[Bug c++/24985] caret diagnostics

[manu at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24985

--- Comment #24 from Manuel López-Ibáñez <manu at gcc dot gnu.org> 2012-04-05 22:05:15 UTC ---
(In reply to comment #23)
> (In reply to comment #22)
> > * DejaGNU trims leading whitespace before passing the text to prune.
> 
> So it does.  Bizarre.
> 
>             set comp_output [string trimleft $comp_output]
> 
> I guess we can just remove the initial space from that prune pattern.
> 
> > The problem is that the default regexp of DejaGNU is:
> > 
> > "(^|\n)(\[^\n\]+$line\[^\n\]*($pattern)\[^\n\]*\n?)+"
> > 
> > the "+" at the end means that it matches as many lines as possible.
> 
> Yes, but only lines that match the line number followed by the pattern; the
> source line shouldn't have its own line number before the pattern.

There are a few dg- directives with line number 0. They match everywhere.

Another problem with c99-vla-jump-3.c and similar testcases is that it seems as
if DejaGNU limits the output (or has a limited size buffer for text) and
decides to stop parsing in the middle of a source line, which breaks the
pruning (and misses several diagnostics that need to be matched). Could this be
true or am I doing something wrong?

I am starting to reconsider adding -fno-diagnostics-show-caret everywhere.