Re: [bug #43122] texi2dvi does not compile enough times to get toc

[Gavin Smith]

On 8 September 2015 at 22:44, Vincent Belaïche
<address@hidden> wrote:
> Salut Gavin & Karl,
>
> Answers embedded below...
>
> Le 05/09/2015 00:30, Gavin Smith a écrit :
>>> So from the above, I think that with MikTeX, yes it is working, but this
>>> is not because of your patch, but thanks to some MikTeX trick which I
>>> don't know. This is why I am saying that your patch is not working, I am
>>> meaning that it is not working *PER SE*.
>>
>> I see. Could you try it with an @include of another file along with
>> --tidy? MikTeX finds files in the build directory, probably because
>> texi2dvi has changed to that directory, so it might not be able to
>> find files in the original directory, if it's not using TEXINPUTS.
>
> I have tried that, I placed a file dummy_inc.texi in the same dir as
> dummy.texi, dummy_inc.texi contains a single line with a flag DUMMY-SONG
> definition which is used in the main file dummy.texi.
>
> Now, when I compile, MikTeX could not find dummy_inc.texi, I had to enter
> manually the path ../../../dummy_inc.texi, which confirms that the
> current directoy for MikTeX compilation is ./dummy.t2d/pdf/build

That confirms what the problem is.

>> Something I read said that MikTeX uses semicolons as a path separator
>> sometimes, but you have colons in your log. Could you try changing
>> path_sep to ; to see if that makes a difference?
>>
>> Do you know where there is official documentation of where MikTeX
>> looks for input files?
>>
>
> I have attached to this email the MikTeX manual as installed with my
> MikTeX 2.9. Please have a look at sections 3.2.{7,8,9}
>
> Also look at chapter 8: in fact MikTeX knows the TEXINPUTS environment
> variable. The problem is you need to make some directory path list
> translation from MSYS path format to MSWindows path format for MikTeX to
> be able to use them.
>
> One way to do that path translation would be as in the
> test_dirlist_translation script attached to this email.

texi2dvi already can use semicolons as separators. Here's the code it uses:

# Systems which define $COMSPEC or $ComSpec use semicolons to separate
# directories in TEXINPUTS -- except for Cygwin et al., where COMSPEC
# might be inherited, but : is used.
if test -n "$COMSPEC$ComSpec" \
   && uname | $EGREP -iv 'cygwin|mingw|djgpp' >/dev/null; then
  path_sep=";"
else
  path_sep=":"
fi

Would it be enough to extend this test to test for Msys? ( `uname -o`
== Msys in your script)? Or is something else required, maybe problems
to do with volume specifiers?