

Catch-22.. But then, how do you then test your test algorithm?



The regex, above, has been tested (using iconv as the reference) for every integer value from 0x00000 to 0x10FFFF.. This upper value being the maximum integer value of a Unicode Codepoint





  • In November 2003 UTF-8 was restricted by RFC 3629 to four bytes covering only the range U+0000 to U+10FFFF, in order to match the constraints of the UTF-16 character encoding.`


According to this wikipedia UTF-8 page,.




  • UTF-8 encodes each of the 1,112,064 code points in the Unicode character set, using one to four 8-bit bytes


This numeber (1,112,064) equates to a range 0x000000 to 0x10F7FF, which is 0x0800 shy of the actual maximum integer-value for the highest Unicode Codepoint: 0x10FFFF



This block of integers is missing from the Unicode Codepoints spectrum, because of the need for the UTF-16 encoding to step beyond its original design intent via a system called surrogate pairs. A block of 0x0800 integers has been reserved to be used by UTF-16.. This block spans the range0x00D800 to 0x00DFFF. None of these inteters are legal Unicode values, and are therefore invalid UTF-8 values.



In Test 1, the regex has been tested against every number in the range of Unicode Codepoints, and it matches exectly the results of iconv .. ie. 0x010F7FF valid values, and 0x000800 invalid values.



However, the issue now arises of, *How does the regex handle Out-Of-Range UTF-8 Value; above 0x010FFFF (UTF-8 can extend to 6 bytes, with a maximum integer value of 0x7FFFFFFF?

To generate the necessary *non-unicode UTF-8 byte values, I've used the following command: 



  perl -C -e 'print chr 0x'$hexUTF32BE


To test their validity (in some fashion), I've used Gilles' UTF-8 regex... 



  perl -l -ne '/

   ^( [00-177]                 # 1-byte pattern

     |[300-337][200-277]      # 2-byte pattern

     |[340-357][200-277]{2}   # 3-byte pattern

     |[360-367][200-277]{3}   # 4-byte pattern

     |[370-373][200-277]{4}   # 5-byte pattern

     |[374-375][200-277]{5}   # 6-byte pattern

    )*$ /x or print'


The output of 'perl's print chr' matches the filtering of Gilles' regex.. One reinforces the validity of the other..
I can't use iconv because it only handles the valid-Unicode Standard subset of the broader (original) UTF-8 standard...



The nunbers involved are rather large, so I've tested top-of-range, bottom-of-range, and several scans stepping by increments such as, 11111, 13579, 33333, 53441... The results all match, so now all that remains is to test the regex against these out-of-range UTF-8-style values (invalid for Unicode, and therefore also invalid for strict UTF-8 itself) ..




Here are the test modules: 



[[ "$(locale charmap)" != "UTF-8" ]] && { echo "ERROR: locale must be UTF-8, but it is $(locale charmap)"; exit 1; }



# Testing the UTF-8 regex

#

# Tests to check that the observed byte-ranges (above) have

#  been  accurately observed and included in the test code and final regex. 

# =========================================================================

: 2 bytes; B2=0 #  run-test=1   do-not-test=0

: 3 bytes; B3=0 #  run-test=1   do-not-test=0

: 4 bytes; B4=0 #  run-test=1   do-not-test=0 



:   regex; Rx=1 #  run-test=1   do-not-test=0



           ((strict=16)); mode[$strict]=strict # iconv -f UTF-16BE  then iconv -f UTF-32BE beyond 0xFFFF)

           ((   lax=32)); mode[$lax]=lax       # iconv -f UTF-32BE  only)



          # modebits=$strict

                  # UTF-8, in relation to UTF-16 has invalid values

                  # modebits=$strict automatically shifts to modebits=$lax

                  # when the tested integer exceeds 0xFFFF

          # modebits=$lax 

                  # UTF-8, in relation to UTF-32, has no restrictione





           # Test 1 Sequentially tests a range of Big-Endian integers

           #      * Unicode Codepoints are a subset ofBig-Endian integers            

           #        ( based on 'iconv' -f UTF-32BE -f UTF-8 )    

           # Note: strict UTF-8 has a few quirks because of UTF-16

                    #    Set modebits=16 to "strictly" test the low range



             Test=1; modebits=$strict

           # Test=2; modebits=$lax

           # Test=3

              mode3wlo=$(( 1*4)) # minimum chars * 4 ( '4' is for UTF-32BE )

              mode3whi=$((10*4)) # minimum chars * 4 ( '4' is for UTF-32BE )





#########################################################################  



# 1 byte  UTF-8 values: Nothing to do; no complexities.



#########################################################################



#  2 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B2==1)) ; then  

  echo "# Test 2 bytes for Valid UTF-8 values: ie. values which are in range"

  # =========================================================================

  time 

  for d1 in {194..223} ;do

      #     bin       oct  hex  dec

      # lo  11000010  302   C2  194

      # hi  11011111  337   DF  223

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {128..191} ;do

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r  |

              iconv -f UTF-8 >/dev/null || { 

                echo "ERROR: Invalid UTF-8 found: ${B2b1}${B2b2}"; exit 20; }

          #

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 2 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  time 

  for d1 in {128..193} {224..255} ;do 

 #for d1 in {128..194} {224..255} ;do # force a valid UTF-8 (needs $B2b2) 

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {0..127} {192..255} ;do

     #for d2 in {0..128} {192..255} ;do # force a valid UTF-8 (needs $B2b1)

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r |

              iconv -f UTF-8 2>/dev/null && { 

                echo "ERROR: VALID UTF-8 found: ${B2b1}${B2b2}"; exit 21; }

          #

      done

  done

  echo

fi



#########################################################################



#  3 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B3==1)) ; then  

  echo "# Test 3 bytes for Valid UTF-8 values: ie. values which are in range"

  # ========================================================================

  time 

  for d1 in {224..239} ;do

      #     bin       oct  hex  dec

      # lo  11100000  340   E0  224

      # hi  11101111  357   EF  239

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {160..191})"

          #     bin       oct  hex  dec  

          # lo  10100000  240   A0  160  

          # hi  10111111  277   BF  191

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {128..159})"

          #     bin       oct  hex  dec  

          # lo  10000000  200   80  128  

          # hi  10011111  237   9F  159

      else

          B3b2range="$(echo {128..191})"

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

      fi

      # 

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r  |

                  iconv -f UTF-8 >/dev/null || { 

                    echo "ERROR: Invalid UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 30; }

              #

          done

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 3 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  #

  # real     26m28.462s  

  # user     27m12.526s  | stepping by 2

  # sys      13m11.193s /

  #

  # real    239m00.836s 

  # user    225m11.108s  | stepping by 1

  # sys     120m00.538s /

  #

  time 

  for d1 in {128..223..1} {240..255..1} ;do 

 #for d1 in {128..224..1} {239..255..1} ;do # force a valid UTF-8 (needs $B2b2,$B3b3) 

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {0..159..1} {192..255..1})"

         #B3b2range="$(> {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {0..127..1} {160..255..1})"

         #B3b2range="$(echo {0..128..1} {160..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      else

          B3b2range="$(echo {0..127..1} {192..255..1})"

         #B3b2range="$(echo {0..128..1} {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      fi

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {0..127..1} {192..255..1} ;do

         #for d3 in {0..128..1} {192..255..1} ;do # force a valid UTF-8 (needs $B2b1)

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r |

                  iconv -f UTF-8 2>/dev/null && { 

                    echo "ERROR: VALID UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 31; }

              #

          done

      done

  done

  echo



fi



#########################################################################



#  Brute force testing in the Astral Plane will take a VERY LONG time..

#  Perhaps selective testing is more appropriate, now that the previous tests 

#     have panned out okay... 

#  

#  4 Byte  UTF-8 values:

if ((B4==1)) ; then  

  echo "# Test 4 bytes for Valid UTF-8 values: ie. values which are in range"

