Re: [AUCTeX-devel] reftex patch to support resolved labels

[Tim Toolan]

Quoting "Ralf Angeli" <address@hidden>:

> * Tim Toolan (2008-10-21) writes:
>
> > Attached is a patch which adds support for resolved labels as they
> > appear the typeset document to the reftex-reference command.  It
> > places them where the counter used to be, and enables interactive
> > filtering based on these resolved labels.  It makes selecting labels
> > significantly more efficient when looking at typeset document.
>
> I finally had some time to look into your patch.
>
> This looks like an interesting feature.  Something like that might also
> be useful when displaying labels in the table of contents (`C-c =').

It could be useful there, but one thing to consider is that labels are  
not "required" for anything, and the only things that can be resolved  
are things that have a label.  For instance, if there is a section 3,  
but it is not referenced in the document, and the writer decided not  
to label it, there will be no resolved label for it.

> > To reference a label, run reftex-reference "C-c )", then enter the
> > number as it appears in the typeset document (ie. "1.2", "IV", "123").  If
> > there is more than one exact match, such as section 3, equation 3, and
> > figure 3, press tab to cycle between these choices.
>
> While the display of typeset label numbers might be a good hint when
> trying to identify the right label, I think the primary way to select
> them should still be based on labels as they are written in the LaTeX
> sources.  First, a user should not be required to run LaTeX or look at
> the typeset document to identify a label,

True.  Maybe when the .aux files are not present, instead of  
presenting question marks for all labels, it should not display  
anything at all.

> second, the label name in the
> sources often carries more information than the typeset one and

That is true, but I am thinking of how I write, and when I reference  
an equation, I am almost always looking at in the typeset document,  
and my mind says something like
  "combining (17) and (18), we get"
instead of
  "combining (\ref{eq:blah1}) and (\ref{eq:blah2}), we get".

> third,
> the selection from a list (<n>, <p>, <RET> or point and click) is mostly
> more convenient than having to type a string with completion.  (The
> latter argument is obviously subjective and represents my personal
> preference.)  In the label selection process at it is now one can
> activate a prompt for typing a label name with completion by pressing
> <TAB>.  One could either allow completion on typeset labels here as well
> or provide a key to activate a dedicated prompt for asking for the
> typeset label name.

That is actually exactly how the patch behaves if  
`reftex-ref-start-in-filter-mode' is set to nil.

> > When in this filtering mode, any special commands that are printable
> > characters (like "f" to toggle follow mode) must be preceded by a
> > "C-c" prefix.  At any time, press the delete key to toggle between
> > filter mode and the previous behavior for reftex-reference.
>
> As hinted at above, such a change of behavior might be too drastic.  A
> lot of people already have the selection process in their muscle memory.

This is true.

> > It is currently setup to disable prompting for reference macro type
> > and label type by default, and going directly to the filter selection
> > process.  To revert to the old behavior, set the variables:
> >    (setq reftex-ref-type-prompt t)
> >    (setq reftex-ref-start-in-filter-mode nil)
> >    (setq reftex-ref-macro-prompt t)
>
> Why would you want to change these defaults?

The main reason I changed them was to illustrate the changes  
introduced by this patch.  As you said, changing a mature user  
interface is not a great idea because people do have emacs muscle  
memory, so they should probably be left alone.

> The patch also contained some changes where I am not sure if they are
> necessary or good, like commented `message' calls or `defvar' statements
> to silence the byte compiler.  But before talking about those (and me
> looking deeper into the code) we should agree on the user interface.

I agree about the defvar statements.  I saw in reftex-sel.el that the  
same thing was done for the same reason, so I duplicated it, but it  
was probably a bad idea.  The reason for the commented message calls  
could be discussed later as you say, after a revised version of the  
patch is worked out.

One additional thing that must be addressed before this patch can be  
applied is the compatibility with preview-latex.  The reason for this  
is that I made the assumption that the .aux files will be consistent  
with the output that the user is looking at.  Normally, it is very  
unusual that the .aux files will not be consistent, but in the case of  
preview-latex they will never be consistent because preview-latex  
doesn't seem to create .aux files.  The .aux file information must be  
available in some way for this to work with it.

-Tim