This makes the targetPrefix always conform to nixpkgs' expectations
instead of relying on the autotools build system to figure it out
correctly (which is also inconsistent across versions).
See also ca9be0511b.
The binutils build system checks by itself if it is building a cross
toolchain or not and prepends or omits a targetPrefix accordingly. This
means that we can always pass target via configureTargets.
However the binutils build system and our bintools wrapper disagree over
whether we are building a cross toolchain or not sometimes since cross
compilation can be relatively subtle in nixpkgs. For example every use
of crossOverlays will make nixpkgs build a cross toolchain even though
localSystem == crossSystem. The cross infrastructure is also used to
build native binaries with a different stdenv (musl instead of glibc,
clang instead of gcc). In all of these cases stdenv.hostPlatform.config
== stdenv.targetPlatform.config, causing binutils to not prepend a
target prefix. At the same time stdenv.hostPlatform !=
stdenv.targetPlatform causing the bintools wrapper to expect a target
prefix, thus building an incomplete set of bintools. This is why
currently pkgsCross.gnu64 and pkgsCross.musl64 aren't working.
The solution is quite simple however: If we detect that we are building
a cross toolchain in the binutils-unwrapped expression, we force the
targetPrefix with --programprefix and fulfill the expectations of the
bintools wrapper at the same time.
Tested (on x86_64-linux):
* pkgsCross.musl64.hello
* pkgsCross.aarch64-multiplatform.hello
* pkgs.hello
Still not working is pkgsCross.gnu64, since
x86_64-unknown-linux-gnu-stage-final-gcc gets confused about targets
now, so bootstrapping the stdenv fails. Since this wasn't working
previously anyways, it's proably fine to fix this separately.
Instead of always supplying flags, apply the flags as defaults. Use
clang's native flags instead of lifting the linker flags from binutils
with `-Wl,`.
If a project is using clang to drive linking, make clang do the right
thing with MACOSX_DEPLOYMENT_TARGET. This can be overridden by command
line arguments. This will cause modern clang to pass
`-platform_version 10.12 0.0.0`, since it doesn't know about the SDK
settings. Older versions of clang will pass down `-macos_version_min`
flags with no sdk version.
At the linker layer, apply a default value for anything left
ambiguous. If nothing is specified, pass a full
`-platform_version`. If only `-macos_version_min` is specified, then
lock down the sdk_version explicitly with `-sdk_version`. If a min
version and sdk version is passed, do nothing.
Since 03eaa48 added perl.withPackages, there is a canonical way to
create a perl interpreter from a list of libraries, for use in script
shebangs or generic build inputs. This method is declarative (what we
are doing is clear), produces short shebangs[1] and needs not to wrap
existing scripts.
Unfortunately there are a few exceptions that I've found:
1. Scripts that are calling perl with the -T switch. This makes perl
ignore PERL5LIB, which is what perl.withPackages is using to inform
the interpreter of the library paths.
2. Perl packages that depends on libraries in their own path. This
is not possible because perl.withPackages works at build time. The
workaround is to add `-I $out/${perl.libPrefix}` to the shebang.
In all other cases I propose to switch to perl.withPackages.
[1]: https://lwn.net/Articles/779997/