  # ==================================================================

  # real    58m18.531s 

  # user    56m44.317s  | 

  # sys     27m29.867s /

  time 

  for d1 in {240..244} ;do

      #     bin       oct  hex  dec

      # lo  11110000  360   F0  240

      # hi  11110100  364   F4  244  -- F4 encodes some values greater than 0x10FFFF;

      #                                    such a sequence is invalid.

      B4b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B4b1 == "F0" ]] ; then

        B4b2range="$(echo {144..191})" ## f0 90 80 80  to  f0 bf bf bf

        #     bin       oct  hex  dec          010000  --  03FFFF 

        # lo  10010000  220   90  144  

        # hi  10111111  277   BF  191

        #                            

      elif [[ $B4b1 == "F4" ]] ; then

        B4b2range="$(echo {128..143})" ## f4 80 80 80  to  f4 8f bf bf

        #     bin       oct  hex  dec          100000  --  10FFFF 

        # lo  10000000  200   80  128  

        # hi  10001111  217   8F  143  -- F4 encodes some values greater than 0x10FFFF;

        #                                    such a sequence is invalid.

      else

        B4b2range="$(echo {128..191})" ## fx 80 80 80  to  f3 bf bf bf

        #     bin       oct  hex  dec          0C0000  --  0FFFFF

        # lo  10000000  200   80  128          0A0000

        # hi  10111111  277   BF  191

      fi

      #

      for d2 in $B4b2range ;do

          B4b2=$(printf "%0.2X" $d2)

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B4b3=$(printf "%0.2X" $d3)

              echo "${B4b1} ${B4b2} ${B4b3} xx"

              #

              for d4 in {128..191} ;do

                  #     bin       oct  hex  dec

                  # lo  10000000  200   80  128

                  # hi  10111111  277   BF  191

                  B4b4=$(printf "%0.2X" $d4)

                  #

                  echo -n "${B4b1}${B4b2}${B4b3}${B4b4}" |

                    xxd -p -u -r  |

                      iconv -f UTF-8 >/dev/null || { 

                        echo "ERROR: Invalid UTF-8 found: ${B4b1}${B4b2}${B4b3}${B4b4}"; exit 40; }

                  #

              done

          done

      done

  done

  echo "# Test 4 bytes for Valid UTF-8 values: END"

  echo

fi



########################################################################

# There is no test (yet) for negated range values in the astral plane. #  

#                           (all negated range values must be invalid) #

#  I won't bother; This was mainly for me to ge the general feel of    #     

#   the tests, and the final test below should flush anything out..    #

# Traversing the intire UTF-8 range takes quite a while...             #

#   so no need to do it twice (albeit in a slightly different manner)  #

########################################################################



################################

### The construction of:    ####

###  The Regular Expression ####

###      (de-construction?) ####

################################



#     BYTE 1                BYTE 2       BYTE 3      BYTE 4 

# 1: [x00-x7F]

#    ===========

#    ([x00-x7F])

#

# 2: [xC2-xDF]           [x80-xBF]

#    =================================

#    ([xC2-xDF][x80-xBF])

# 

# 3: [xE0]                [xA0-xBF]  [x80-xBF]   

#    [xED]                [x80-x9F]  [x80-xBF]

#    [xE1-xECxEE-xEF]  [x80-xBF]  [x80-xBF]

#    ==============================================

#    ((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))

#

# 4  [xF0]                [x90-xBF]  [x80-xBF]  [x80-xBF]    

#    [xF1-xF3]           [x80-xBF]  [x80-xBF]  [x80-xBF]

#    [xF4]                [x80-x8F]  [x80-xBF]  [x80-xBF]

#    ===========================================================

#    ((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))

#

# The final regex

# ===============

# 1-4:  (([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))

# 4-1:  (((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|([xC2-xDF][x80-xBF])|([x00-x7F]))





#######################################################################

#  The final Test; for a single character (multi chars to follow)     #  

#   Compare the return code of 'iconv' against the 'regex'            #

#   for the full range of 0x000000 to 0x10FFFF                        #

#                                                                     #     

#  Note; this script has 3 modes:                                     #

#        Run this test TWICE, set each mode Manually!                 #     

#                                                                     #     

#     1. Sequentially test every value from 0x000000 to 0x10FFFF      #     

#     2. Throw a spanner into the works! Force random byte patterns   #     

#     2. Throw a spanner into the works! Force random longer strings  #     

#        ==============================                               #     

#                                                                     #     

#  Note: The purpose of this routine is to determine if there is any  #

#        difference how 'iconv' and 'regex' handle the same data      #  

#                                                                     #     

#######################################################################

if ((Rx==1)) ; then

  # real    191m34.826s

  # user    158m24.114s

  # sys      83m10.676s

  time { 

  invalCt=0

  validCt=0

   failCt=0

  decBeg=$((0x00110000)) # incement by decimal integer

  decMax=$((0x7FFFFFFF)) # incement by decimal integer

  # 

  for ((CPDec=decBeg;CPDec<=decMax;CPDec+=13247)) ;do

      ((D==1)) && echo "=========================================================="

      #

      # Convert decimal integer '$CPDec' to Hex-digits; 6-long  (dec2hex)

      hexUTF32BE=$(printf '%0.8Xn' $CPDec)  # hexUTF32BE



      # progress count  

      if (((CPDec%$((0x1000)))==0)) ;then

          ((Test>2)) && echo

          echo "$hexUTF32BE  Test=$Test mode=${mode[$modebits]}            "

      fi

      if   ((Test==1 || Test==2 ))

      then # Test 1. Sequentially test every value from 0x000000 to 0x10FFFF

          #

          if   ((Test==2)) ; then

              bits=32

              UTF8="$( perl -C -e 'print chr 0x'$hexUTF32BE |

                perl -l -ne '/^(  [00-177]

                                | [300-337][200-277]

                                | [340-357][200-277]{2}

                                | [360-367][200-277]{3}

                                | [370-373][200-277]{4}

                                | [374-375][200-277]{5}

                               )*$/x and print' |xxd -p )"

              UTF8="${UTF8%0a}"

              [[ -n "$UTF8" ]] 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=



          elif ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              bits=16

              UTF8="$( echo -n "${hexUTF32BE:4}" |

                xxd -p -u -r |

                  iconv -f UTF-16BE -t UTF-8 2>/dev/null)" 

                    && rcIco16=0 || rcIco16=1  

                       rcIco32=

          else

              bits=32

              UTF8="$( echo -n "$hexUTF32BE" |

                xxd -p -u -r |

                  iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=

          fi

          # echo "1 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((${rcIco16}${rcIco32}!=0)) ;then

              # 'iconv -f UTF-16BE' failed produce a reliable UTF-8

              if ((bits==16)) ;then

                  ((D==1)) &&           echo "bits-$bits rcIconv: error    $hexUTF32BE .. 'strict' failed, now trying 'lax'"

                  #  iconv failed to create a  'srict' UTF-8 so   

                  #      try UTF-32BE to get a   'lax' UTF-8 pattern    

                  UTF8="$( echo -n "$hexUTF32BE" |

                    xxd -p -u -r |

                      iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                        && rcIco32=0 || rcIco32=1

                  #echo "2 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

                  if ((rcIco32!=0)) ;then

                      ((D==1)) &&               echo -n "bits-$bits rcIconv: Cannot gen UTF-8 for: $hexUTF32BE"

                      rcIco32=1

                  fi

              fi

          fi

          # echo "3 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((rcIco16==0 || rcIco32==0)) ;then

              # 'strict(16)' OR 'lax(32)'... 'iconv' managed to generate a UTF-8 pattern  

                  ((D==1)) &&       echo -n "bits-$bits rcIconv: pattern* $hexUTF32BE"

                  ((D==1)) &&       if [[ $bits == "16" && $rcIco32 == "0" ]] ;then 

                  echo " .. 'lax' UTF-8 produced a pattern"

              else

                  echo

              fi

               # regex test

              if ((modebits==strict)) ;then

                 #rxOut="$(echo -n "$UTF8" |perl -l -ne '/^(([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))*$/ or print' )"

                                     rxOut="$(echo -n "$UTF8" |

                  perl -l -ne '/^( ([x00-x7F])             # 1-byte pattern

                                  |([xC2-xDF][x80-xBF])  # 2-byte pattern

                                  |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))  # 3-byte pattern

                                  |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))        # 4-byte pattern

                                 )*$ /x or print' )"

               else

                  if ((Test==2)) ;then

                      rx="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ and print')"

                      [[ "$UTF8" != "$rx" ]] && rxOut="$UTF8" || rxOut=

                      rx="$(echo -n "$rx" |sed -e "s/(..)/1 /g")"  

                  else 

                      rxOut="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ or print' )"

                  fi

              fi

              if [[ "$rxOut" == "" ]] ;then

                ((D==1)) &&           echo "        rcRegex: ok"

                  rcRegex=0

              else

                  ((D==1)) &&           echo -n "bits-$bits rcRegex: error    $hexUTF32BE .. 'strict' failed,"

                  ((D==1)) &&           if [[  "12" == *$Test* ]] ;then 

                                            echo # "  (codepoint) Test $Test" 

                                        else

                                            echo

                                        fi

                  rcRegex=1

              fi

          fi

          #

      elif [[ $Test == 2 ]]

      then # Test 2. Throw a randomizing spanner into the works! 

