On Fri, Dec 19, 2008 at 8:59 AM, Fariborz Jahanian <[email protected]> wrote: > On Dec 18, 2008, at 5:43 PM, Daniel Dunbar wrote: > > Sure. These patches came because of dejagnu test failures. There are still > more issues > that I am looking at. But tests will come first. Unfortunately, encoding > mismatches are > hard to find because they cause silent runtime failures. And dejagnu tests > use them > for other purposes than testing @encode.
This is why I wrote the print-class-info.m, it uses the Objective-C runtime library to iterate over certain interfaces and print all their metadata information. I tried to put examples of every kind of construct in the classes we iterate over, so that we exercise all the encoding code. By comparing the output of this program to gcc, we get a pretty good idea that we are encoding things correctly. > (from private email) > This is a clang specific issue. llvm-gcc does not have this issue > as it uses gcc's front-end for encoding. > Why should this test goes into llvm source tree? Not the llvm source tree, the llvm test suite source tree: http://llvm.org/docs/TestingGuide.html#testsuiterun The llvm test suite is our only current infrastructure for doing executable / end-to-end tests. It makes sense to share these tests because they are only relying on the "user visible" side of the compiler. In this specific case, both llvm-gcc and clang should emit metadata which matches gcc. - Daniel _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
