Take the 2-minute tour ×
Unix & Linux Stack Exchange is a question and answer site for users of Linux, FreeBSD and other Un*x-like operating systems.. It's 100% free, no registration required.

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.

share|improve this question

This question has an open bounty worth +100 reputation from Gilles ending in 2 days.

One or more of the answers is exemplary and worthy of an additional bounty.

    
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

5 Answers 5

up vote 13 down vote accepted

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-\xEC\xEE-\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

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 '/
   ^( [\000-\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-\xEC\xEE-\xEF]  [\x80-\xBF]  [\x80-\xBF]
#    ==============================================
#    ((([\xE0][\xA0-\xBF])|([\xED][\x80-\x9F])|([\xE1-\xEC\xEE-\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-\xEC\xEE-\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-\xEC\xEE-\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 '%0.8X\n' $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'$hexUTF32BE |
                perl -l -ne '/^(  [\000-\177]
                                | [\300-\337][\200-\277]
                                | [\340-\357][\200-\277]{2}
                                | [\360-\367][\200-\277]{3}
                                | [\370-\373][\200-\277]{4}
                                | [\374-\375][\200-\277]{5}
                               )*$/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 "$UTF8" |perl -l -ne '/^(([\x00-\x7F])|([\xC2-\xDF][\x80-\xBF])|((([\xE0][\xA0-\xBF])|([\xED][\x80-\x9F])|([\xE1-\xEC\xEE-\xEF][\x80-\xBF]))([\x80-\xBF]))|((([\xF0][\x90-\xBF])|([\xF1-\xF3][\x80-\xBF])|([\xF4][\x80-\x8F]))([\x80-\xBF]{2})))*$/ 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-\xEC\xEE-\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 '/^([\000-\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 '/^([\000-\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/(.)(.)(.)(.)(.)(.)(.)(.)/\1\n\2\n\3\n\4\n\5\n\6\n\7\n\8/" |
              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}${rcIco32}=,$rcRegex\t: $(printf "%0.8X\n" $invalCt)\t$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 \300\200 (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

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

If you want to use grep, you can do:

grep -av '^.*$' 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 2 days ago
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 2 days ago
    
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 2 days ago
    
@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 2 days ago

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

$ print '\\xE9 \xe9 \u20ac \ud800\udc00 \U110000' |
    uconv --callback escape-c -t us
\xE9 \xE9 \u20ac \xED\xA0\x80\xED\xB0\x80 \xF4\x90\x80\x80

(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 yesterday

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 2 days ago
    
@StéphaneChazelas Dammit. Thanks. I've only run nominal tests so far, I'll run Peter's test battery later. –  Gilles 2 days ago

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.