          #          Then test the  arbitary bytes ASIS

          #

          hexLineRand="$(echo -n "$hexUTF32BE" |

            sed -re "s/(.)(.)(.)(.)(.)(.)(.)(.)/1n2n3n4n5n6n7n8/" |

              sort -R |

                tr -d 'n')"

          # 

      elif [[ $Test == 3 ]]

      then # Test 3. Test single UTF-16BE bytes in the range 0x00000000 to 0x7FFFFFFF

          #

          echo "Test 3 is not properly implemented yet.. Exiting"

          exit 99 

      else

          echo "ERROR: Invalid mode"

          exit

      fi

      #

      #

      if ((Test==1 || Test=2)) ;then

          if ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              ((rcIconv=rcIco16))

          else

              ((rcIconv=rcIco32))

          fi

          if ((rcRegex!=rcIconv)) ;then

              [[ $Test != 1 ]] && echo

              if ((rcRegex==1)) ;then

                  echo "ERROR: 'regex' ok, but NOT 'iconv': ${hexUTF32BE} "

              else

                  echo "ERROR: 'iconv' ok, but NOT 'regex': ${hexUTF32BE} "

              fi

              ((failCt++));

          elif ((rcRegex!=0)) ;then

            # ((invalCt++)); echo -ne "$hexUTF32BE  exit-codes $${rcIco16}${rcIco32}=,$rcRegext: $(printf "%0.8Xn" $invalCt)t$hexLine$(printf "%$(((mode3whi*2)-${#hexLine}))s")r"

              ((invalCt++)) 

          else

              ((validCt++)) 

          fi

          if   ((Test==1)) ;then

              echo -ne "$hexUTF32BE "    "mode=${mode[$modebits]}  test-return=($rcIconv,$rcRegex)   valid,invalid,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))          r"

          else 

              echo -ne "$hexUTF32BE $rx mode=${mode[$modebits]} test-return=($rcIconv,$rcRegex)  val,inval,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))r"

          fi

      fi

  done

  } # End time

fi

exit





share|improve this answer















Edit: I've fixed a typo-bug in the regex.. It needed a 'x80` not 80.



The regex to filter out invalid UTF-8 forms, for strict adherance to UTF-8, is as follows 



perl -l -ne '/

 ^( ([x00-x7F])              # 1-byte pattern

   |([xC2-xDF][x80-xBF])   # 2-byte pattern

   |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF])) # 3-byte pattern

   |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))       # 4-byte pattern

  )*$ /x or print'


Output (of key lines.from Test 1):



Codepoint

=========  

00001000  Test=1 mode=strict               valid,invalid,fail=(1000,0,0)          

0000E000  Test=1 mode=strict               valid,invalid,fail=(D800,800,0)          

0010FFFF  mode=strict  test-return=(0,0)   valid,invalid,fail=(10F800,800,0)



Q. How does one create test data to test a regex which filters invalid Unicode ?

A. Create your own UTF-8 test algorithm, and break it's rules...

Catch-22.. But then, how do you then test your test algorithm?



The regex, above, has been tested (using iconv as the reference) for every integer value from 0x00000 to 0x10FFFF.. This upper value being the maximum integer value of a Unicode Codepoint





  • In November 2003 UTF-8 was restricted by RFC 3629 to four bytes covering only the range U+0000 to U+10FFFF, in order to match the constraints of the UTF-16 character encoding.`


According to this wikipedia UTF-8 page,.




  • UTF-8 encodes each of the 1,112,064 code points in the Unicode character set, using one to four 8-bit bytes


This numeber (1,112,064) equates to a range 0x000000 to 0x10F7FF, which is 0x0800 shy of the actual maximum integer-value for the highest Unicode Codepoint: 0x10FFFF



This block of integers is missing from the Unicode Codepoints spectrum, because of the need for the UTF-16 encoding to step beyond its original design intent via a system called surrogate pairs. A block of 0x0800 integers has been reserved to be used by UTF-16.. This block spans the range0x00D800 to 0x00DFFF. None of these inteters are legal Unicode values, and are therefore invalid UTF-8 values.



In Test 1, the regex has been tested against every number in the range of Unicode Codepoints, and it matches exectly the results of iconv .. ie. 0x010F7FF valid values, and 0x000800 invalid values.



However, the issue now arises of, *How does the regex handle Out-Of-Range UTF-8 Value; above 0x010FFFF (UTF-8 can extend to 6 bytes, with a maximum integer value of 0x7FFFFFFF?

To generate the necessary *non-unicode UTF-8 byte values, I've used the following command: 



  perl -C -e 'print chr 0x'$hexUTF32BE


To test their validity (in some fashion), I've used Gilles' UTF-8 regex... 



  perl -l -ne '/

   ^( [00-177]                 # 1-byte pattern

     |[300-337][200-277]      # 2-byte pattern

     |[340-357][200-277]{2}   # 3-byte pattern

     |[360-367][200-277]{3}   # 4-byte pattern

     |[370-373][200-277]{4}   # 5-byte pattern

     |[374-375][200-277]{5}   # 6-byte pattern

    )*$ /x or print'


The output of 'perl's print chr' matches the filtering of Gilles' regex.. One reinforces the validity of the other..
I can't use iconv because it only handles the valid-Unicode Standard subset of the broader (original) UTF-8 standard...



The nunbers involved are rather large, so I've tested top-of-range, bottom-of-range, and several scans stepping by increments such as, 11111, 13579, 33333, 53441... The results all match, so now all that remains is to test the regex against these out-of-range UTF-8-style values (invalid for Unicode, and therefore also invalid for strict UTF-8 itself) ..




Here are the test modules: 



[[ "$(locale charmap)" != "UTF-8" ]] && { echo "ERROR: locale must be UTF-8, but it is $(locale charmap)"; exit 1; }



# Testing the UTF-8 regex

#

# Tests to check that the observed byte-ranges (above) have

#  been  accurately observed and included in the test code and final regex. 

# =========================================================================

: 2 bytes; B2=0 #  run-test=1   do-not-test=0

: 3 bytes; B3=0 #  run-test=1   do-not-test=0

: 4 bytes; B4=0 #  run-test=1   do-not-test=0 



:   regex; Rx=1 #  run-test=1   do-not-test=0



           ((strict=16)); mode[$strict]=strict # iconv -f UTF-16BE  then iconv -f UTF-32BE beyond 0xFFFF)

           ((   lax=32)); mode[$lax]=lax       # iconv -f UTF-32BE  only)



          # modebits=$strict

                  # UTF-8, in relation to UTF-16 has invalid values

                  # modebits=$strict automatically shifts to modebits=$lax

                  # when the tested integer exceeds 0xFFFF

          # modebits=$lax 

                  # UTF-8, in relation to UTF-32, has no restrictione





           # Test 1 Sequentially tests a range of Big-Endian integers

           #      * Unicode Codepoints are a subset ofBig-Endian integers            

           #        ( based on 'iconv' -f UTF-32BE -f UTF-8 )    

           # Note: strict UTF-8 has a few quirks because of UTF-16

                    #    Set modebits=16 to "strictly" test the low range



             Test=1; modebits=$strict

           # Test=2; modebits=$lax

           # Test=3

              mode3wlo=$(( 1*4)) # minimum chars * 4 ( '4' is for UTF-32BE )

              mode3whi=$((10*4)) # minimum chars * 4 ( '4' is for UTF-32BE )





