In https://github.com/NixOS/nixpkgs/pull/209870 I tried to unify the
treatment of clang and gcc in cc-wrapper as much as possible.
However it appears that I went too far.
Clang requires -isystem flags in order to be able to find gcc's
libstdc++. Gcc does not need these flags. If they are added,
gfortran will get confused:
https://github.com/NixOS/nixpkgs/pull/209870#issuecomment-1500550903
This commit deunifies the chunk of code that adds the -isystem
flags, and explains why this chunk applies only to clang.
When wrapping `clang` and using a `gccForLibs` whose `libgcc` is in
its own output (rather than the `lib` output), this commit will adds
`-L${gccForLibs.libgcc}/lib` to `cc-ldflags`.
If that flag is not added, `firefox` will fail to compile because it
invokes `clang-wrapper` with `-fuse-ld=lld` and passes `-lgcc_s` to
`lld`, but does not tell `lld` where to find `libgcc_s.so`. In that
situation, firefox will fail to link.
Without this change $target-cpp is used unwrapped and is missing
standard header search paths among other things).
Example failure:
$ nix build -f. -L pkgsStatic.netbsd.compat
...
> checking how to run the C preprocessor... x86_64-unknown-linux-musl-cpp
> configure: error: in `/build/cvs-export/tools/compat':
> configure: error: C preprocessor "x86_64-unknown-linux-musl-cpp" fails sanity check
> See `config.log' for more details
cc-wrapper has essentially two separate codepaths: the `gccForLibs`
codepath, used only by non-gcc (i.e. clang) compilers, and the
"other" codepath.
This PR allows non-clang compilers to opt-in to the `gccForLibs`
codepath (off by default). To allow this, a new parameter
`ccForLibs` is exposed, since it would be extremely confusing for
gcc to be able to use `gccForLibs` but not do so by default.
This reverts commit 8c80bd08b7
("build-support/cc-wrapper: pass in non-existent --sysroot= to untangle
from libc").
This change was good in spirit: we caught a few genuine problems with
`scons` based packages (`godot`, `fluxus`) and unexpected `-idirafter`
includes in various boot loadres (`ipxe`, wimboot`):
https://github.com/NixOS/nixpkgs/pull/210004#issuecomment-1407162693
Unfortunately `--sysroot=` also has a negative impact on libary search
order for DT_NEEDED libraries and RUNPATHs of linked libraries. This
unexpectedly broke `dmd`, `d-seams`, `llvmPackages_rocm.compiler-rt`).
An interesting case of unexpected breakage is `usbmuxd2` where the bug
exposed incomplete library move on `libstdc++fs` in `gcc`.
The library breakage is very non-intuitive (on top of already unusual
layout of `cc-wrapper` driver). Let's revert this change for now.
Once it lands we can undo `--sysroot=/` workarounds merged for
`staging-next`.
After https://github.com/NixOS/nixpkgs/pull/210004 `usbmuxd2` started
failing to build as:
usbmuxd2-unstable> .../ld: cannot find -lstdc++fs: No such file or directory
usbmuxd2-unstable> clang-11: error: linker command failed with exit code 1 (use -v to see invocation)
This started happening because #210004 exposed a long-standing bug in
`gcc` derivation: `cc.lib` is missing `libstdc++fs` library:
$ find $(nix-build --no-link -A stdenv.cc.cc.lib) | fgrep libstdc | unnix
/<<NIX>>/gcc-11.3.0-lib/lib/libstdc++fs.la
/<<NIX>>/gcc-11.3.0-lib/lib/libstdc++.la
/<<NIX>>/gcc-11.3.0-lib/lib/libstdc++.so.6.0.29
/<<NIX>>/gcc-11.3.0-lib/lib/libstdc++.so
/<<NIX>>/gcc-11.3.0-lib/lib/libstdc++.so.6
It was not moved from `cc.out` output:
$ find $(nix-build --no-link -A stdenv.cc.cc) | fgrep libstdc | unnix
/<<NIX>>/gcc-11.3.0/lib/libstdc++.a
/<<NIX>>/gcc-11.3.0/lib/libstdc++fs.a
This change adds `cc` library lookup path back to `staging-next` until
`gcc` is fixed.`
FreeBSD doesn't use LLVM's cxxabi implementation, for backwards
compatibility reasons. Software expects the libcxxrt API when
building on FreeBSD. This fixes the build of
pkgsCross.x86_64-freebsd.boost.
