bug#9319: Compilation fontification regression

bug#9319: Compilation fontification regression

[Juri Linkov]

1. Comparing etc/compilation.txt in side-by-side frames
   with versions 23.1 and 24.0 (current Bzr trunk)
   reveals the regression in fontification
   of some compilation error/warning messages.
   I see differences in the following sections:

* IBM jikes, Sparc Pascal, GCOV, Podchecker
  - errors/warnings not highlighted,
    so this looks like a bug.

* ant
  - warning highlighted with `error' face.  But maybe this change
    is based on the user's request, I don't know.

Other changes between 23.1 and 24.0 look like bug fixes
that provide more correct fontification.

2. Sometimes errors/warnings are not fontified at all.
   For instance,

   C-x C-f etc/compilation.txt RET
   C-s RET symbol: sun RET

i.e. non-incremental search that goes directly to the given section,
fontifies nothing.

But scrolling slowly with C-v from the beginning of the
etc/compilation.txt buffer, fontifies everything.

bug#9319: Compilation fontification regression

[Chong Yidong]

Juri Linkov <juri@jurta.org> writes:

> * IBM jikes, Sparc Pascal, GCOV, Podchecker
>   - errors/warnings not highlighted,
>     so this looks like a bug.
>
> * ant
>   - warning highlighted with `error' face.  But maybe this change
>     is based on the user's request, I don't know.
>
> Other changes between 23.1 and 24.0 look like bug fixes
> that provide more correct fontification.
>
> 2. Sometimes errors/warnings are not fontified at all.
>    For instance,
>
>    C-x C-f etc/compilation.txt RET
>    C-s RET symbol: sun RET
>
> i.e. non-incremental search that goes directly to the given section,
> fontifies nothing.

Both problems were due to incorrect code in compilation-error-properties
causing it to signal an error for error definitions with nil filenames,
introduced when Stefan revamped compile.el.  I've committed a fix.

bug#9319: Compilation fontification regression

[Juri Linkov]

> Both problems were due to incorrect code in compilation-error-properties
> causing it to signal an error for error definitions with nil filenames,
> introduced when Stefan revamped compile.el.  I've committed a fix.

Thanks, now this is fixed.

I noticed another regression: in etc/compilation.txt the line

  tool1 -output=foo foo.x

is highlighted incorrectly.

This is caused by revno:102131 from:
http://lists.gnu.org/archive/html/emacs-devel/2010-10/msg01018.html

I fixed this by reverting to the old regexp and removing just one
unnecessary question mark.  Now all these samples are highlighted
correctly:

tool -o foo.o foo.c
tool -o=foo.o foo.c
tool -output foo.o foo.c
tool -output=foo.o foo.c
tool -outfile foo.o foo.c
tool -outfile=foo.o foo.c
tool --output foo.o foo.c
tool --output=foo.o foo.c
tool --outfile foo.o foo.c
tool --outfile=foo.o foo.c

whereas the following are not highlighted:

tool --omega foo.c
tool --output-html-file foo.c