Jeremy Morse via llvm-dev
2021-Feb-10 11:29 UTC
[llvm-dev] Repeated names for architecture, aarch64 and arm64, needed in XFail list
Hi llvm-dev@, I've run into a situation where an architecture has two different names. While editing a debug-info test recently, I needed to XFail it for aarch64 [0]. However, it then turned out that when building on Apple M1s their target architecture is spelt "arm64", which wasn't caught by the XFail list, leading to some unexpected failures [1]. Adding another name to the XFail list isn't a big deal, but having two names for one thing isn't ideal. Is this is a known limitation, or is there a better approach to xfailing things for aarch64?. [0] e05c10380ce7 and 4fd29e4fd3c [1] https://reviews.llvm.org/D95617#2550717 -- Thanks, Jeremy
via llvm-dev
2021-Feb-10 13:47 UTC
[llvm-dev] Repeated names for architecture, aarch64 and arm64, needed in XFail list
> -----Original Message----- > From: llvm-dev <llvm-dev-bounces at lists.llvm.org> On Behalf Of Jeremy Morse > via llvm-dev > Sent: Wednesday, February 10, 2021 6:29 AM > To: llvm-dev <llvm-dev at lists.llvm.org> > Subject: [llvm-dev] Repeated names for architecture, aarch64 and arm64, > needed in XFail list > > Hi llvm-dev@, > > I've run into a situation where an architecture has two different > names. While editing a debug-info test recently, I needed to XFail it > for aarch64 [0]. However, it then turned out that when building on > Apple M1s their target architecture is spelt "arm64", which wasn't > caught by the XFail list, leading to some unexpected failures [1].XFAIL doesn't do anything sophisticated, it's just a substring match on the triple. That's why it didn't work. I ran into something similar at one point, though. There's a test somewhere (which I'm too lazy to look up right now) that has both 'arm' and 'arm64' triples in it; and this doesn't work if you have enabled ARM but not AARCH64. Moving the arm64 RUN line to an aarch64-equivalent test... didn't work. Somewhere in the commit history you can see my attempt to deal with it and the fairly quick revert. The problem wasn't annoying enough to be worth digging deeper, but it would be nice to know what's going on. --paulr