[libcxx-commits] [libcxxabi] [libc++abi] Handle null pointer-to-object: Issue #64593 (PR #68076)

[Iain Sandoe via libcxx-commits]

iains wrote:

> Interesting. This actually highlights a blind spot we have in the expressiveness of our Lit features. We don't have a way to express "libc++ configured as Apple system library, targeting , but using the tip-of-trunk version of the library". In other words, we generally assume that `apple-system` + `target=<XYZ>` implies that we are back-deploying (i.e. running against the system library that was shipped on `<XYZ>`). 

in this case it was just the plain `apple-system` i.e. running natively on the box.  Given that it intentionally installs libc++ and libc++abi, (and then points to them with DYLD_LIBRARY_PATH) I guess it is intentional that this exercises the built tip of trunk.

>I'm not sure why that never bit us, but I suspect this is because these fixes are somewhat rare.

I suppose libc++abi moves slowly, not sure why it does not bite more with libc++ tho - if the assumption of the lit config is that it's the installed libs being tested, but actually we're testing the new ones.

> To avoid blocking you on this (which really has nothing to do with this patch), you could just use `UNSUPPORTED:` instead of `XFAIL`, that way the test just won't run in that configuration.

OK - I guess that means that the test will not run in any CI - but should still run when people test the library locally?


https://github.com/llvm/llvm-project/pull/68076