The old stdenv adapters were subtly wrong in two ways:
- `overrideAttrs` leaked the original, unoverridden `mkDerivation`.
- `stdenv.override` would throw away any manually-set `mkDerivation`
from a stdenv reverting to the original.
Now, `mkDerivation` is controlled (nearly directly) via an argument, and
always correctly closes over the final ("self") stdenv. This means the
adapters can work entirely via `.override` without any manual `stdenv //
...`, and both those issues are fixed.
Note hashes are changed, because stdenvs no previously overridden like
`stdenvNoCC` and `crossLibcStdenv` now are. I had to add some
`dontDisableStatic = true` accordingly. The flip side however is that
since the overrides compose, we no longer need to override anything but
the default `stdenv` from which all the others are created.
With removeUnknownConfigureFlags, it's impossible to express a package
that needs --enable-static, but will not accept --disable-shared,
without overriding the result of removeUnknownConfigureFlags _again_
in pkgs/top-level/static.nix.
It would be much better (and more in line with the rest of Nixpkgs) if
we encoded changes needed for static builds in package definitions
themselves, rather than in an ever-expanding list in static.nix. This
is especially true when doing it in static.nix is going to require
multiple overrides to express what could be expressed with stdenv
options.
So as a step in that direction, and to fix the problem described
above, here I replace removeUnknownConfigureFlags with a new stdenv
option, dontAddStaticConfigureFlags. With this mechanism, a package
that needs one but not both of the flags just needs to set
dontAddStaticConfigureFlags and then set up configureFlags manually
based on stdenv.hostPlatform.isStatic.
This logic should be in the pkgs/top-level/static.nix. We don’t want
to pollute Nixpkgs with =if stdenv.static=. Also, "static" is not
descriptive. We have two types of static stdenvs, ‘makeStaticLibaries’
and ‘makeStaticBinaries’. We shouldn’t rely on a static boolean like
this.
This allows one to force a compiler to use native machine optimizations. This
goes contrary to all the usual guarantees of Nix and so should be used only by
end-user and only in specific cases when they know what are they doing.
In my case this is needed to get a noticeable FPS boost in RPCS3 which is very
CPU-hungry PlayStation 3 emulator.
Existing "mips64el" should be "mipsel".
This is just the barest minimum so that nixpkgs can recognize them as
systems - although required for building individual derivations onto
MIPS boards, it is not sufficient if you want to actually build nixos on
those targets
By default, all previous overrides are discarded as before, as they
would only apply to the old host platform. But sometimes it is useful to
add some new ones, and this optional parameter allows that.
This is needed when cross-compiling for iOS (Aarch64 + Darwin). I also
changed the syntax of the Linux stdenv for visual consistency, though
that has no effect on semantics as the os is already guaranteed to be
Linux.
Eventually the adapter will be removed. Moved is
- Name suffix from hostPlatform
- configurePlatforms
To not cause more breakage, the default is currently [], but
eventually it will be [ "build" "host" ], as the cross adapter makes
it today.
Packages get --host and --target by default, but can explicitly request
any subset to be passed as needed. See docs for more info.
rustc: Avoid hash breakage by using the old (ignored)
dontSetConfigureCross when not cross building
See previous commit for what was done to `binutils` to make this
possible.
There were some uses of `forcedNativePackages` added. The
combination of overrides with that attribute is highly spooky: it's
often important that if an overridden package comes from it, the
replaced arguments for that package come from it. Long term this
package set and all the spookiness should be gone and irrelevant:
"Move along, nothing to see here!"
No hashes should be changed with this commit
Before all overrides were also pruned in the previous stage, now
only gcc and binutils are, because they alone care about about the
target platform. The rest of the overrides don't, so it's better to
preserve them in order to avoid spurious rebuilds.
This is required for Aarch64 since a lot of source tarballs ship with
outdated configure scripts that don't recognize aarch64. Simply
replacing the config.guess and config.sub with new versions from
upstream makes them build again.
This same approach is used by at least Buildroot and Fedora. In
principle this could be enabled for all architectures but
conditionalizing this on aarch64 avoids a mass rebuild on x86.
The long term goal is a big replace:
{ inherit system platform; } => buildPlatform
crossSystem => hostPlatform
stdenv.cross => targetPlatform
And additionally making sure each is defined even when not cross compiling.
This commit refactors the bootstrapping code along that vision, but leaves
the old identifiers with their null semantics in place so packages can be
modernized incrementally.
[N.B., this package also applies to the commits that follow it in the same
PR.]
In most cases, buildPackages = pkgs so things work just as before. For
cross compiling, however, buildPackages is resolved as the previous
bootstrapping stage. This allows us to avoid the mkDerivation hacks cross
compiling currently uses today.
To avoid a massive refactor, callPackage will splice together both package
sets. Again to avoid churn, it uses the old `nativeDrv` vs `crossDrv` to do
so. So now, whether cross compiling or not, packages with get a `nativeDrv`
and `crossDrv`---in the non-cross-compiling case they are simply the same
derivation. This is good because it reduces the divergence between the
cross and non-cross dataflow. See `pkgs/top-level/splice.nix` for a comment
along the lines of the preceding paragraph, and the code that does this
splicing.
Also, `forceNativeDrv` is replaced with `forceNativePackages`. The latter
resolves `pkgs` unless the host platform is different from the build
platform, in which case it resolves to `buildPackages`. Note that the
target platform is not important here---it will not prevent
`forcedNativePackages` from resolving to `pkgs`.
--------
Temporarily, we make preserve some dubious decisions in the name of preserving
hashes:
Most importantly, we don't distinguish between "host" and "target" in the
autoconf sense. This leads to the proliferation of *Cross derivations
currently used. What we ought to is resolve native deps of the cross "build
packages" (build = host != target) package set against the "vanilla
packages" (build = host = target) package set. Instead, "build packages"
uses itself, with (informally) target != build in all cases.
This is wrong because it violates the "sliding window" principle of
bootstrapping stages that shifting the platform triple of one stage to the
left coincides with the next stage's platform triple. Only because we don't
explicitly distinguish between "host" and "target" does it appear that the
"sliding window" principle is preserved--indeed it is over the reductionary
"platform double" of just "build" and "host/target".
Additionally, we build libc, libgcc, etc in the same stage as the compilers
themselves, which is wrong because they are used at runtime, not build
time. Fixing this is somewhat subtle, and the solution and problem will be
better explained in the commit that does fix it.
Commits after this will solve both these issues, at the expense of breaking
cross hashes. Native hashes won't be broken, thankfully.
--------
Did the temporary ugliness pan out? Of the packages that currently build in
`release-cross.nix`, the only ones that have their hash changed are
`*.gcc.crossDrv` and `bootstrapTools.*.coreutilsMinimal`. In both cases I
think it doesn't matter.
1. GCC when doing a `build = host = target = foreign` build (maximally
cross), still defines environment variables like `CPATH`[1] with
packages. This seems assuredly wrong because whether gcc dynamically
links those, or the programs built by gcc dynamically link those---I
have no idea which case is reality---they should be foreign. Therefore,
in all likelihood, I just made the gcc less broken.
2. Coreutils (ab)used the old cross-compiling infrastructure to depend on
a native version of itself. When coreutils was overwritten to be built
with fewer features, the native version it used would also be
overwritten because the binding was tight. Now it uses the much looser
`BuildPackages.coreutils` which is just fine as a richer build dep
doesn't cause any problems and avoids a rebuild.
So, in conclusion I'd say the conservatism payed off. Onward to actually
raking the muck in the next PR!
[1]: https://gcc.gnu.org/onlinedocs/gcc/Environment-Variables.html
