[AUCTeX] Re: reftex-label-alist not consistent

[Ralf Angeli]

* Emilio Jesús Gallego Arias (2008-04-13) writes:

> Ralf Angeli <address@hidden> writes:
>
>> The \begin{lemma} like all the other section stuff from
>> `reftex-section-levels' and `reftex-label-alist' is only found if it is
>> followed by "[[{ \t\r\n]" (regexp-wise).  I have to check if this
>> restriction can be loosened.
>
> I wonder what's the rationale behind this.

This part of the regexp is supposed to prevent stuff like \sectionxxx to
be matched if e.g. "section" is in `reftex-section-levels{-all}'.

> I changed that to be 
> "[[{ \t\r\n]?" and indeed the lemmas show now correctly in the TOC.
>
> I have checked with some documents I've got here and I couldn't see any
> adverse effect, but sure we should first know the rationale behind the
> old regex before changing that.

There will likely be false positives as described above.

I've know implemented a change where environments and macros are treated
differently when the regexp is built.  The check for "[[{ \t\r\n]" is
now only done for macros.  Not perfect, but this should be better now.

>> The first \begin{lemma} is found and since it is treated as a section,
>> the bound for looking backwards during parsing is set to the point right
>> after it (in `reftex-parse-from-file').  Now when the label after it is
[...]
> I modified the code to set the bound to the start of the match, and then
> everything work as expected in my files, but unfortunately I don't think
> they are complex enough to say this works in every case.

I did the same thing.  The change seems harmless but one never knows.

>> I'll need to check more closely how the bugs can be fixed.
>
> I hope my experiment is useful. At least I now understand what the code
> is doing thanks to your explanations.

Well, I was mostly through with implementing the changes when your
message arrived, but they gave me some confirmation with respect to what
I've done.

You can get the fixes through CVS.

> BTW, it seems we don't have a test case for reftex. I think it would be
> cool to have a pair (file.tex,file.rel) so any change to parsing can be
> tested.

(Automated) test harnesses and simular stuff are always a good idea,
especially for regression testing.  But it can also be quite time
intensive creating them.  This is likely the reason we don't have them
yet.

-- 
Ralf