Gcc does not allow `-march=` on PowerPC:
https://gcc.gnu.org/onlinedocs/gcc-12.2.0/gcc/RS_002f6000-and-PowerPC-Options.html#RS_002f6000-and-PowerPC-Options
Instead, `-mcpu=` should be used to set the minimum instruction set
and `-mtune=` is used to optimize instruction scheduling for a
specific processor. Both flags take the same set of valid values,
which includes `native`.
This commit causes `isGccArchSupported` to return `false` for PowerPC
targets so we never pass an `-march=` flag, since that will always be
rejected by gcc.
I would like to add an extra `gcc` build step during linux bootstrap
(https://github.com/NixOS/nixpkgs/issues/208412). This makes it early
bootstrap compiler linked and targeted against `bootstrapTools` `glibc`
including it's headers.
Without this change `gcc`'s spec files always prefer `bootstrapTools` `glibc`
for header search path (passed in as --with-native-system-header-dir=). We'can't
override it with:
- `-I` option as it gets stacked before gcc-specific headers, we need to keep
glibc headers after gcc as gcc cleans namespace up for C standard by using
#include_next and by undefining system macros.
- `-idirafter` option as it gets appended after existing `glibc`-includes
This `--sysroot=/nix/store/does/not/exist` hack allows us to remove existing
`glibc` headers and add new ones with `-idirafter`.
We use `cc-cflags-before` instead of `libc-cflags` to allow user to define
their own `--sysroot=` (like `firefox` does).
To keep it working prerequisite cross-symlink in gcc.libs is required:
https://github.com/NixOS/nixpkgs/pull/209153
Switches that gnatmake needs to pass to gcc must be given as
"-cargs <gcc_switches>" after at least the files to compile (see the
gnatmake docs for all the subtleties). This commit makes that happen,
in a way that triggers rebuilds only for things that actually depend on
GNAT, and not the other compilers contained in GCC.
On darwin clang driver always sets -D_FORTIFY_SOURCE=0 under asan.
This causes -Werror to trip over macro redefinition:
<command line>:1:9: error: '_FORTIFY_SOURCE' macro redefined [-Werror,-Wmacro-redefined]
#define _FORTIFY_SOURCE 2
^
To avoid it let's always explicitly undefine it first before redefining.
Currently, clang++ statically links against libstdc++ instead of
dynamically links, because the -L path included in the cc-wrapper is
incorrect. The gccForLibs.lib output only contains the architecture
subdirectory if the target and host platform are not the same. (See
targetConfig set in gcc/<version>/default.nix and the gcc/builder.nix)
This fixes the incorrect linking by using the correct path for both the
cross and native cases. This also matches the cc_solib set above in
cc-wrapper/default.nix
Tested by compiling a simple cpp binary and noting that it now correctly
dynamically links against libstdc++, natively on x86_64 and arm64, as
well as x86_64 -> arm64 cross compilation.
Co-Authored-By: Sebastian Ullrich <sebasti@nullri.ch>
There is context here that I needed when resolving an issue in which
libc was added to NIX_CFLAGS_COMPILE before the C++ stdlib that took
me awhile to understand.
It was suggested to me that this context be included as a comment,
since it is not obvious and could help others in the future.
POSIX sh (and `bash`) impose a restriction on environment variable name
format and disallow hypheps in the names. Normally it's not a problem
as nothing usually tries to refer nyphenated names.
One exception is `nix develop` (https://github.com/NixOS/nix/issues/6848):
$ nix develop -f. gcc -L
gcc-wrapper> ...-get-env.sh: line 70: expand-response-params: bad substitution
Note that bash usually uses explicitly created `expandResponseParams`
variant of the same variable.
