* gnu/packages/rust.scm (rust-1.44): Do not replace LLVM version, for
simplicity. It used to be required at this point because LLVM 7 was no longer
supported for rust-1.44, but we're now using LLVM 9.
(rust-1.46): Move all additional phases, doc output to rust-1.54.
(rust-1.53, rust-1.54): New variables.
(rust-1.54): Adjust snippet field.
(rust): Set default Rust to the latest version, 1.54.0.
[native-inputs]: Use the latest GDB.
Only stage 1 (rather than stage 2) rustc and cargo are built and the
installation phase rewritten to not invoke the build tool, which helps to
shorten the build time to about 10 minutes per Rust on a fast machine. The
total build time should take less than 4h30, down from the current 8 hours on
a Ryzen 3900X CPU.
* gnu/packages/patches/rust-reproducible-builds.patch: Delete file.
* gnu/packages/patches/rustc-1.39.0-src.patch: New file.
* gnu/local.mk (dist_patch_DATA): Register it, and un-register
rust-reproducible-builds.patch.
* gnu/packages/rust.scm (%mrustc-commit): New variable.
(%mrustc-source): Update to latest commit.
(rust-1.29): Morph into...
(rust-1.39): ... this.
[source]: Adjust the snippet, patches and patch-flags fields.
[inputs]: Replace llvm-7 by llvm (9).
[make-flags]: Add the RUSTC_VERSION, MRUSTC_TARGET_VER and OUTDIR_SUF make
variables. Remove the RUSTCSRC make variable.
[phases]{copy-mrustc-and-patch}: Rename to...
{setup-mrustc-sources}: ... this. A symbolic link is created inside the
mrustc directory, pointing to the Rust 1.39.0 sources.
{patch-makefiles}: Adjust directory. Patch date and git definitions. Edit
out the RUSTC_SRC_DL prerequisite. Adjust the patching of a shebang.
{patch-cargo-checksums}: Adjust.
{configure-cargo-home}: New phase.
{configure}: Create and add a 'cc' shim to PATH.
{build}: Do not invoke make in parallel mode inside the run_rustc directory.
(rust-1.30): Morph into...
(rust-1.40): ... this, integrating the changes introduced between 1.30 and
1.40.
[modules]: Properly import (guix build cargo-utils).
[phases]{add-cc-shim-to-path}: New phase.
{configure}: Increase the codegen-units value to 256.
{build}: Only build stage 1 rustc and cargo, and group the commands into one
invocation.
{install}: Manually install the stage 1 build artifacts.
{patch-cargo-checksums}: Remove the ad-hoc use-modules, no longer needed.
[source]: Adjust.
[arguments]: Set validate-runpath? to #f.
[phases]{patch-cargo-checksums}: Remove phase.
{configure}: Repatriate the jemalloc configuration changes from 1.40.
[native-inputs]: Replace the rust-1.29 inputs by rust-1.39. Use regular
jemalloc and llvm versions.
(rust-1.41)[phases]: Add the patch-cargo-checksums phase.
(rust-1.31, rust-1.32, rust-1.33, rust-1.34, rust-1.35, rust-1.36, rust-1.37)
(rust-1.38, rust-1.39, rust-1.40): Delete variables.
(rust-1.46)[phases]{install}: Group build, test and install commands.
(rust-1.47)[phases]{build}: Override to adjust for the relocation of the
standard library source directory.
Fixes <https://issues.guix.gnu.org/50105>.
* gnu/packages/aux-files/python/sitecustomize.py: Add a comment explaining the
general idea, and use sys.prefix instead of sys.executable.
(major_minor): Use the unpacking operator (*) to provide the arguments.
(site_packages_prefix): Use os.path.join to form the path.
(python_site): Likewise. Use sys.prefix instead of sys.executable.
(all_sites_raw): Split on os.path.pathsep.
(sys.path): Directly splice the result in the list.
Suggested-by: Hartmut Goebel <h.goebel@crazy-compilers.com>
Reported-by: Mathieu Othacehe <othacehe@gnu.org>
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
Fixes <https://issues.guix.gnu.org/50066>.
* guix/packages.scm (patch-and-repack): Test for a tarball using tarball? and
move the plain file copy to the else clause.
Reported-by: Mathieu Othacehe <othacehe@gnu.org>
This allows creating wrappers for other linkers than "ld", such as "ld.gold".
* gnu/packages/base.scm (make-ld-wrapper): New variable.
[linker]: New argument. Update doc.
<ld>: Use LINKER to derive the linker source.
GCC 5 has a misaligned access to a vector pointer which causes a build failure
when using GCC 8 and later. Backport upstream fix that was applied on the GCC
6 branch.
* gnu/packages/gcc.scm (gcc-5)[source]: Apply
gcc-5-fix-powerpc64le-build.patch.
* gnu/packages/patches/gcc-5-fix-powerpc64le-build.patch: New file.
* gnu/local.mk (dist_patch_DATA): Register it.
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
* guix/build/qt-utils.scm (find-files-to-wrap): Exclude already wrapped
programs. This is forbidden in wrap-program now due to a738a663a9.
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
‘coreutils-gnulib-tests.patch’ is applied in the gnulib version shipped
with 3.8.
* gnu/packages/base.scm (diffutils): Update to 3.8.
[source]: Do not apply coreutils-gnulib-tests.patch.
