On Jan 16, 2014, at 12:28 PM, Argyrios Kyrtzidis <[email protected]> wrote:
> On Jan 16, 2014, at 11:04 AM, Ben Langmuir <[email protected]> wrote: > >> On Jan 16, 2014, at 10:15 AM, Argyrios Kyrtzidis <[email protected]> wrote: >>> >>> Very nice! I like the import stack in the notes. >>> Could you also add a note saying something like "please rebuild the >>> precompiled header ’foo.h.pch’” ? >> >> Done. I emitted it conditionally upon Diags.isDiagnosticsInFlight(). I >> wasn’t sure if that’s desired, but it seems to be the way the other notes >> are being emitted in this file so I did it too. > > (disregard the previous email..) > > The test is problematic. It has > > -fmodule-map-file=modified-module-dependency.module.map > > But with no absolute path it will just try to find it in the current > directory where the test suite is executing and fail. You later touch it > (therefore create it) and a second run will pass. > In general you should avoid relative paths in the test run, if you are going > to touch something, copy it to the test output directory and use it from > there. > > Also I don’t see any module created in the test’s module cache path. This is > also indicated by the error message which only contains about the precompiled > header, no module shows up in the imported stack. > > I’d suggest > - Create a PCH that imports a module which contains a header > - touch the module header > - Verify that the error complains about the PCH and that the module shows up > in the import stack. Ah, sorry about the bogus test. The attached includes a rewritten test per your suggestions. I’ve verified that it passes from a fresh state (and fails appropriately without the changes from this patch). Ben
pch-error.patch
Description: Binary data
_______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