To work the problem around let's avoid environment variable export and move
it to `passthru` for `cc` (used ina few places) and remove it completely for
`binutils` (does not seem to be used at all).
Noticed this bug when was trying to bootstrap m4 on darwin. That fixes
line 163: no such file or directory error
That does not solve all problems staging has on darwin.
This enables users to make use of clang's multi-platform/target support
without having to go through full cross system setup. This is especially useful
for generating bpf object files, I'm not even usre what would a no-userland
cross compile system tuple even look like to even try going that route.
Fixes#176128
Otherwise, these warnings are emitted:
command-line option '-Wformat=1' is valid for C/C++/ObjC/ObjC++ but not for Fortran
command-line option '-Wformat-security' is valid for C/C++/ObjC/ObjC++ but not for Fortran
'-Werror=' argument '-Werror=format-security' is not valid for Fortran
Fixes part of #27218
Take a "nixSupport" argument that is an attrset of lists to append
to "$out/nix-support/${name}" where name is the name of the
attribute. This attrset is available from the passthru of the
wrapped compiler.
This is an alternative to imperatively issuing flags with
extraBuildCommands and makes it possible to append or filter
flags with an override.
In https://github.com/NixOS/nixpkgs/issues/158042 I noticed order
mismatch as a bootstrap build failure when building x86_64-linux
against glibc-2.35 in nixpkgs (bootstrap libs has glibc-2.27):
expand-response-params> ld: /nix/store/p4s4jf7aq6v6z9iazll1aiqwb34aqxq9-bootstrap-tools/lib/crt1.o: in function `_start':
expand-response-params> /build/glibc-2.27/csu/../sysdeps/x86_64/start.S:101: undefined reference to `__libc_csu_fini'
expand-response-params> ld: /build/glibc-2.27/csu/../sysdeps/x86_64/start.S:102: undefined reference to `__libc_csu_init'
expand-response-params> collect2: error: ld returned 1 exit status
Here crt1.o from glibc-2.27 links against libc.so.6 from glibc-2.35.
This happens because ordering of `-L` (influences `libc.so` lookup) and
`-B` (influences `crt1.o` lookup) flags differs:
expand-response-params> -B/...-bootstrap-tools/lib
expand-response-params> -B/...-glibc-2.35/lib/
...
expand-response-params> -L/...-glibc-2.35/lib
expand-response-params> -L/...-bootstrap-tools/lib
The change makes consistent ordering of `-L`/`-B` and allows getting to
stage4 for `glibc-2.35` target.
So far we've ignored response files in arguments, and did not
check linkType against expanded parameters. This means if
we have `-static` in a @reponse-file, linkType will not be
set to `-static` as we never check against the expanded arguments
from response files.
fixes e.g.:
pkgsMusl.libfsm
pkgsMusl.libiscsi
pkgsMusl.nsjail
pkgsMusl.pv
match strings have whitespace on either side, which wasn't
matching leading/trailing arguments previously
fixes:
pkgsMusl.bulletml
pkgsMusl.proot
pkgsMusl.python3
Debian explains this issue well in the dpkg-buildflags manpage:
-fPIE
Can be linked into any program, but not a shared library (recommended).
-fPIC
Can be linked into any program and shared library.
On projects that build both programs and shared libraries you might need to
make sure that when building the shared libraries -fPIC is always passed last
(so that it overrides any previous -PIE) to compilation flags such as CFLAGS.
(from https://manpages.debian.org/bullseye/dpkg-dev/dpkg-buildflags.1.en.html#hardening)
In newer versions of mingw, programs compiled with FORTIFY_SOURCE need
to link to libssp or they will have link-time errors.
gmp has been broken since @pstn updated mingw-64 in c60a0b0447
GPRbuild is a multi language build system developed by AdaCore which
is mostly used for build Ada-related projects using GNAT.
