Eu tenho um arquivo de texto em uma codificação desconhecida ou mista. Quero ver as linhas que contêm uma sequência de bytes que não é UTF-8 válida (canalizando o arquivo de texto em algum programa). Equivalentemente, quero filtrar as linhas que são válidas UTF-8. Em outras palavras, estou procurando .grep [notutf8]

Uma solução ideal seria portátil, curta e generalizável para outras codificações, mas se você achar que a melhor maneira é assar na definição de UTF-8 , vá em frente.

Se você deseja usar grep, você pode:

grep -axv '.*' file

nos locais UTF-8 para obter as linhas que tenham pelo menos uma sequência UTF-8 inválida (isso funciona pelo menos com o GNU Grep).

Eu acho que você provavelmente quer iconv . É para converter entre conjuntos de códigos e suporta um número absurdo de formatos. Por exemplo, para remover qualquer coisa inválida no UTF-8, você pode usar:

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

Sem a opção -c, ele reportará problemas na conversão para stderr; portanto, com a direção do processo, você pode salvar uma lista deles. Outra maneira seria retirar o material não-UTF8 e depois

diff input.txt output.txt

para obter uma lista de onde as alterações foram feitas.

Edit: Eu corrigi um erro de digitação no regex. Ele precisava de um '\ x80` e não \ 80 .

O regex para filtrar formulários UTF-8 inválidos, para aderência estrita a UTF-8, é o seguinte

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'

Saída (das linhas principais. Do teste 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. Como alguém cria dados de teste para testar uma regex que filtra Unicode inválido?
A. Crie seu próprio algoritmo de teste UTF-8 e quebre suas regras ...
Catch-22 .. Mas então, como você testa seu algoritmo de teste?

O regex, acima, foi testado (usando iconvcomo referência) para cada valor inteiro de 0x00000até 0x10FFFF.. Esse valor superior é o valor inteiro máximo de um ponto de código Unicode

• Em novembro de 2003, o UTF-8 foi restringido pelo RFC 3629 a quatro bytes, cobrindo apenas o intervalo U + 0000 a U + 10FFFF, a fim de corresponder às restrições da codificação de caracteres UTF-16. `

De acordo com esta página UTF-8 da wikipedia .

• O UTF-8 codifica cada um dos 1.112.064 pontos de código no conjunto de caracteres Unicode, usando um a quatro bytes de 8 bits

Esse número (1.112.064) equivale a um intervalo 0x000000para 0x10F7FF, que é 0x0800 tímido do valor inteiro máximo real para o ponto de código Unicode mais alto:0x10FFFF

Esse bloco de números inteiros está ausente no espectro Unicode Codepoints, devido à necessidade da codificação UTF-16 ir além de sua intenção de design original por meio de um sistema chamado pares substitutos . Um bloco de 0x0800números inteiros foi reservado para ser usado pelo UTF-16. Esse bloco abrange o intervalo 0x00D800para 0x00DFFF. Nenhum desses intérpretes são valores Unicode legais e, portanto, são valores UTF-8 inválidos.

No Teste 1 , regexele foi testado contra todos os números no intervalo de pontos de código Unicode e corresponde exatamente aos resultados de iconv .. ie. Valores válidos de 0x010F7FF e valores inválidos de 0x000800 .

No entanto, agora surge o problema de: * Como o regex lida com o valor UTF-8 fora da faixa; acima 0x010FFFF(UTF-8 pode se estender para 6 bytes, com um valor inteiro máximo de 0x7FFFFFFF ?
Para gerar os necessários * valores de bytes UTF-8 não unicode necessários , usei o seguinte comando:

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

Para testar sua validade (de alguma forma), usei o Gilles'regex UTF-8 ...

  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'

A saída do 'do Perl chr print' corresponde a filtragem de regex Gilles' .. Um reforça a validade das outras .. Eu não posso usar iconvporque ele só lida com o Unicode válido-subconjunto do mais amplo (original) padrão UTF-8 padrão...

Os nunbers envolvidos são bastante grandes, então eu testei as varreduras de alto e baixo alcance e diversas varreduras, aumentando em incrementos como 11111, 13579, 33333, 53441 ... Os resultados são todos iguais, agora tudo o que resta é testar a regex em relação a esses valores fora do intervalo do estilo UTF-8 (inválido para Unicode e, portanto, também inválido para o próprio UTF-8 estrito).

Aqui estão os módulos de teste:

[[ "$(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

Eu acho uconv(no icu-devtoolspacote Debian) útil para inspecionar dados UTF-8:

$ 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

(Os \xajudam a identificar os caracteres inválidos (exceto o falso positivo introduzido voluntariamente com um literal \xE9acima)).

(muitos outros usos agradáveis).

Python teve um built-in unicodefunção desde a versão 2.0.

#!/usr/bin/env python2
import sys
for line in sys.stdin:
    try:
        unicode(line, 'utf-8')
    except UnicodeDecodeError:
        sys.stdout.write(line)

No Python 3, unicodefoi dobrado str. Ele precisa receber um objeto semelhante a bytes , aqui os bufferobjetos subjacentes para os descritores padrão .

#!/usr/bin/env python3
import sys
for line in sys.stdin.buffer:
    try:
        str(line, 'utf-8')
    except UnicodeDecodeError:
        sys.stdout.buffer.write(line)

Me deparei com um problema semelhante (detalhe na seção "Contexto") e cheguei com a seguinte solução ftfy_line_by_line.py :

#!/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"))

Usando codificar + substituir + ftfy para corrigir automaticamente o Mojibake e outras correções.

Contexto

Eu coletei> 10GiB CSV de metadados básicos do sistema de arquivos usando o seguinte script gen_basic_files_metadata.csv.sh , executando essencialmente:

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

O problema que tive foi com a codificação inconsistente de nomes de arquivos nos sistemas de arquivos, causando o UnicodeDecodeErrorprocessamento adicional com aplicativos python ( csvsql para ser mais específico).

Portanto, eu apliquei acima do script ftfy, e levou

Observe que o ftfy é bem lento, o processamento daqueles> 10GiB levou:

real    147m35.182s
user    146m14.329s
sys     2m8.713s

enquanto sha256sum para comparação:

real    6m28.897s
user    1m9.273s
sys     0m6.210s

na CPU Intel (R) Core (i) i7-3520M a 2,90 GHz + 16 GiB de RAM (e dados na unidade externa)