#########################################################################  



# 1 byte  UTF-8 values: Nothing to do; no complexities.



#########################################################################



#  2 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B2==1)) ; then  

  echo "# Test 2 bytes for Valid UTF-8 values: ie. values which are in range"

  # =========================================================================

  time 

  for d1 in {194..223} ;do

      #     bin       oct  hex  dec

      # lo  11000010  302   C2  194

      # hi  11011111  337   DF  223

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {128..191} ;do

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r  |

              iconv -f UTF-8 >/dev/null || { 

                echo "ERROR: Invalid UTF-8 found: ${B2b1}${B2b2}"; exit 20; }

          #

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 2 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  time 

  for d1 in {128..193} {224..255} ;do 

 #for d1 in {128..194} {224..255} ;do # force a valid UTF-8 (needs $B2b2) 

      B2b1=$(printf "%0.2X" $d1)

      #

      for d2 in {0..127} {192..255} ;do

     #for d2 in {0..128} {192..255} ;do # force a valid UTF-8 (needs $B2b1)

          B2b2=$(printf "%0.2X" $d2)

          #

          echo -n "${B2b1}${B2b2}" |

            xxd -p -u -r |

              iconv -f UTF-8 2>/dev/null && { 

                echo "ERROR: VALID UTF-8 found: ${B2b1}${B2b2}"; exit 21; }

          #

      done

  done

  echo

fi



#########################################################################



#  3 Byte  UTF-8 values:  Verifying that I've got the right range values.

if ((B3==1)) ; then  

  echo "# Test 3 bytes for Valid UTF-8 values: ie. values which are in range"

  # ========================================================================

  time 

  for d1 in {224..239} ;do

      #     bin       oct  hex  dec

      # lo  11100000  340   E0  224

      # hi  11101111  357   EF  239

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {160..191})"

          #     bin       oct  hex  dec  

          # lo  10100000  240   A0  160  

          # hi  10111111  277   BF  191

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {128..159})"

          #     bin       oct  hex  dec  

          # lo  10000000  200   80  128  

          # hi  10011111  237   9F  159

      else

          B3b2range="$(echo {128..191})"

          #     bin       oct  hex  dec

          # lo  10000000  200   80  128

          # hi  10111111  277   BF  191

      fi

      # 

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r  |

                  iconv -f UTF-8 >/dev/null || { 

                    echo "ERROR: Invalid UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 30; }

              #

          done

      done

  done

  echo



  # Now do a negated test.. This takes longer, because there are more values.

  echo "# Test 3 bytes for Invalid values: ie. values which are out of range"

  # =========================================================================

  # Note: 'iconv' will treat a leading  x00-x7F as a valid leading single,

  #   so this negated test primes the first UTF-8 byte with values starting at x80

  #

  # real     26m28.462s  

  # user     27m12.526s  | stepping by 2

  # sys      13m11.193s /

  #

  # real    239m00.836s 

  # user    225m11.108s  | stepping by 1

  # sys     120m00.538s /

  #

  time 

  for d1 in {128..223..1} {240..255..1} ;do 

 #for d1 in {128..224..1} {239..255..1} ;do # force a valid UTF-8 (needs $B2b2,$B3b3) 

      B3b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B3b1 == "E0" ]] ; then

          B3b2range="$(echo {0..159..1} {192..255..1})"

         #B3b2range="$(> {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      elif [[ $B3b1 == "ED" ]] ; then

          B3b2range="$(echo {0..127..1} {160..255..1})"

         #B3b2range="$(echo {0..128..1} {160..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      else

          B3b2range="$(echo {0..127..1} {192..255..1})"

         #B3b2range="$(echo {0..128..1} {192..255..1})" # force a valid UTF-8 (needs $B3b1,$B3b3)

      fi

      for d2 in $B3b2range ;do

          B3b2=$(printf "%0.2X" $d2)

          echo "${B3b1} ${B3b2} xx"

          #

          for d3 in {0..127..1} {192..255..1} ;do

         #for d3 in {0..128..1} {192..255..1} ;do # force a valid UTF-8 (needs $B2b1)

              B3b3=$(printf "%0.2X" $d3)

              #

              echo -n "${B3b1}${B3b2}${B3b3}" |

                xxd -p -u -r |

                  iconv -f UTF-8 2>/dev/null && { 

                    echo "ERROR: VALID UTF-8 found: ${B3b1}${B3b2}${B3b3}"; exit 31; }

              #

          done

      done

  done

  echo



fi



#########################################################################



#  Brute force testing in the Astral Plane will take a VERY LONG time..

#  Perhaps selective testing is more appropriate, now that the previous tests 

#     have panned out okay... 

#  

#  4 Byte  UTF-8 values:

if ((B4==1)) ; then  

  echo "# Test 4 bytes for Valid UTF-8 values: ie. values which are in range"

  # ==================================================================

  # real    58m18.531s 

  # user    56m44.317s  | 

  # sys     27m29.867s /

  time 

  for d1 in {240..244} ;do

      #     bin       oct  hex  dec

      # lo  11110000  360   F0  240

      # hi  11110100  364   F4  244  -- F4 encodes some values greater than 0x10FFFF;

      #                                    such a sequence is invalid.

      B4b1=$(printf "%0.2X" $d1)

      #

      if   [[ $B4b1 == "F0" ]] ; then

        B4b2range="$(echo {144..191})" ## f0 90 80 80  to  f0 bf bf bf

        #     bin       oct  hex  dec          010000  --  03FFFF 

        # lo  10010000  220   90  144  

        # hi  10111111  277   BF  191

        #                            

      elif [[ $B4b1 == "F4" ]] ; then

        B4b2range="$(echo {128..143})" ## f4 80 80 80  to  f4 8f bf bf

        #     bin       oct  hex  dec          100000  --  10FFFF 

        # lo  10000000  200   80  128  

        # hi  10001111  217   8F  143  -- F4 encodes some values greater than 0x10FFFF;

        #                                    such a sequence is invalid.

      else

        B4b2range="$(echo {128..191})" ## fx 80 80 80  to  f3 bf bf bf

        #     bin       oct  hex  dec          0C0000  --  0FFFFF

        # lo  10000000  200   80  128          0A0000

        # hi  10111111  277   BF  191

      fi

      #

      for d2 in $B4b2range ;do

          B4b2=$(printf "%0.2X" $d2)

          #

          for d3 in {128..191} ;do

              #     bin       oct  hex  dec

              # lo  10000000  200   80  128

              # hi  10111111  277   BF  191

              B4b3=$(printf "%0.2X" $d3)

              echo "${B4b1} ${B4b2} ${B4b3} xx"

              #

              for d4 in {128..191} ;do

                  #     bin       oct  hex  dec

                  # lo  10000000  200   80  128

                  # hi  10111111  277   BF  191

                  B4b4=$(printf "%0.2X" $d4)

                  #

                  echo -n "${B4b1}${B4b2}${B4b3}${B4b4}" |

                    xxd -p -u -r  |

                      iconv -f UTF-8 >/dev/null || { 

                        echo "ERROR: Invalid UTF-8 found: ${B4b1}${B4b2}${B4b3}${B4b4}"; exit 40; }

                  #

              done

          done

      done

  done

  echo "# Test 4 bytes for Valid UTF-8 values: END"