Since GPRbuild is used to build itself and its dependency library
XML/Ada we first build a bootstrap version of it using the provided
bash build script bootstrap.sh as the gprbuild-boot derivation.
gprbuild-boot is then used to build xmlada and the proper gprbuild
derivation.
GPRbuild has its own search path mechanism via GPR_PROJECT_PATH which
we address via a setupHook. It currently works quite similar to the
pkg-config one: It accumulates all inputs into GPR_PROJECT_PATH,
GPR_PROJECT_PATH_FOR_BUILD etc. However this is quite limited at the
moment as we don't have a gprbuild wrapper yet which understands the
_FOR_BUILD suffix. However, we'll need to address this in the future
as it is currently basically impossible to test since the distinction
only affects cross-compilation, but it is not possible to build a GNAT
cross-compiler in nixpkgs at the moment (I'm working on changing that,
however).
Another issue we had to solve was GPRbuild not finding the right GNAT
via its gprconfig tool: GPRbuild has a knowledge base with compiler
definitions which run some checks and collect info about binaries
which are in PATH. In the end the first compiler in PATH that supports
the desired language is selected.
We want GPRbuild to discover our wrapped GNAT since the unwrapped one
is incapable of producing working binaries since it won't find the
crt*.o objects distributed with libc. GPRbuild however needs to find
the Ada runtime distributed with GNAT which is not part of the wrapper
derivation, so it will skip the wrapper and select the unwrapped GNAT.
Symlinking the unwrapped's lib directory into the wrapper fixes this
problem, but breaks linking in some cases (e. g. when linking against
OMP from gcc, the runtime variant will shadow the problem dynamic lib
from buildInputs). Additionally it uses gnatls as an indicator it has
found GNAT which is not part of the wrapper.
The solution we opted to adopt here is to install a custom compiler
description into gprbuild's knowledge base which properly detects the
nixpkgs GNAT wrapper: It uses gnatmake to detect GNAT instead of
gnatls and discovers the runtime via a symlink we add to
`$out/nix-support`. This additional definition is enough to properly
detect GNAT, since the plain wrapped gcc detection works out of the
box. It may, however, be necessary to add special definitions for
other languages in the future where gprbuild also needs to discover
the runtime.
One future improvement would be to install libgpr into a separate
output or split it into a separate derivation (which would require to
link gprbuild statically always since otherwise we end up with a
cyclical dependency).
This will begin the process of breaking up the `useLLVM` monolith. That
is good in general, but I hope will be good for NetBSD and Darwin in
particular.
Co-authored-by: sterni <sternenseemann@systemli.org>
This PR adds a new aarch64 android toolchain, which leverages the
existing crossSystem infrastructure and LLVM builders to generate a
working toolchain with minimal prebuilt components.
The only thing that is prebuilt is the bionic libc. This is because it
is practically impossible to compile bionic outside of an AOSP tree. I
tried and failed, braver souls may prevail. For now I just grab the
relevant binaries from https://android.googlesource.com/.
I also grab the msm kernel sources from there to generate headers. I've
included a minor patch to the existing kernel-headers derivation in
order to expose an internal function.
Everything else, from binutils up, is using stock code. Many thanks to
@Ericson2314 for his help on this, and for building such a powerful
system in the first place!
One motivation for this is to be able to build a toolchain which will
work on an aarch64 linux machine. To my knowledge, there is no existing
toolchain for an aarch64-linux builder and an aarch64-android target.
Also begin to start work on cross compilation, though that will have to
be finished later.
The patches are based on the first version of
https://reviews.llvm.org/D99484. It's very annoying to do the
back-porting but the review has uncovered nothing super major so I'm
fine sticking with what I've got.
Beyond making the outputs work, I also strove to re-sync the packages,
as they have been drifting pointlessly apart for some time.
----
Other misc notes, highly incomplete
- lvm-config-native and llvm-config are put in `dev` because they are
tools just for build time.
- Clang no longer has an lld dep. That was introduced in
db29857eb3, but if clang needs help
finding lld when it is used we should just pass it flags / put in the
resource dir. Providing it at build time increases critical path
length for no good reason.
