Previously we relied on the presence of a `python3` binary in the
PATH that has the correct minor version to build the port.
We now first check for the presence of a `python3.minor` binary
in the PATH and use that if found.
This allows users that have multiple Python versions installed
simultaneously (like from a PPA) to build the port without having
to change their main version.
While at it, pass the `--3way` option to give the user something to work
with. Otherwise, Git will just prepare the commit metadata and nothing
else. Note that this has a relatively low chance of working, since the
recorded blob hashes aren't known to a freshly imported repository.
Use single quotes for non-interpolated strings in the 'configopts' and
'depends' sections.
Let each dependency of the 'depends' section be on a separate line.
This is likely because we usually used the host's version of the file.
I happen to use a newer libgpg-error (which is relevant because that
particular executable has been renamed, so my system no longer contains
a file of the name gpg-error-config), which revealed this latent bug.
Normally this is supposed to be installed from gdb or gcc. If a port
wants to link against libbfd though, we need to make sure libiberty is
actually available in the root filesytem without requiring the port to
depend on those larger packages.
We need a `wxallowed` bind mount for the `ioquake3` binary to be able to
compile and run its executable scripts in memory. Instead of instructing
the user how to do so in `/etc/fstab`, we can now use the fancy
`/etc/fstab.d` facility :^)
The ports `libvorbis`, `readline` and `timidity` would not install on
macOS as a result of using `sed -i` without an extension provided. GNU
sed is available through Homebrew, but it does not replace `sed` by
default.
Instead, provide a new `sed_in_place` function that calls `sed` with the
right arguments.
Sets `$LD` to `$HOST_LD` in `.port_include.sh` if it is defined as well,
else it will be set to `ld`.
Makes libiconv build with the Clang toolchain.
This also impacts other ports when building them with the Clang
toolchain, and might result in more ports building correctly.
The ScummVM icon repository no longer accepts "1970-01-01" as a valid
start date for the icon pack generation. We now use the oldest commit
date in the repository which _is_ accepted.
If a developer built cmake (or we built it for them) from source, make
sure that port builds can find that version of CMake that has the
SerenityOS platform files included.
:yakkie:
The build process for the Zig compiler is more involved than most of
the other ports, because the Zig compiler is mostly self-hosting. In
order to build it, the zig-bootstrap build system is used, which does
the following:
1) Build LLVM for the host OS;
2) Build Zig for the host OS with the SerenityOS target enabled;
3) Build zlib, zstd and LLVM for SerenityOS using `zig cc` as the C/C++
compiler;
4) Build Zig for SerenityOS using the host Zig.
A few hacks are required in order to tell `zig cc` and zig about what
Serenity's libc looks like in the build process, but other than that
it's fairly straightforward. All of the patches that are included with
this commit are Zig-upstream ready once the LLVM patches are upstreamed.
When using cmake --build, CMake will look for this environment variable
to enable parallelism. The Zig port, for example, uses cmake --build,
and will otherwise use a single core if cmake selects Make as the build
system. This should help with all ports which use cmake --build.
We have a new, improved string type coming up in AK (OOM aware, no null
state), and while it's going to use UTF-8, the name UTF8String is a
mouthful - so let's free up the String name by renaming the existing
class.
Making the old one have an annoying name will hopefully also help with
quick adoption :^)
This makes the port install drop the CMake install files into the
sysroot, which is friendlier to macOS users. Homebrew CMake really
likes to pick homebrew zstd, even for cross-builds.
When building a port with `useconfigure="false"`, the `do_configure`
function invokes a `buildstep` with multiple positional arguments as the
command to execute.
It then tests whether the positional arguments evaluate to an emtpy
string, but could fail when multiple positional arguments were provided.
This resulted in the following warning when building the Composer port,
for example:
../.port_include.sh: line 16: [: echo: binary operator expected
Prevent this warning by testing against the number of positional
arguments, instead.
Since ac40197047 LibCrypt has its own library. PHP tries to detect the
`crypt_data` struct but fails to do so since it cannot find `crypt.h`.
By adding this explicit include path, PHP builds again.
These were no longer being picked up after some recent changes. Since
port builds happen in subshells nowadays, we can get rid of the export /
unset combo anyway.
This fixes ScummVM crashing on startup, caused by `-fvisibility` not
being set.
This now requires `--host` and `--with-build-python` to be passed to the
configure script when cross compiling; the former we simply do like in
many other package.sh scripts as well, the latter we point to `python3`,
which is expected to match the port's version anyway.
previously every buildstep would return a success error code. As a
result, all the steps would run even if previous steps failed.
I've also added a red status message when this happens.
This patch fixes libopenal's build by disabling the alsoft-config
utility which required qt6, and disabling the example programs that were
causing compiler errors.
This also forces CMake to build with an SDL2 backend so audio can
actually be played.
We used to do whole bunch of unnecessary things in the install sequence
which the default port_include script sequence can do just fine,
therefore the install sequence is removed from the port script.
