------- Comment #9 from manu at gcc dot gnu dot org 2007-01-31 02:28 ------- (In reply to comment #8) > This is nice, Manuel, I hadn't considered changing the expressions as they are > added to the messages list.
We already wrap dg-test (see the end of lib/gcc-dg.exp). So why not wrapping dg-warning and dg-error? Unfortunately my approach is quite brute, it would be better to call the original dg-warning/dg-error, catch its output and adjust it (or replace it). This way we shouldn't need to copy the whole function. Something similar is done for dg-test. I don't know how to do this because I have no idea of Tcl programming. > Another possibility is to add two new test directives, for example > dg-gcc-warning and dg-gcc-error, and leave dg-warning and dg-error as they > are. > This has the advantage of not confusing people who are accustomed to the > existing behavior of the DejaGnu versions of these procs. That way you could > start using the new ones immediately, and we can gradually move other tests to > use them as well. We could do that. Will people use them? Will they be obligatory for new testcases? Anyway, I am afraid that the above patch is not correct. It produces spurious FAILs for some testcases that are fine. I am testing a new version, let's see how many testcases fail then. Also, we will need a new way to handle the output of inform and other messages that gcc produces, such as "cc1: warnings being treated as errors". (I think that particular message is inconsistent with the rest of our diagnostics, but that is another issue). -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25241