  echo

fi



########################################################################

# There is no test (yet) for negated range values in the astral plane. #  

#                           (all negated range values must be invalid) #

#  I won't bother; This was mainly for me to ge the general feel of    #     

#   the tests, and the final test below should flush anything out..    #

# Traversing the intire UTF-8 range takes quite a while...             #

#   so no need to do it twice (albeit in a slightly different manner)  #

########################################################################



################################

### The construction of:    ####

###  The Regular Expression ####

###      (de-construction?) ####

################################



#     BYTE 1                BYTE 2       BYTE 3      BYTE 4 

# 1: [x00-x7F]

#    ===========

#    ([x00-x7F])

#

# 2: [xC2-xDF]           [x80-xBF]

#    =================================

#    ([xC2-xDF][x80-xBF])

# 

# 3: [xE0]                [xA0-xBF]  [x80-xBF]   

#    [xED]                [x80-x9F]  [x80-xBF]

#    [xE1-xECxEE-xEF]  [x80-xBF]  [x80-xBF]

#    ==============================================

#    ((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))

#

# 4  [xF0]                [x90-xBF]  [x80-xBF]  [x80-xBF]    

#    [xF1-xF3]           [x80-xBF]  [x80-xBF]  [x80-xBF]

#    [xF4]                [x80-x8F]  [x80-xBF]  [x80-xBF]

#    ===========================================================

#    ((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))

#

# The final regex

# ===============

# 1-4:  (([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))

# 4-1:  (((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|([xC2-xDF][x80-xBF])|([x00-x7F]))





#######################################################################

#  The final Test; for a single character (multi chars to follow)     #  

#   Compare the return code of 'iconv' against the 'regex'            #

#   for the full range of 0x000000 to 0x10FFFF                        #

#                                                                     #     

#  Note; this script has 3 modes:                                     #

#        Run this test TWICE, set each mode Manually!                 #     

#                                                                     #     

#     1. Sequentially test every value from 0x000000 to 0x10FFFF      #     

#     2. Throw a spanner into the works! Force random byte patterns   #     

#     2. Throw a spanner into the works! Force random longer strings  #     

#        ==============================                               #     

#                                                                     #     

#  Note: The purpose of this routine is to determine if there is any  #

#        difference how 'iconv' and 'regex' handle the same data      #  

#                                                                     #     

#######################################################################

if ((Rx==1)) ; then

  # real    191m34.826s

  # user    158m24.114s

  # sys      83m10.676s

  time { 

  invalCt=0

  validCt=0

   failCt=0

  decBeg=$((0x00110000)) # incement by decimal integer

  decMax=$((0x7FFFFFFF)) # incement by decimal integer

  # 

  for ((CPDec=decBeg;CPDec<=decMax;CPDec+=13247)) ;do

      ((D==1)) && echo "=========================================================="

      #

      # Convert decimal integer '$CPDec' to Hex-digits; 6-long  (dec2hex)

      hexUTF32BE=$(printf '%0.8Xn' $CPDec)  # hexUTF32BE



      # progress count  

      if (((CPDec%$((0x1000)))==0)) ;then

          ((Test>2)) && echo

          echo "$hexUTF32BE  Test=$Test mode=${mode[$modebits]}            "

      fi

      if   ((Test==1 || Test==2 ))

      then # Test 1. Sequentially test every value from 0x000000 to 0x10FFFF

          #

          if   ((Test==2)) ; then

              bits=32

              UTF8="$( perl -C -e 'print chr 0x'$hexUTF32BE |

                perl -l -ne '/^(  [00-177]

                                | [300-337][200-277]

                                | [340-357][200-277]{2}

                                | [360-367][200-277]{3}

                                | [370-373][200-277]{4}

                                | [374-375][200-277]{5}

                               )*$/x and print' |xxd -p )"

              UTF8="${UTF8%0a}"

              [[ -n "$UTF8" ]] 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=



          elif ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              bits=16

              UTF8="$( echo -n "${hexUTF32BE:4}" |

                xxd -p -u -r |

                  iconv -f UTF-16BE -t UTF-8 2>/dev/null)" 

                    && rcIco16=0 || rcIco16=1  

                       rcIco32=

          else

              bits=32

              UTF8="$( echo -n "$hexUTF32BE" |

                xxd -p -u -r |

                  iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                    && rcIco32=0 || rcIco32=1

                       rcIco16=

          fi

          # echo "1 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((${rcIco16}${rcIco32}!=0)) ;then

              # 'iconv -f UTF-16BE' failed produce a reliable UTF-8

              if ((bits==16)) ;then

                  ((D==1)) &&           echo "bits-$bits rcIconv: error    $hexUTF32BE .. 'strict' failed, now trying 'lax'"

                  #  iconv failed to create a  'srict' UTF-8 so   

                  #      try UTF-32BE to get a   'lax' UTF-8 pattern    

                  UTF8="$( echo -n "$hexUTF32BE" |

                    xxd -p -u -r |

                      iconv -f UTF-32BE -t UTF-8 2>/dev/null)" 

                        && rcIco32=0 || rcIco32=1

                  #echo "2 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

                  if ((rcIco32!=0)) ;then

                      ((D==1)) &&               echo -n "bits-$bits rcIconv: Cannot gen UTF-8 for: $hexUTF32BE"

                      rcIco32=1

                  fi

              fi

          fi

          # echo "3 mode=${mode[$modebits]}-$bits  rcIconv: (${rcIco16},${rcIco32})  $hexUTF32BE "

          #

          #

          #

          if ((rcIco16==0 || rcIco32==0)) ;then

              # 'strict(16)' OR 'lax(32)'... 'iconv' managed to generate a UTF-8 pattern  

                  ((D==1)) &&       echo -n "bits-$bits rcIconv: pattern* $hexUTF32BE"

                  ((D==1)) &&       if [[ $bits == "16" && $rcIco32 == "0" ]] ;then 

                  echo " .. 'lax' UTF-8 produced a pattern"

              else

                  echo

              fi

               # regex test

              if ((modebits==strict)) ;then

                 #rxOut="$(echo -n "$UTF8" |perl -l -ne '/^(([x00-x7F])|([xC2-xDF][x80-xBF])|((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))|((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2})))*$/ or print' )"

                                     rxOut="$(echo -n "$UTF8" |

                  perl -l -ne '/^( ([x00-x7F])             # 1-byte pattern

                                  |([xC2-xDF][x80-xBF])  # 2-byte pattern

                                  |((([xE0][xA0-xBF])|([xED][x80-x9F])|([xE1-xECxEE-xEF][x80-xBF]))([x80-xBF]))  # 3-byte pattern

                                  |((([xF0][x90-xBF])|([xF1-xF3][x80-xBF])|([xF4][x80-x8F]))([x80-xBF]{2}))        # 4-byte pattern

                                 )*$ /x or print' )"

               else

                  if ((Test==2)) ;then

                      rx="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ and print')"

                      [[ "$UTF8" != "$rx" ]] && rxOut="$UTF8" || rxOut=

                      rx="$(echo -n "$rx" |sed -e "s/(..)/1 /g")"  

                  else 

                      rxOut="$(echo -n "$UTF8" |perl -l -ne '/^([00-177]|[300-337][200-277]|[340-357][200-277]{2}|[360-367][200-277]{3}|[370-373][200-277]{4}|[374-375][200-277]{5})*$/ or print' )"

                  fi

              fi

              if [[ "$rxOut" == "" ]] ;then

                ((D==1)) &&           echo "        rcRegex: ok"

                  rcRegex=0

              else

                  ((D==1)) &&           echo -n "bits-$bits rcRegex: error    $hexUTF32BE .. 'strict' failed,"

                  ((D==1)) &&           if [[  "12" == *$Test* ]] ;then 

                                            echo # "  (codepoint) Test $Test" 

                                        else

                                            echo

                                        fi

                  rcRegex=1

              fi

          fi

          #

      elif [[ $Test == 2 ]]

      then # Test 2. Throw a randomizing spanner into the works! 