Modified-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
diffutils has a race condition in its signal processing code which is easy to
trigger on powerpc64le-linux. More often than not, it causes the ‘colors’
test to fail and therefore the build of the package fails as well.
Add the patch proposed in Debian bug 922552 which fixes the problem.
* gnu/packages/patches/diffutils-fix-signal-processing.patch: New file.
* gnu/local.mk (dist_patch_DATA): Add it.
* gnu/packages/base.scm (diffutils)[source]: Use it.
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
The signal callbacks in the ‘builder’ testcase have wrong prototypes.
This causes it to fail the “/Builder/Signal Autoconnect” test on
powerpc64le-linux.
Solve the problem by backporting the upstream fix that was applied to
GTK+ 3.0.
* gnu/packages/patches/gtk2-fix-builder-test.patch: New file.
* gnu/local.mk (dist_patch_DATA): Add it.
* gnu/packages/gtk.scm (gtk+-2): Use it.
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
Fixes <https://issues.guix.gnu.org/50830>.
Reported by podiki on #guix.
Previously, glibc 2.33 would look for
/gnu/store/...-glibc-2.33/etc/localtime instead of /etc/localtime.
* gnu/packages/base.scm (glibc)[arguments]: Remove #:make-flags.
[phases]: In 'pre-configure', modify 'inet/Makefile' instead of
'sunrpc/Makefile' since this is where these bits are in 2.33.
(glibc-2.31)[arguments]: Add 'set-etc-rpc-installation-directory'
phase.
(glibc-2.30): Inherit from GLIBC-2.31.
* gnu/tests/base.scm (run-basic-test)["libc honors /etc/localtime"]: New
test.
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
This change had no effect already in glibc 2.29, which has proper
versioning.
* gnu/packages/base.scm (glibc)[arguments]: Remove obsolete
'nscd/nscd_stat.c' 'substitute*' statement.
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
This resolves a build failure following the GStreamer update to 1.18.5.
* gnu/packages/networking.scm (libnice): Update to 0.1.18-0.47a9633.
[phases]{disable-failing-tests}: Comment out rather than delete
test-set-port-range; disable the bind test, which is flaky.
There are still many parts of the build that happen sequentially, but at least
this hastens the long build somewhat.
* gnu/packages/machine-learning.scm (tensorflow)[phases]: Remove trailing #t.
{build-pip-package}: Honor the PARALLEL-BUILD? argument.
This adjustment is made considering the contributions to the (gnu packages
polkit) module in commits cb151c68bf and 222e3319bd.
Suggested-by: Morgan Smith <Morgan.J.Smith@outlook.com>
But keep the older variant for Tensorflow.
* gnu/packages/cpp.scm (abseil-cpp-20200923.3): Renamed older variant.
(abseil-cpp): New variable.
* gnu/packages/machine-learning.scm (tensorflow-lite)
[inputs]: Adjust to use the older abseil-cpp-20200923.3.
* gnu/packages/rpc.scm (grpc-1.16.1)[native-inputs]: Likewise.
Co-authored-by: Greg Hogan <code@greghogan.com>
Fixes <https://issues.guix.gnu.org/50957>.
This reverts the following commits:
- 42231bc (gnu: cheese: Propagate gdk-pixbuf+svg rather than gdk-pixbuf)
- ad6a950 (gnu: mutter: Propagate gdk-pixbuf+svg rather than gdk-pixbuf)
- 5e02d1a (gnu: cogl: Propagate gdk-pixbuf+svg rather than gdk-pixbuf)
- 5c467b2 (gnu: libgweather: Propagate gdk-pixbuf+svg rather than gdk-pixbuf)
- 26487f6 (gnu: librsvg-next: Do not propagate gdk-pixbuf)
- d891922 (gnu: librsvg: Do not propagate gdk-pixbuf)
- 22476d3 (gnu: libgsf: Do not propagate gdk-pixbuf)
- 099ca0e (gnu: libnotify: Propagate gdk-pixbuf+svg rather than gdk-pixbuf)
- fb6a1ff (gnu: appstream-glib: Propagate gdk-pixbuf+svg rather than gdk-pixbuf)
- 4fb525d (gnu: weasyprint: Propagate gdk-pixbuf+svg rather than gdk-pixbuf)
These commits were made to resolve a gdk-pixbuf loaders clash that could occur
in a profile when both gdk-pixbuf and gdk-pixbuf+svg were propagated.
Unfortunately, librsvg really needs to propagate gdk-pixbuf itself, which
would introduce a cycle. Packages such as 'efl' were broken as a result of
not propagating gdk-pixbuf.
A more definitive fix will be proposed for core-updates, which includes a
profile hook to generate the gdk-pixbuf loaders.cache file, as well as a build
phase and search path specification, that should resolve the issue for good.
The 'bluez5' configuration options is removed to ease transformation of the
package by users. It is auto-enabled by Meson based on the inputs available,
so it's still enabled in this package.
* gnu/packages/pulseaudio.scm (pulseaudio): Update to 15.0.
[arguments]: Fix the 'oss-output' value and remove the 'bluez5' option.
[inputs]: Add 'elogind'.
Signed-off-by: Maxim Cournoyer <maxim.cournoyer@gmail.com>
This resolves a test failure on armhf-linux (when building on a aarch64-linux
system).
* gnu/packages/linux.scm (e2fsprogs): Update to 1.46.4. Remove trailing #t.