----
A note on `nativeCC`:
`stdenv` takes tools from the previous stage, so:
1. `pkgsBuildBuild`: `(?1, x, x)`
2. `pkgsBuildBuild.stdenv.cc`: `(?0, ?1, x)`
while:
1. `pkgsBuildBuild`: `(?1, x, x)`
2. `pkgsBuildBuild.targetPackages`: `(x, x, ?2)`
3. `pkgsBuildBuild.targetPackages.stdenv.cc`: `(?1, x, x)`
In a typical build environment the toolchain will use the value of the
MACOSX_DEPLOYMENT_TARGET environment variable to determine the version
of macOS to support. When cross compiling there are two distinct
toolchains, but they will look at this single environment variable. To
avoid contamination, we always set the equivalent command line flag
which effectively disables the toolchain's internal handling.
Prior to this change, the MACOSX_DEPLOYMENT_TARGET variable was
ignored, and the toolchains always used the Nix platform
definition (`darwinMinVersion`) unless overridden with command line
arguments.
This change restores support for MACOSX_DEPLOYMENT_TARGET, and adds
nix-specific MACOSX_DEPLOYMENT_TARGET_FOR_BUILD and
MACOSX_DEPLOYMENT_TARGET_FOR_TARGET for cross compilation.
The check for including the C++ standard library headers was nested inside the
check for linking with the C++ standard library. As a result, the `-nostdlib`
flag incorrectly implied `-nostdinc++`, which made it virtually impossible to
partially link C++ objects.
Fixes build failures with clang:
clang-7: error: unknown argument: '-fPIC -target'
clang-7: error: no such file or directory: '@<(printf %qn -O2'
clang-7: error: no such file or directory: 'x86_64-apple-darwin'
Introduced by 60c5cf9cea in #112449
The `platform` field is pointless nesting: it's just stuff that happens
to be defined together, and that should be an implementation detail.
This instead makes `linux-kernel` and `gcc` top level fields in platform
configs. They join `rustc` there [all are optional], which was put there
and not in `platform` in anticipation of a change like this.
`linux-kernel.arch` in particular also becomes `linuxArch`, to match the
other `*Arch`es.
The next step after is this to combine the *specific* machines from
`lib.systems.platforms` with `lib.systems.examples`, keeping just the
"multiplatform" ones for defaulting.
When invoking a simple Ada program with `gcc` from `gnats10`, the
following warnings are shown:
```
$ gcc -c conftest.adb
gnat1: warning: command-line option ‘-Wformat=1’ is valid for C/C++/ObjC/ObjC++ but not for Ada
gnat1: warning: command-line option ‘-Wformat-security’ is valid for C/C++/ObjC/ObjC++ but not for Ada
gnat1: warning: ‘-Werror=’ argument ‘-Werror=format-security’ is not valid for Ada
$ echo $?
0
```
This is only spammy when compiling Ada programs inside a Nix derivation,
but certain configure scripts (such as the ./configure script from the
gcc that's built by coreboot's `make crossgcc` command) fail entirely
when getting that warning output.
https://nixos.wiki/wiki/Coreboot currently suggests manually running
> NIX_HARDENING_ENABLE="${NIX_HARDENING_ENABLE/ format/}" make crossgcc
… but actually teaching the nixpkgs-provided cc wrapper that `format`
isn't supported as a hardening flag seems to be the more canonical way
to do this in nixpgks.
After this, Ada programs still compile:
```
$ gcc -c conftest.adb
$ echo $?
0
```
And the compiler output is empty.
We need to set FC so that CMake and other tools can find the fortran
compiler. Also we need to limit the hardening flags since fortify and
format don’t work with fortran.
Fixes#88449
I hate the thing too even though I made it, and rather just get rid of
it. But we can't do that yet. In the meantime, this brings us more
inline with autoconf and will make it slightly easier for me to write a
pkg-config wrapper, which we need.