          #          Then test the  arbitary bytes ASIS

          #

          hexLineRand="$(echo -n "$hexUTF32BE" |

            sed -re "s/(.)(.)(.)(.)(.)(.)(.)(.)/1n2n3n4n5n6n7n8/" |

              sort -R |

                tr -d 'n')"

          # 

      elif [[ $Test == 3 ]]

      then # Test 3. Test single UTF-16BE bytes in the range 0x00000000 to 0x7FFFFFFF

          #

          echo "Test 3 is not properly implemented yet.. Exiting"

          exit 99 

      else

          echo "ERROR: Invalid mode"

          exit

      fi

      #

      #

      if ((Test==1 || Test=2)) ;then

          if ((modebits==strict && CPDec<=$((0xFFFF)))) ;then

              ((rcIconv=rcIco16))

          else

              ((rcIconv=rcIco32))

          fi

          if ((rcRegex!=rcIconv)) ;then

              [[ $Test != 1 ]] && echo

              if ((rcRegex==1)) ;then

                  echo "ERROR: 'regex' ok, but NOT 'iconv': ${hexUTF32BE} "

              else

                  echo "ERROR: 'iconv' ok, but NOT 'regex': ${hexUTF32BE} "

              fi

              ((failCt++));

          elif ((rcRegex!=0)) ;then

            # ((invalCt++)); echo -ne "$hexUTF32BE  exit-codes $${rcIco16}${rcIco32}=,$rcRegext: $(printf "%0.8Xn" $invalCt)t$hexLine$(printf "%$(((mode3whi*2)-${#hexLine}))s")r"

              ((invalCt++)) 

          else

              ((validCt++)) 

          fi

          if   ((Test==1)) ;then

              echo -ne "$hexUTF32BE "    "mode=${mode[$modebits]}  test-return=($rcIconv,$rcRegex)   valid,invalid,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))          r"

          else 

              echo -ne "$hexUTF32BE $rx mode=${mode[$modebits]} test-return=($rcIconv,$rcRegex)  val,inval,fail=($(printf "%X" $validCt),$(printf "%X" $invalCt),$(printf "%X" $failCt))r"

          fi

      fi

  done

  } # End time

fi

exit






share|improve this answer














share|improve this answer



share|improve this answer








edited May 6 '11 at 8:29

























answered May 3 '11 at 18:50









Peter.OPeter.O

18.9k1791144




18.9k1791144













  • The main problem with my regexp is that it allowed some forbidden sequences such as 300200 (really bad: that's code point 0 not expressed with a null byte!). I think your regexp rejects them correctly.

    – Gilles
    May 8 '11 at 12:38



















  • The main problem with my regexp is that it allowed some forbidden sequences such as 300200 (really bad: that's code point 0 not expressed with a null byte!). I think your regexp rejects them correctly.

    – Gilles
    May 8 '11 at 12:38

















The main problem with my regexp is that it allowed some forbidden sequences such as 300200 (really bad: that's code point 0 not expressed with a null byte!). I think your regexp rejects them correctly.

– Gilles
May 8 '11 at 12:38





The main problem with my regexp is that it allowed some forbidden sequences such as 300200 (really bad: that's code point 0 not expressed with a null byte!). I think your regexp rejects them correctly.

– Gilles
May 8 '11 at 12:38











6





+100









I find uconv (in icu-devtools package in Debian) useful to inspect UTF-8 data:



$ print '\xE9 xe9 u20ac ud800udc00 U110000' |

    uconv --callback escape-c -t us

xE9 xE9 u20ac xEDxA0x80xEDxB0x80 xF4x90x80x80


(The xs help spotting the invalid characters (except for the false positive voluntarily introduced with a literal xE9 above)).



(plenty of other nice usages).






share|improve this answer


























  • I think recode can be used similarly - except that I think it should fail if asked to translate an invalid multibyte sequence. I'm not sure though; it won't fail for print...|recode u8..u8/x4 for example (which just does a hexdump as you do above) because it doesn't do anything but iconv data data, but it does fail like recode u8..u2..u8/x4 because it translates then prints. But I don't know enough about it to be sure - and there are a lot of possibilities.

    – mikeserv
    Dec 20 '14 at 19:29











  • If I have a file, say, test.txt. How should I suppose to find the invalid character using your solution? What does us in your code mean?

    – jdhao
    Dec 26 '17 at 7:51











  • @Hao, us means United States, that is short for ASCII. It converts the input into a ASCII one where the non-ASCII characters are converted to uXXXX notation and the non-characters to xXX.

    – Stéphane Chazelas
    Dec 26 '17 at 14:19











  • Where should I put my file to use your script? Is the last line in the code block the output of your code? It is a little confusing to me.

    – jdhao
    Dec 26 '17 at 14:27
















6





+100









I find uconv (in icu-devtools package in Debian) useful to inspect UTF-8 data:



$ print '\xE9 xe9 u20ac ud800udc00 U110000' |

    uconv --callback escape-c -t us

xE9 xE9 u20ac xEDxA0x80xEDxB0x80 xF4x90x80x80


(The xs help spotting the invalid characters (except for the false positive voluntarily introduced with a literal xE9 above)).



(plenty of other nice usages).






share|improve this answer


























  • I think recode can be used similarly - except that I think it should fail if asked to translate an invalid multibyte sequence. I'm not sure though; it won't fail for print...|recode u8..u8/x4 for example (which just does a hexdump as you do above) because it doesn't do anything but iconv data data, but it does fail like recode u8..u2..u8/x4 because it translates then prints. But I don't know enough about it to be sure - and there are a lot of possibilities.

    – mikeserv
    Dec 20 '14 at 19:29











  • If I have a file, say, test.txt. How should I suppose to find the invalid character using your solution? What does us in your code mean?

    – jdhao
    Dec 26 '17 at 7:51











  • @Hao, us means United States, that is short for ASCII. It converts the input into a ASCII one where the non-ASCII characters are converted to uXXXX notation and the non-characters to xXX.

    – Stéphane Chazelas
    Dec 26 '17 at 14:19











  • Where should I put my file to use your script? Is the last line in the code block the output of your code? It is a little confusing to me.

    – jdhao
    Dec 26 '17 at 14:27














6





+100







6





+100



6




+100





I find uconv (in icu-devtools package in Debian) useful to inspect UTF-8 data:



$ print '\xE9 xe9 u20ac ud800udc00 U110000' |

    uconv --callback escape-c -t us

xE9 xE9 u20ac xEDxA0x80xEDxB0x80 xF4x90x80x80


(The xs help spotting the invalid characters (except for the false positive voluntarily introduced with a literal xE9 above)).



(plenty of other nice usages).






share|improve this answer















I find uconv (in icu-devtools package in Debian) useful to inspect UTF-8 data:



$ print '\xE9 xe9 u20ac ud800udc00 U110000' |

    uconv --callback escape-c -t us

xE9 xE9 u20ac xEDxA0x80xEDxB0x80 xF4x90x80x80


(The xs help spotting the invalid characters (except for the false positive voluntarily introduced with a literal xE9 above)).



(plenty of other nice usages).







share|improve this answer














share|improve this answer



share|improve this answer








edited Dec 20 '14 at 19:09

























answered Dec 19 '14 at 23:46









Stéphane ChazelasStéphane Chazelas

302k56567919




302k56567919













  • I think recode can be used similarly - except that I think it should fail if asked to translate an invalid multibyte sequence. I'm not sure though; it won't fail for print...|recode u8..u8/x4 for example (which just does a hexdump as you do above) because it doesn't do anything but iconv data data, but it does fail like recode u8..u2..u8/x4 because it translates then prints. But I don't know enough about it to be sure - and there are a lot of possibilities.

    – mikeserv
    Dec 20 '14 at 19:29











  • If I have a file, say, test.txt. How should I suppose to find the invalid character using your solution? What does us in your code mean?

    – jdhao
    Dec 26 '17 at 7:51











  • @Hao, us means United States, that is short for ASCII. It converts the input into a ASCII one where the non-ASCII characters are converted to uXXXX notation and the non-characters to xXX.

    – Stéphane Chazelas
    Dec 26 '17 at 14:19











  • Where should I put my file to use your script? Is the last line in the code block the output of your code? It is a little confusing to me.

    – jdhao
    Dec 26 '17 at 14:27



















  • I think recode can be used similarly - except that I think it should fail if asked to translate an invalid multibyte sequence. I'm not sure though; it won't fail for print...|recode u8..u8/x4 for example (which just does a hexdump as you do above) because it doesn't do anything but iconv data data, but it does fail like recode u8..u2..u8/x4 because it translates then prints. But I don't know enough about it to be sure - and there are a lot of possibilities.

    – mikeserv
    Dec 20 '14 at 19:29











  • If I have a file, say, test.txt. How should I suppose to find the invalid character using your solution? What does us in your code mean?

    – jdhao
    Dec 26 '17 at 7:51











  • @Hao, us means United States, that is short for ASCII. It converts the input into a ASCII one where the non-ASCII characters are converted to uXXXX notation and the non-characters to xXX.

    – Stéphane Chazelas
    Dec 26 '17 at 14:19











  • Where should I put my file to use your script? Is the last line in the code block the output of your code? It is a little confusing to me.

