2014-11-28 19:10 GMT+01:00 Joseph Myers <jos...@codesourcery.com>: > On Fri, 28 Nov 2014, Kai Tietz wrote: > >> Some diagnostics are different and some constructs getting allowed >> with enabled ms-extensions flag. Additionally is the pedantic-flag >> not automatically set for *-*-mingw* targets. So for enforcing >> ISO-C++ pedantic checks the *-*-mingw* targets need to have explicit >> set the -pedantic flag (which is on for some other targets by >> default). > > It seems very odd for the default dg-options to vary depending on target - > where is that target-dependent default set?
True, I was curious to see this difference too. > But for my main point: -std=c11 -pedantic-errors must diagnose with an > error anything where C11 requires a diagnostic. Not "anything where C11 > requires a diagnostic, except for certain extensions on MinGW targets". > It must not be necessary to add -fno-ms-extensions for such diagnostics; > the options for conformance must not depend on the target. So for the > cases where the test uses -std=<some ISO standard> > -pedantic/-pedantic-errors, you need to fix the port: the test failures > are showing up an actual bug in the port, and the tests are correct as-is. > The same principle of course applies to C++ tests, though I didn't examine > your changes to those to see if the same issue applies to any of those > changes. How so? That ms-extensions are enabled by default is all the port does. The rest is done in FEs, and indeed the diagnostic here about handling ms-extensions, or on standard-case seems to be buggy. To depend in testsuite on implicit set flags seems to be a bug too. So I can't follow you that it is a port issue. .. Kai Tietz kti...@redhat.com > -- > Joseph S. Myers > jos...@codesourcery.com
