Closed tiennou closed 3 years ago
Updated to have builds against multiple LLVM versions. I've moved the macOS build out of the matrix so we don't end up with 40+ jobs as well.
Looks like the PR picked up the workflow already 👍.
It looks like Ruby 2.6 sometimes fails to grab all the groups, but somewhat inconsistently. I wonder what's up with that and whether it's something we need to/can fix.
Yeah, I suspect it's spotty because of stuff like /var/folders/24/8k48jl6d249_n_qfxwsl6xvm0000gn/T/HEAD-20210603-2774-1tb1dgk/types.h:36:1: error: unknown type name 'GIT_BEGIN_DECL'
, causing whatever "group" header was currently being processed to be entirely missing. I still haven't pinpointed if it's a clang bug that causes it to read an include as a 0-length file, or a clobbering issue on our side, but I know I've seen utterly nonsensical stuff, like unknown type name 'ypere'
.
I have a patch that implements a retry loop around that to alleviate this issue, and I might upstream it separately.
This moves all our CI infrastructure over to Github Actions and removes Travis from the equation, with a matrix of Ubuntu/macOS vs. Ruby 2.4/2.5/2.6/head.
Note that I cherrypicked one commit from #56 to handle Ruby 3/
head
, and that I've run the actions while pushing to my fork'smaster
branch — see here for the runs.Loosely inspired from #47 (had an issue with finding libclang on Linux).