Iain Buclaw

> actually could you do me a favour and upstream the relaxations I'm looking at the first failure in dmd ``` assert(logGamma(-50.2L) == log(fabs(gamma(-50.2L)))); ``` I don't see any deviation...

> I'll have a try on aarch64, as there's sure to be failures there that don't exist on i386/x86_64. gammafunction hasn't been ported to 128-bit, so that's a dead end....

> @thewilsonator can you rebase this? Tests will fail if you remove INLINE_Y2LX, so rebasing would be a futile effort. https://github.com/dlang/phobos/pull/7942#issuecomment-817138577

> But they are separate sections, aren't they ? Hmm.

@Geod24 doing the inverse then, as having the top-level changelog would just be confusing.

> I think it'd be better to have them at the top. I would rather have a way to distinguish them, e.g. have `pre-commit` enforce that all changelog entry start...

> Scoring major points with this one :P Where's my medal? :-)

> Note also that issue [22568](https://issues.dlang.org/show_bug.cgi?id=22568) affects `-os` equally. Actually, `-os=` works. ``` $ dmd hello.d -c -os=windows $ file hello.obj hello.obj: Intel amd64 COFF object file, not stripped, 8...

> This should be fixed properly not removed. Any effort trying to improve dmd-specific driver handling is better spent elsewhere.

> We should be careful of downstream wrappers of DMD arguments and fix them before doing this. -target doesn't work, so there are doubts that there are even any downstream...