Re: Fwd: REVIEW REQUEST - user manual generation

[Vincent Belaïche]

Answering to myself...

>Hello,
>
>Le 03/06/2016 à 18:15, Gavin Smith a écrit :
>>> The problem is that my MSYS sed was version 3.02, and it did not
>>> recognise \n in the replacement string when you want to make each funny
>>> character on one line. I made some test with my git bash --- which has
>>> sed version 4.2.1 --- and recognising \n is OK
>>>
>>
>> Apparently it should be \NEWLINE, where NEWLINE is a newline
>> character. Quoting is tricky, because \NEWLINE is interpreted as a
>> line continuation except in single quotes or if the backslash is
>> escaped. We are in single quotes, but those single quotes are within
>> backquotes. 

Yes quoting can be tricky...

The following works for me w/o going through a list_funnies function:

--8<----8<----8<----8<----8<-- begin -->8---->8---->8---->8---->8----
  in_input_funnies=`echo "$in_input" \
    | $SED -e 's![^}#$%&^_{~]!!g; s!\(.\)!\1\'"$newline"'!g' \
    | uniq`
--8<----8<----8<----8<----8<--  end  -->8---->8---->8---->8---->8----

One must write "$newline", not just $newline, otherwise the expansion
happens too early and the NEWLINE is not passed as part of the sed
argument.

Please note that the variable newline is defined at the beginning of
texi2dvi.

VBR,
        Vincent.


>> I avoided this complication by using a function in the following
>> patch:
>>
>> --- texi2dvi    (revision 7200)
>> +++ texi2dvi    (working copy)
>> @@ -918,6 +918,13 @@
>>    fi
>>  }
>>
>> +list_funnies ()
>> +{
>> +  echo "$in_input" \
>> +    | $SED -e 's![^}#$%&^_{~]!!g' -e 's!\(.\)!\1\
>> +!g' | uniq
>> +}
>> +
>
>Just to say that I have tried the following both with sed 3.02 and sed
>4.2.1:
>
>--8<----8<----8<----8<----8<-- begin -->8---->8---->8---->8---->8----
>$ echo 'ab' | sed -e 's!\(.\)!\1\
>> !g'
>a
>b
>--8<----8<----8<----8<----8<--  end  -->8---->8---->8---->8---->8----
>
>and it worked. So, your patch will make it.
>
>Just as a minor comment, one could have the two instructions of the sed
>script in the same -e argument provided that they are semi-colon
>separated, like this:
>
>--8<----8<----8<----8<----8<-- begin -->8---->8---->8---->8---->8----
>list_funnies ()
>{
>  echo "$in_input" \
>    | $SED -e 's![^}#$%&^_{~]!!g; s!\(.\)!\1\
>!g' | uniq
>}
>--8<----8<----8<----8<----8<--  end  -->8---->8---->8---->8---->8----
>
>This would be marginally more readable.
>
>>  # run_tex - Run TeX as "$tex $in_input", taking care of errors and logs.
>>  run_tex ()
>>  {
>> @@ -926,9 +933,8 @@
>>    # Check for any unusual characters in the filename.
>>    # However, >, \ and any whitespace characters are not supported
>>    # filenames.
>> -  in_input_funnies=`echo "$in_input" \
>> -    | $SED -e 's![^}#$%&^_{~]!!g' -e 's!\(.\)!\1\n!g' \
>> -    | uniq`
>> +  in_input_funnies=`list_funnies`
>> +
>>    if test -n "$in_input_funnies" ; then
>>      # Make > an end group character, as it's unlikely to appear in
>>      # a filename.
>>
>>
>>> Ok, one sh only solution would be to do this :
>>>
>>> --8<----8<----8<----8<----8<-- begin -->8---->8---->8---->8---->8----
>>>   in_input_funnies=`echo "$in_input" \
>>>     | $SED -e 's![^}#$%&^_{~]!!g' \
>>>     | { while IFS= read -r -n1 -d '' c; do printf '%s\n' "$c"; done } \
>>>     | uniq`
>>> --8<----8<----8<----8<----8<--  end  -->8---->8---->8---->8---->8----
>>>
>>> Hopefully this is portable code ;-) ...
>>
>> Options to "read" aren't portable AFAIK.
>
>Thank you for clarifying the point. Anyway your solution is clearly
>better.
>
>VBR,
>       Vincent.
>