The post_install sequence wrongly called "make install-bin" which could
be done in the default install sequence, as well as to create the /bin
directory which is completely unnecessary to do because the image build
script already does that for us. Also, now /usr/local/bin is in the PATH
environment variable, so the installed binaries are runnable without
creating symlinks in the /bin directory, therefore making the sequence
of post_install completely unnecessary in the script so it is removed
too.
Replace the manual creation of symlinks with the
`--enable-compat-symlinks` configure option, which automatically creates
the symlinks that we want and more.
This is a preparation to check if our users find noticeable bugs in the
x86-64 target, before we can decide if we want to remove the i686 target
for good.
Per the release notes for 1.5.0, the CMake build is preferred going
forward. This lets us drop some Makefile patches and pass them as CMake
options instead, with the exception of disabling mold-wrapper.so.
While this loses quite a bit of accuracy (although to no apparent
decrease in emulation quality) , it helps avoiding the additional
overhead of the `clock_gettime` syscall (as `CLOCK_MONOTONIC_COARSE`
is forwarded using the mapped time page) and we don't have to do a
HPET timer read for each tick.
This results in a decrease of Serenity boot time from 1h16m down to
42m when running on Serenity.
We were previously comparing the hash against the hash after the initial
import, which caused us to regenerate patches every time as long as we
did have patches (even if they haven't changed at all) and the script
entirely missing that it should remove patches if the current commit is
the "import" commit.
I've lost more changes to "you forgot to push the changed commits to the
remote" than I'd like to admit, so let's just unify both and only ever
use the actual working repository for detemining whether any patches
have changed.
This stops us from mangling our LLVM patch names and titles when using
`./package.sh dev`, as they like to put their category names in square
brackets.
Those have been added to the Clang build script in
d4d6f2d945, and now that the Clang version
has been updated, everyone should have those symlinks already anyways.
There were only two packages making use of this functionality, but it
seems more sensible to have a fixed implementation for the cleaning of
ports - especially now they delete the entire build directory and/or
dist files.
The functionality for `./package.sh clean` was a bit weird: based on
whether you were working in dev mode, it would try to delete either
`$workdir` or `$nongit_workdir` and `*.out` from your `pwd`.
The new functionality is pretty clear: `./package.sh clean` deletes the
entire build directory for the port regardless of what mode you're in,
`./package.sh clean_dist` removes all `$files`, and
`./package.sh clean_all` does both.
Instead of overwriting the existing `-j` makeopt, we only append
options. This brings the build time for the Quake port down from 24.3s
to 4.4s on my machine.
Music (MIDI) works through TiMidity++ and OpenGL support works, but is
still very buggy. Sometimes PrBoom+ fails to load its own configuration
file, which can be worked around by deleting it.
Also add it as a dependency for the `SDL_mixer` and `SDL2_mixer` ports.
Eawpats GUS patches are installed as part of the port, because without
patches TiMidity++ is not able to generate any sound. The license for
these is "fameware", credit to Andrew Suffield:
https://lists.debian.org/debian-legal/2002/09/msg00137.html
Add two patches to allow Python's package manager to work on Serenity:
- The first one enables zlib module, which is needed for `ensurepip`
command;
- The second patch fixes pip downloads, so it's possible to install
packages from the PyPI repository.
SDL2 applications are prone to reinitializing the SDL components when,
for example, changing the display resolution. This would lead to
crashes since we were doing things in the wrong order.
* We now decouple `GLContext` from `SDL_Window` a bit more, allowing
the window to be destroyed before the GLContext.
* Do not quit the `GUI::Application`, let our event loop handle exiting
These changes allow changing the display resolution in the Quake3 port.
On Arch Linux, the build was picking up the system libsamplerate, which
is undesirable. Unlikely that it is needed on Serenity, so disabling it
is a good workaround.
The GLib repository contains several submodules which is part of the
build process, but the source code for these submodule is not part of
the Source code tarball generated by Gitlab. Switching to download the
Release tarball from download.gnome.org solves this issue. It first
became apparent in version 2.73.x.
Additional information at: https://gitlab.gnome.org/GNOME/glib/-/issues/2716
These libtool archives incorrectly pointed to `/usr/local` causing
other ports not to be able to find the libvorbis libraries.
We cannot use `configure --prefix=...` since that will add up with our
fixed `make DESTDIR=...` logic, causing the path to be duplicated.
We can also not change that `DESTDIR` logic without influencing all
other port builds.
Finally, `configure --with-sysroot=...` doesn't work since not all
other ports (such as SDL_mixer) have a recent enough libtool to
understand the sysroot syntax.
So let's `sed` this and be done with it :^)
bzip2's Makefile uses the '=' operator to set these variables so they
cannot be overridden by just the environment variables; we have to pass
them on the command line.
This change ensures that the system ar/ranlib are no longer used, so the
port can be build on macOS or non-x86 Linux.
This was as simple as copying over the libtool patches from libvorbis
and removing now-unneeded Serenity-awareness patches.
Co-authored-by: Tim Schumacher <timschumi@gmx.de>