    – jdhao
    Dec 26 '17 at 14:27

















I think recode can be used similarly - except that I think it should fail if asked to translate an invalid multibyte sequence. I'm not sure though; it won't fail for print...|recode u8..u8/x4 for example (which just does a hexdump as you do above) because it doesn't do anything but iconv data data, but it does fail like recode u8..u2..u8/x4 because it translates then prints. But I don't know enough about it to be sure - and there are a lot of possibilities.

– mikeserv
Dec 20 '14 at 19:29





I think recode can be used similarly - except that I think it should fail if asked to translate an invalid multibyte sequence. I'm not sure though; it won't fail for print...|recode u8..u8/x4 for example (which just does a hexdump as you do above) because it doesn't do anything but iconv data data, but it does fail like recode u8..u2..u8/x4 because it translates then prints. But I don't know enough about it to be sure - and there are a lot of possibilities.

– mikeserv
Dec 20 '14 at 19:29













If I have a file, say, test.txt. How should I suppose to find the invalid character using your solution? What does us in your code mean?

– jdhao
Dec 26 '17 at 7:51





If I have a file, say, test.txt. How should I suppose to find the invalid character using your solution? What does us in your code mean?

– jdhao
Dec 26 '17 at 7:51













@Hao, us means United States, that is short for ASCII. It converts the input into a ASCII one where the non-ASCII characters are converted to uXXXX notation and the non-characters to xXX.

– Stéphane Chazelas
Dec 26 '17 at 14:19





@Hao, us means United States, that is short for ASCII. It converts the input into a ASCII one where the non-ASCII characters are converted to uXXXX notation and the non-characters to xXX.

– Stéphane Chazelas
Dec 26 '17 at 14:19













Where should I put my file to use your script? Is the last line in the code block the output of your code? It is a little confusing to me.

– jdhao
Dec 26 '17 at 14:27





Where should I put my file to use your script? Is the last line in the code block the output of your code? It is a little confusing to me.

– jdhao
Dec 26 '17 at 14:27











3














Python has had a built-in unicode function since version 2.0.



#!/usr/bin/env python2

import sys

for line in sys.stdin:

    try:

        unicode(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.write(line)


In Python 3, unicode has been folded into str. It needs to be passed a bytes-like object, here the underlying buffer objects for the standard descriptors.



#!/usr/bin/env python3

import sys

for line in sys.stdin.buffer:

    try:

        str(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.buffer.write(line)





share|improve this answer
























  • The python 2 one fails to flag UTF-8 encoded UTF-16 surrogate non-characters (at least with 2.7.6).

    – Stéphane Chazelas
    Dec 19 '14 at 23:12











  • @StéphaneChazelas Dammit. Thanks. I've only run nominal tests so far, I'll run Peter's test battery later.

    – Gilles
    Dec 19 '14 at 23:14
















3














Python has had a built-in unicode function since version 2.0.



#!/usr/bin/env python2

import sys

for line in sys.stdin:

    try:

        unicode(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.write(line)


In Python 3, unicode has been folded into str. It needs to be passed a bytes-like object, here the underlying buffer objects for the standard descriptors.



#!/usr/bin/env python3

import sys

for line in sys.stdin.buffer:

    try:

        str(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.buffer.write(line)





share|improve this answer
























  • The python 2 one fails to flag UTF-8 encoded UTF-16 surrogate non-characters (at least with 2.7.6).

    – Stéphane Chazelas
    Dec 19 '14 at 23:12











  • @StéphaneChazelas Dammit. Thanks. I've only run nominal tests so far, I'll run Peter's test battery later.

    – Gilles
    Dec 19 '14 at 23:14














3












3








3







Python has had a built-in unicode function since version 2.0.



#!/usr/bin/env python2

import sys

for line in sys.stdin:

    try:

        unicode(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.write(line)


In Python 3, unicode has been folded into str. It needs to be passed a bytes-like object, here the underlying buffer objects for the standard descriptors.



#!/usr/bin/env python3

import sys

for line in sys.stdin.buffer:

    try:

        str(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.buffer.write(line)





share|improve this answer













Python has had a built-in unicode function since version 2.0.



#!/usr/bin/env python2

import sys

for line in sys.stdin:

    try:

        unicode(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.write(line)


In Python 3, unicode has been folded into str. It needs to be passed a bytes-like object, here the underlying buffer objects for the standard descriptors.



#!/usr/bin/env python3

import sys

for line in sys.stdin.buffer:

    try:

        str(line, 'utf-8')

    except UnicodeDecodeError:

        sys.stdout.buffer.write(line)






share|improve this answer












share|improve this answer



share|improve this answer










answered Dec 19 '14 at 23:05









GillesGilles

533k12810721594




533k12810721594













  • The python 2 one fails to flag UTF-8 encoded UTF-16 surrogate non-characters (at least with 2.7.6).

    – Stéphane Chazelas
    Dec 19 '14 at 23:12











  • @StéphaneChazelas Dammit. Thanks. I've only run nominal tests so far, I'll run Peter's test battery later.

    – Gilles
    Dec 19 '14 at 23:14



















  • The python 2 one fails to flag UTF-8 encoded UTF-16 surrogate non-characters (at least with 2.7.6).

    – Stéphane Chazelas
    Dec 19 '14 at 23:12











  • @StéphaneChazelas Dammit. Thanks. I've only run nominal tests so far, I'll run Peter's test battery later.

    – Gilles
    Dec 19 '14 at 23:14

















The python 2 one fails to flag UTF-8 encoded UTF-16 surrogate non-characters (at least with 2.7.6).

– Stéphane Chazelas
Dec 19 '14 at 23:12





The python 2 one fails to flag UTF-8 encoded UTF-16 surrogate non-characters (at least with 2.7.6).

– Stéphane Chazelas
Dec 19 '14 at 23:12













@StéphaneChazelas Dammit. Thanks. I've only run nominal tests so far, I'll run Peter's test battery later.

– Gilles
Dec 19 '14 at 23:14





@StéphaneChazelas Dammit. Thanks. I've only run nominal tests so far, I'll run Peter's test battery later.

– Gilles
Dec 19 '14 at 23:14











0














I came across similar problem (detail in "Context" section) and arrived with following ftfy_line_by_line.py solution:



#!/usr/bin/env python3

import ftfy, sys

with open(sys.argv[1], mode='rt', encoding='utf8', errors='replace') as f:

  for line in f:

    sys.stdout.buffer.write(ftfy.fix_text(line).encode('utf8', 'replace'))

    #print(ftfy.fix_text(line).rstrip().decode(encoding="utf-8", errors="replace"))


Using encode+replace + ftfy to auto-fix Mojibake and other corrections.



Context



I've collected >10GiB CSV of basic filesystem metadata using following gen_basic_files_metadata.csv.sh script, running essentially:



find "${path}" -type f -exec stat --format="%i,%Y,%s,${hostname},%m,%n" "{}" ;


The trouble I had was with inconsistent encoding of filenames across file systems, causing UnicodeDecodeError when processing further with python applications (csvsql to be more specific).



Therefore I applied above ftfy script, and it took



Please note ftfy is pretty slow, processing those >10GiB took:



real    147m35.182s

user    146m14.329s

sys     2m8.713s


while sha256sum for comparison:



real    6m28.897s

user    1m9.273s

sys     0m6.210s


on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz + 16GiB RAM (and data on external drive)






share|improve this answer


























  • And yes, I know that this find command will not properly encode filenames containing quotes according to csv standard

    – Grzegorz Wierzowiecki
    May 25 '17 at 14:12
















0














I came across similar problem (detail in "Context" section) and arrived with following ftfy_line_by_line.py solution:



#!/usr/bin/env python3

import ftfy, sys

with open(sys.argv[1], mode='rt', encoding='utf8', errors='replace') as f:

  for line in f:

    sys.stdout.buffer.write(ftfy.fix_text(line).encode('utf8', 'replace'))

    #print(ftfy.fix_text(line).rstrip().decode(encoding="utf-8", errors="replace"))


Using encode+replace + ftfy to auto-fix Mojibake and other corrections.



