On Wed, 31 Jul 2024 at 21:59, Jonathan Wakely <jwak...@redhat.com> wrote: > > On Wed, 31 Jul 2024 at 21:44, François Dumont <frs.dum...@gmail.com> wrote: > > > > Committed as trivial. > > > > Fix a compilation error that is not expected by the tests preserving > > the expected ones. > > > > The 'test' variable declaration is missing since commit > > a9260b7eb688df43a724e25421ba40f35a89fee9 that removed the test global > > variable in testsuite files. > > Oh good catch! > > The problem is that the dg-error "no" is matching two errors on that > line, the "'test' was not declared" one and the "no match" one. > > I think we should get rid of the 'test' variable again, and use a > better dg-error pattern. "no" is much too short and matches too much.
Those are the only tests we have with a two-letter dg-error pattern, there are none with one letter or three letters. There are lots with four-letter patterns, but they're all { dg-error "here" } which is fine. There are lots of decimal FP tests with { dg-error "error" } which is silly, it might as well be "." and match anything. They should be fixed, although everything for DFP is low priority. > > There's no need to assign the expressions to a variable, they're > ill-formed anyway. > > So: > > itr != mapByName.end(); // { dg-error "no match" } > itr == mapByName.end(); // { dg-error "no match" } > > > > > > libstdc++-v3/ChangeLog: > > > > * testsuite/23_containers/map/operators/1_neg.cc (test01): Add test > > variable > > declaration. > > * testsuite/23_containers/set/operators/1_neg.cc (test01): Likewise. > > > > François