std.Target.Query: Fix parse test on ABIs like gnueabi, gnuabi64, etc.

[alexrp]

The zigTriple() implementation simply returns gnu when a glibc version is provided but a more specific ABI isn't.

zig/lib/std/Target/Query.zig

Lines 433 to 445 in bb70501

	if (self.glibc_version) |v| {
	const name = if (self.abi) |abi| @tagName(abi) else "gnu";
	try result.ensureUnusedCapacity(name.len + 2);
	result.appendAssumeCapacity('-');
	result.appendSliceAssumeCapacity(name);
	result.appendAssumeCapacity('.');
	try formatVersion(v, result.writer());
	} else if (self.abi) |abi| {
	const name = @tagName(abi);
	try result.ensureUnusedCapacity(name.len + 1);
	result.appendAssumeCapacity('-');
	result.appendSliceAssumeCapacity(name);
	}

I do realize I sound like a broken record at this point, but this is another problem we have because we conflate libc and ABI.

Note: I think this test has been broken for a long time, but it was not failing in CI because you need to be testing for a target like arm-linux-gnueabi, mips64-linux-gnuabi64, etc, and AFAIK the CI machines do not have non-native glibcs installed (which is a bit of a shame IMO, but I digress).

[andrewrk]

You're trying to fix this on arm-linux-gnueabi but then the expected value is wrong because the zip triple should be "arm-linux-gnueabi".

[alexrp]

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 for the ABI. So right now, it just returns gnu. And if we're targeting e.g. arm-linux-gnueabi (or gnueabihf, gnuabi64, ...), that means we construct the expected target triple native-native-gnueabi.2.1.1 which will not match native-native-gnu.2.1.1 as returned by zigTriple().

So as I see it, either a) the test needs to be fixed like I've done here, or b) the test is actually unworkable and should be deleted.

[andrewrk]

There's a third option, you can skip the test on a non-applicable host.

[alexrp]

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 PR as-is? 🤷 I don't actually have a strong opinion either way; I'd just like this test to stop failing my local multi-glibc builds.

[andrewrk]

I see your point, OK 👍