Re: FYI: more locations (PR/360) + question

[Akim Demaille]

| Index: automake.in
| ===================================================================
| RCS file: /cvs/automake/automake/automake.in,v
| retrieving revision 1.1371
| diff -u -r1.1371 automake.in
| --- automake.in       30 Sep 2002 12:32:20 -0000      1.1371
| +++ automake.in       30 Sep 2002 13:21:06 -0000
| @@ -291,7 +291,7 @@
|  
|  # We can't always associate a location to a variable or a rule,
|  # when its defined by Automake.  We use INTERNAL in this case.
| -use constant INTERNAL => new Automake::Location 'internal';
| +use constant INTERNAL => new Automake::Location;
|  

:) :)  I love your stuff :)

|  >> Now I'm open to suggestions about how these diagnostics could be
|  >> output in a way which is more readable.  Would it be better to
|  >> add a new line between each diagnostic, as follow?
| 
|  Akim> Well, let me speak as the devil :)  Make the real error a read message
|  Akim> when the TTY supports it :)
| 
| What's a read message?  "hit any key"?

Sorry, typo: s/read/red/.

|  Akim> Maybe we should extend to GCS to support shifting the secondary
|  Akim> locations to the right?  Something like:
| 
|  Akim> /home/adl/projs/cvs/automake/HEAD2/tests/../lib/am/library.am: change 
your target to read `libfoo.a$(EXEEXT)'
|  Akim>    Makefile.am:3:   while processing library `libfoo.a'
| 
| This could be nice.  (So far I've been indenting only the text of 
| secondary locations, but that obviously doesn't make it easier to read.)

That would be quite easy to have accepted from RMS provided you send a
patch for compile.el.  Bundle this with a patch for the GCS.

| There are other questions related to diagnostic formating that could
| be answered by the GCS:
| 
|   - how should multiline diagnostics be displayed?
|     Presently Automake repeats the location, as in
|       foo:3: this message spans
|       foo:3: over two lines
|     (that was what Automake did before I introduced Automake::Channels so
|     I simply copied that)

I'm afraid we ought to obey the tradition set with recent GCC (which
is what you display).  But, BTW, I must say that I hate this tradition
for several reasons:

1. when I run the compilation under Emacs, I prefer infinite width
rather than arbitrary 80 col.  But I agree this can be qualified as `a
matter of taste' (I don't agree though), the following is more
problematic...

2. because there can be message spread on several lines, Emacs makes
no difference with a line with *two* (or more) error messages, and
*skips* them when C-`.  For instance, compile this:

int foo = 1++ + 1++;
int bar = 1++;

You'll see that next error goes to line 2, although:

foo.c:1: invalid lvalue in increment
foo.c:1: invalid lvalue in increment
foo.c:1: initializer element is not constant
foo.c:2: invalid lvalue in increment

I think the GCS should chose something that helps distinguishing the
two cases.  Note that fine grained locations (with col and range) does
not eliminate the case of a single being responsible for *several*
errors.

|   - how should errors and warnings be indicated?  Presently the
|     Automake user has no way to differentiate warnings from errors --
|     well, except $?.

Ah!  I missed that point!  Undoubtedly there should be `warning: '.