On Tue, May 13, 2014 at 1:26 PM, Nico Weber <[email protected]> wrote: > On Tue, May 13, 2014 at 1:21 PM, Eric Christopher <[email protected]> > wrote: >> >> On Tue, May 13, 2014 at 1:20 PM, Rafael Espíndola >> <[email protected]> wrote: >> > On 13 May 2014 16:09, Eric Christopher <[email protected]> wrote: >> >> Also as a note, driver tests shouldn't require registered targets if >> >> they're just checking command lines. Also, no clang test should >> >> require a registered target. I'm not sure why some do at the moment. >> > >> > This is a case where the -cc1 test requiring asm printing is somewhat >> > reasonable. I don't think we have a better way to check that a codegen >> > option passed to -cc1 is being used. >> > >> >> Right. We should only test that we've passed it. Make no claims about >> it working :) > > > Why is that? Isn't it more interesting to test that -masm=intel actually > produces intel asm, instead of testing some implementation detail for how > that's done? >
Because then you're testing the backend. If you want to test that it actually happened you just need a backend test with the associated .ll file for this. Probably should have one if we don't. -eric _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
