Alex Rønne Petersen
Alex Rønne Petersen
~~Probably would be good to merge #20944 before this is done, if possible.~~ Done.
@Beyley are you still able to repro https://github.com/ziglang/zig/issues/20313 with Zig master? I just tried switching to Zig master on my repro project and it [seems to work](https://github.com/alexrp/zig-20313-repro/actions/workflows/build.yml).
Yeah, I don't think this necessarily needs to block the PR considering how much of an edge case it is. IIUC, it seems like the worst case scenario is that...
Follow-up filed at #23366.
This is properly ready for review now.
Rebased after #20725 was merged (tiny conflict in `std.os.linux.tls`). No functional changes, and CI was [green](https://github.com/ziglang/zig/actions/runs/10038451491) prior in case we want to avoid wasting another full CI run on this.
The issue is that the resulting Zig triple *can't* become `*-*-gnueabi`; there is simply not enough information provided for `zigTriple()` to determine whether it should return `gnu`, `gnuabi64`, `gnueabi`, etc...
That works too I suppose. Although with #20690 accepted, the "API" component of the target becomes just `gnu` anyway, so in that light, perhaps it makes sense to take the...
@kubkon I updated the PR with a more detailed comment explaining why I'm making it unconditionally emit the reference regardless of target arch/ABI.
> I don't get it. The symbol is both provided in the static libc_nonshared.a as well as a dynamic reference to libc.so? The symbol is *defined* in `Scrt1.o` and *referenced*...