Fixes#256769
Repro + test of fix here: https://gitlab.com/ramirez7/bug-repros/-/merge_requests/1
Adds a `srcModifier` argument to `callCabal2nixWithOptions` to allow
customizing the source files used to generate the cabal file (e.g. to
support `hpack`/`package.yaml`).
The original change missed that we used to erroneously provide LLVM on
aarch64-linux, but not on aarch64-darwin which was later fixed
in #246045. Since the change was tested on aarch64-linux prior to this,
tls ended up being broken on both aarch64-linux and aarch64-darwin.
This commit resolves the situation by introducing a helper function to
achieve the effect of passing -fllvm while ensuring LLVM is provided.
Resolves#260013.
ghcWithHoogle was using unoverridden packages pulled from the
buildHaskellPackages package set. This commit fixes that by pulling the
packages from self instead.
ghc and also ghcWithPackages (when taken from a haskell package set) are
a bit weird—in the same way stdenv is: ghc is actually from
buildPackages (pkgsBuildHost) wheras the main package set belongs to
pkgsHostTarget. ghc (and stdenv) is included in the package set due to
its special relation to the set: it is built by that ghc, so constituted
by the compiler in a manner of speaking.
For ghc this works in a straightforward way: It is packaged
independently from the haskell package sets and passed to
make-package-set.nix to create the different sets we expose.
With ghcWithPackages an error crept in, though: Since it needs to
receive the haskellPackages fix point (and thus can't be instantiated
before the package set), it is defined in make-package-set.nix. Here it
was neglected to make sure that it also has the same scope as ghc, i.e.
buildHaskellPackages/buildPackages: Otherwise the shell the wrapper
scripts use to invoke ghc (originally from buildPackages) would be from
pkgsHostTarget—in the cross case, the wrapper scripts would be
executable by neither host nor build platform. We want them to work on
the build platform, though.
Note that this creates a weird mismatch where it is hard to see which of
the alternatives would be more natural: ghcWithPackages and
ghcWithHoogle now use packages from the package set they are a member
of, but have *-ghc and hoogle executables that are executable on the
build platform. This works because ghc originates from buildPackages (as
discussed) and hoogleWithPackages is taken from buildHaskellPackages.
This does imply though that while set.ghcWithHoogle will be executable
on the build platform, set.hoogleWithPackages will be executable on the
host platform—both will use the fix point of set for the package
selector function. This is maybe a confusing asymmetry, but it seems
like a valid use case to cross-compile a hoogle instance. Most
development tools use ghcWithHoogle (or equivalent), so that use case is
covered as well in principle.
Packages built with `haskellPackages.callHackage` won't be rebuilt when
updating `all-cabal-hashes`.
The removed comment was keeping a reference to the `cabal2nix` call,
which itself depends on `all-cabal-hashes`, in order to keep this file
during a garbage collection.
The tradeoff is between:
- The current behavior: a mass rebuild, any change of `all-cabal-hashes`
triggers a rebuild of all the packages built with `callHackage` and
packages which depend on them. This can take hours, and may happen
after a "small" unrelated change (i.e. an user is bumping
`all-cabal-hashes` in order to use a new package from hackage). It
also have global impacts in a project (long rebuild in CI, new entries
in cache, developers need to fetch the new entries, ...). In this
context, `cabal2nix` entries are not garbage collected.
- The new behavior: No mass rebuild, but `cabal2nix` derivations need to
be recomputed after a garbage collection. This is usually fast (a few
seconds by call), linear with the number of calls and should not
happen a lot (i.e. users are not garbage collecting everyday).
See https://github.com/NixOS/nixpkgs/issues/194751 for details.
Deprecate haskell.lib{,.compose}.generateOptparseApplicativeCompletion*
in favor of the newly added
haskell.packages.*.generateOptparseApplicativeCompletions (plural!)
which takes into account whether we are cross-compiling or not. If we
are, generating completions is disabled, since we can't execute software
built for a different platform.
The move is necessary, so we can receive the /same/ stdenv as the
package we are overriding in order to accurately check whether we can
execute produced binaries.
Resolves#174040.
Resolves#49648.
* all-cabal-hahses can be a directory
Motivating problem: we want to use the flake.lock for all-cabal-hashes
as a github repository as our revision pin, and then import that to pass
as all-cabal-hashes. To do that we need it to accept a directory rather
than a tarball.
* all-cabal-hashes: add some comments
This is achieved by passing the entire package set to the respective
wrappers and passing the select function as a second attribute. Together
with the new support for callPackage-ing functions this allows for
things like `ghcWithPackages.override { useLLVM = true; } (p: [ … ])`.
To make this possible for `ghcWithHoogle` as well, we need to make the
wrapper a bit more bespoke and inline the hoogle feature as well. The
hoogle wrapper, however, can remain separate and is exposed as
`hoogleWithPackages` additionally, as it can also serve standalone use.
`hoogleLocal` is kept for backwards compatibility (including the old,
suboptimal API), but will inform users about the better alternative via
a warning.
Overrideable functions are possible by wrapping them as functors in an
attribute set first and appending the overrideScope and override
attributes later.
cabal only allows one cabal file per directory and cabal2nix will find
the correct file easily. Thus we don't need to know ahead of time which
cabal file to select. One downside of this is of course that we may
include unnecessary files in subdirectories, which could possibly be
fixed by enforcing depth.
This is a new argument to the Haskell builder's 'shellFor' which allows
the caller to adjust the 'genericBuilderArgs' after they are built.
One immediate application for this is that it allows users to disable
tests, enable benchmarks, etc. for _all_ packages in a given Haskell
package set.
This is often useful when so many of the derivations in the
package set have changed that most of the cache has been invalidated,
and there is no need for the user to run tests on their dependencies,
but they still want these dependencies available in their 'nix-shell'
environment for some package-under-development.
Current, the `cabal2nix` derivation contains both the executable, and a wrapper
that adds `nix` and `nix-prefetch-scripts`, which are required for some
features.
However, when calling `callCabal2nix` to create a derivation from a cabal file
at evaluation time,
these features are not actually used, but the huge closure of
`nix-prefetch-scripts` (which includes multiple vcs, as well as python and perl)
still needs to be fetched.
This commit splits cabal2nix into a lightweight version that is a standalone
static binary (`cabal2nix-unwrapped`), and a wrapper that includes the proper
dependencies in the path for full usage of the command line
utility (`cabal2nix`).
This commit also switches to the default ghc, to reduce the likelyhood of
building a different ghc when calling `callCabal2nix`.
This makes it work like work-on-multi from Reflex Platform. In
particular, rather than making `.env` from `shellFor`, we make `.env`
the primitive, and `shellFor` works by combining together the arguments
of all the packages to `generic-builder` and taking the `.env` of the
resulting mashup-package.
There are 2 benefits of this:
1. The dependency logic is deduplicated. generic builder just concatted
lists, whereas all the envs until now would sieve apart haskell and
system build inputs. Now, they both decide haskell vs system the same
way: according to the argument list and without reflection.
Consistency is good, especially because it mean that if the build
works, the shell is more likely to work.
2. Cross is handled better. For native builds, because the
`ghcWithPackages` calls would shadow, we through both the regular
component (lib, exe, test, bench) haskell deps and Setup.hs haskell
deps in the same `ghcWithPackages` call. But for cross builds we use
`buildPackages.ghcWithPackages` to get the setup deps. This ensures
everything works correctly.