Context



I've collected >10GiB CSV of basic filesystem metadata using following gen_basic_files_metadata.csv.sh script, running essentially:



find "${path}" -type f -exec stat --format="%i,%Y,%s,${hostname},%m,%n" "{}" ;


The trouble I had was with inconsistent encoding of filenames across file systems, causing UnicodeDecodeError when processing further with python applications (csvsql to be more specific).



Therefore I applied above ftfy script, and it took



Please note ftfy is pretty slow, processing those >10GiB took:



real    147m35.182s

user    146m14.329s

sys     2m8.713s


while sha256sum for comparison:



real    6m28.897s

user    1m9.273s

sys     0m6.210s


on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz + 16GiB RAM (and data on external drive)






share|improve this answer


























  • And yes, I know that this find command will not properly encode filenames containing quotes according to csv standard

    – Grzegorz Wierzowiecki
    May 25 '17 at 14:12














0












0








0







I came across similar problem (detail in "Context" section) and arrived with following ftfy_line_by_line.py solution:



#!/usr/bin/env python3

import ftfy, sys

with open(sys.argv[1], mode='rt', encoding='utf8', errors='replace') as f:

  for line in f:

    sys.stdout.buffer.write(ftfy.fix_text(line).encode('utf8', 'replace'))

    #print(ftfy.fix_text(line).rstrip().decode(encoding="utf-8", errors="replace"))


Using encode+replace + ftfy to auto-fix Mojibake and other corrections.



Context



I've collected >10GiB CSV of basic filesystem metadata using following gen_basic_files_metadata.csv.sh script, running essentially:



find "${path}" -type f -exec stat --format="%i,%Y,%s,${hostname},%m,%n" "{}" ;


The trouble I had was with inconsistent encoding of filenames across file systems, causing UnicodeDecodeError when processing further with python applications (csvsql to be more specific).



Therefore I applied above ftfy script, and it took



Please note ftfy is pretty slow, processing those >10GiB took:



real    147m35.182s

user    146m14.329s

sys     2m8.713s


while sha256sum for comparison:



real    6m28.897s

user    1m9.273s

sys     0m6.210s


on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz + 16GiB RAM (and data on external drive)






share|improve this answer















I came across similar problem (detail in "Context" section) and arrived with following ftfy_line_by_line.py solution:



#!/usr/bin/env python3

import ftfy, sys

with open(sys.argv[1], mode='rt', encoding='utf8', errors='replace') as f:

  for line in f:

    sys.stdout.buffer.write(ftfy.fix_text(line).encode('utf8', 'replace'))

    #print(ftfy.fix_text(line).rstrip().decode(encoding="utf-8", errors="replace"))


Using encode+replace + ftfy to auto-fix Mojibake and other corrections.



Context



I've collected >10GiB CSV of basic filesystem metadata using following gen_basic_files_metadata.csv.sh script, running essentially:



find "${path}" -type f -exec stat --format="%i,%Y,%s,${hostname},%m,%n" "{}" ;


The trouble I had was with inconsistent encoding of filenames across file systems, causing UnicodeDecodeError when processing further with python applications (csvsql to be more specific).



Therefore I applied above ftfy script, and it took



Please note ftfy is pretty slow, processing those >10GiB took:



real    147m35.182s

user    146m14.329s

sys     2m8.713s


while sha256sum for comparison:



real    6m28.897s

user    1m9.273s

sys     0m6.210s


on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz + 16GiB RAM (and data on external drive)







share|improve this answer














share|improve this answer



share|improve this answer








edited May 25 '17 at 18:33

























answered May 25 '17 at 14:11









Grzegorz WierzowieckiGrzegorz Wierzowiecki

5,2421363105




5,2421363105













  • And yes, I know that this find command will not properly encode filenames containing quotes according to csv standard

    – Grzegorz Wierzowiecki
    May 25 '17 at 14:12



















  • And yes, I know that this find command will not properly encode filenames containing quotes according to csv standard

    – Grzegorz Wierzowiecki
    May 25 '17 at 14:12

















And yes, I know that this find command will not properly encode filenames containing quotes according to csv standard

– Grzegorz Wierzowiecki
May 25 '17 at 14:12





And yes, I know that this find command will not properly encode filenames containing quotes according to csv standard

– Grzegorz Wierzowiecki
May 25 '17 at 14:12


















draft saved

draft discarded




















































Thanks for contributing an answer to Unix & Linux Stack Exchange!


  • Please be sure to answer the question. Provide details and share your research!

But avoid …



  • Asking for help, clarification, or responding to other answers.

  • Making statements based on opinion; back them up with references or personal experience.


To learn more, see our tips on writing great answers.




draft saved


draft discarded














StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2funix.stackexchange.com%2fquestions%2f6516%2ffiltering-invalid-utf8%23new-answer', 'question_page');
}
);

Post as a guest















Required, but never shown





















































Required, but never shown














Required, but never shown












Required, but never shown







Required, but never shown

































Required, but never shown














Required, but never shown












Required, but never shown







Required, but never shown







-

Popular posts from this blog

How to make a Squid Proxy server?

Image
1 I want to set up squid proxy server for monitoring user activity in a network and Internet Web URL filtering for user. I also want to configure the user login, user log, user access group, and block unwanted web sites in my network. Please help me and guide me the full process to do this. server proxy squid share | improve this question edited Jul 31 '12 at 14:45 Jorge Castro 37.1k 107 422 617 asked Jul 31 '12 at 5:16 Goivind Tiwari Goivind Tiwari ...
Read more

Is this a new Fibonacci Identity?

Image
6 1 $begingroup$ I have found the following Fibonacci Identity (and proved it). If $F_n$ denotes the nth Fibonacci Number, we have the following identity begin{equation} F_{n-r+h}F_{n+k+g+1} - F_{n-r+g}F_{n+k+h+1} = (-1)^{n+r+h+1} F_{g-h}F_{k+r+1} end{equation} where $F_1 = F_2 = 1$ , $r leq n$ , $h leq g$ , and $n, g, k in mathbb{N}$ . It is not too hard to show that this identity subsumes Cassini's Identity, Catalan's Identity, Vajda's Idenity, and d'Ocagne's identity to name a few. I have done a pretty thorough literature review, and I have not found anything like this, but I am still wondering if anyone has seen this identity before? I found this by accident after noticing some patterns in some analysis work I was doing, so if this is already known I would be curious to see w...
Read more

19世紀

Image
この記事は検証可能な参考文献や出典が全く示されていないか、不十分です。 出典を追加して記事の信頼性向上にご協力ください。 （ 2012年3月 ） 千年紀: 2千年紀 世紀: 18世紀 - 19世紀 - 20世紀 十年紀: 1800年代 1810年代 1820年代 1830年代 1840年代 1850年代 1860年代 1870年代 1880年代 1890年代 19世紀に君臨した大英帝国。 19世紀 （じゅうきゅうせいき）は、西暦1801年から西暦1900年までの100年間を指す世紀。 目次 1 19世紀の歴史 1.1 国民国家の成立 1.2 帝国主義の興隆 1.3 列強の植民地争奪戦 2 できごと 2.1 1800年代 2.2 1810年代 2.3 1820年代 2.4 1830年代 2.5 1840年代 2.6 1850年代 2.7 1860年代 2.8 1870年代 2.9 1880年代 2.10 1890年代 2.11 1900年代 3 文化 3.1 文学 3.2 音楽 3.3 思想 3.4 科学 3.5 技術 4 人物 4.1 ヨーロッパ 4.1.1 政治家・王族 4.1.1.1 フランス 4.1.1.2 オーストリア＝ハンガリー 4.1.1.3 ロシア 4.1.1.4 イギリス 4.1.1.5 ドイツ（プロイセンほかドイツ領邦を含む） 4.1.1.6 北欧 4.1.1.7 イタリア 4.1.1.8 スペイン 4.1.1.9 ベルギー 4.1.1.10 ギリシア 4.1.2 軍人 4.1.3 実業家 4.1.4 科学と技術 4.1.5 思想と哲学・人文諸学 4.1.6 宗教 4.1.7 文学 4.1.8 美術 4.1.9 音楽 4.1.10 社会事業家 4.1.11 探検家・旅行家 4.1.12 料...
Read more