On 18 September 2015 at 14:30, Gavin Smith <[email protected]> wrote: > After mulling it over, I've come up with the following: > > Index: texi2dvi > =================================================================== > --- texi2dvi (revision 6632) > +++ texi2dvi (working copy) > @@ -1366,7 +1366,7 @@ > run_core_conversion () > { > # Backup a sorted index file > - test -f "$in_noext.fls" && cp "$in_noext.fls" "$in_noext.f0s" > + test -f "$in_noext.fls" && mv "$in_noext.fls" "$in_noext.f0s" > > case $in_lang:$latex2html:`out_lang_tex` in > *:dvi|*:pdf|latex:tex4ht:html) > > ==================== > > That is, force the *.fls file to be read from the backup directory > because no such file will exist in the current directory. > > Of course, that relies on the recorder file not being created > immediately and identified for opening for reading. That's possible, > with what you found with MikTeX seemingly being able to read from a > file of one filename while writing out a recorder file with exactly > the same name at the same time. Alas, even if the format of the > recorder file is consistent across TeX derivatives, the exact timing > of files being opened and closed may not be.
So this doesn't work. I reordered TEXINPUTS to put the working directory first, and the index displays as non-existent. I put in a breakpoint in texinfo.tex in \printindex (with \show), and had a look at the files in the directory. test.fls had already been created, but was empty.
