| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
| |
So far we only matched the file name of the dependency but not its
architecture, so if for example there is one shared object for
i686-linux and another one with the same name but for x86_64-linux,
chances are that the wrong architecture is chosen.
Now we're checking the architecture of the shared object file and only
pick it, if it matches the architecture of the file to patchelf.
Signed-off-by: aszlig <aszlig@nix.build>
|
|
|
|
|
|
|
| |
This makes sure that ldd doesn't print warnings like if the file isn't
executable and also makes the status messages a bit more clear.
Signed-off-by: aszlig <aszlig@nix.build>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The semantics of adding env hooks has changed in
NixOS/nixpkgs@7f3ca3e21a22fd3101b40cadb86899542dec2e35 so that it's now
broken up into several hook variables (envBuildBuildHook,
envBuildHostHook, envBuildTargetHook, envHostHostHook, envHostTargetHook
and envTargetTargetHook).
Fortunately commit NixOS/nixpkgs@a036473a0a0c6100fce316e1444fc33ec6674b
adds a helper function (addEnvHooks) to deal this without referencing
those hook variables directly and rather just providing an offset.
Signed-off-by: aszlig <aszlig@nix.build>
|
|
|
|
|
|
|
| |
This is not only useful for packaging games, so let's make it available
from the vuizvui scope, so we can use it from other packages as well.
Signed-off-by: aszlig <aszlig@nix.build>
|
|
|
|
|
|
|
|
| |
If the saveDir isn't set the game only gets access to all the saveDirs
from other Unity games, so missing that attribute shouldn't be a deal
breaker.
Signed-off-by: aszlig <aszlig@nix.build>
|
|
|
|
|
|
|
|
| |
This is used for sandboxing and instead of letting all Unity games
access the data of every other Unity game we now restrict it to its own
save directory only.
Signed-off-by: aszlig <aszlig@nix.build>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We now distinguish between paths that have to exist and paths that are
fine to skip during bind mounting.
So far we had hard failures whenever a path that needed to be mounted
didn't exist, for example something like $XDG_CONFIG_HOME/unity3d failed
whenever the directory didn't exist.
Apart from that we now have a more clean attribute structure for sandbox
parameters, which are now:
* paths.required: Created prior to bind-mounting
* paths.wanted: Skipped if it doesn't exist
* paths.runtimeVars: Extracted from PATH-like environment variables
Signed-off-by: aszlig <aszlig@nix.build>
|
|
|
|
|
|
|
|
| |
We already mount /etc, but it isn't quite enough, because it contains
files symlinked to /etc/static, which in turn is a store path so we need
to mount the closure of that path as well.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
| |
First of all this is to bring down the amount of syscalls we're doing
but it's also useful to avoid errors when we try to mount a path over an
already mounted path.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
| |
This is using buildSandbox with the addition of adding LD_LIBRARY_PATH
to the default runtimePathVars. I've also renamed the attribute to be
called runtimePathVars instead of runtimePathVariables, simply because
it's shorter.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
On NixOS the LD_LIBRARY_PATH looks similar to this (depending on the
configuration):
/run/opengl-driver/lib:/run/opengl-driver-32/lib
However, we don't have these paths available within the sandbox, because
so far we've only used exportReferencesGraph to gather the runtime
dependencies after the build has succeeded.
This obviously doesn't take into account runtime dependencies from the
system itself.
We are now taking care of this by using the Nix store library to query
the requisities of all the paths that are contained inside path-like
variables (multiple paths delimited by colons) and mount them during
sandbox setup.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
| |
This leaves sandbox.c with only the main() function and nothing else, so
that whenever we have a lot of binaries to generate, the compilation
time should be much lower now.
The change doesn't change anything in functionality.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
| |
We're now using a makefile for building the sandbox and use pkg-config
to pass in the flags we need for compiling against lib(nix)store.
Right now the sandbox itself doesn't do anything different because we're
not actually using the (incomplete) code for querying the store.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
|
| |
These paths are things such as /etc and /run but also the .Xauthority
file, which contains the cookie to connect to the X server.
What is still missing is access to the DRI libraries, which is a bit
trickier, because we need to add those store paths at runtime and we
need to also mount all of the dependencies.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
I've already pointed out in the previous commit that using /tmp for the
root directory isn't a very good idea, mainly because we can't access
sockets from /tmp (eg. the X server sockets).
So what we're now doing is using the store path that contains the
sandbox wrappers, because that very path won't be mounted into the
sandbox anyway, so we get a free directory just as an entry point.
This has the main advantage that we don't need to create any temporary
directories which we later need to clean up nor do we need to assume
that some paths might exist in the system. For example if we'd use /usr
we still have /usr/bin/env on NixOS, but if that's going to go away in
the future or we are on a distro that doesn't have it at all, the
sandbox setup will fail.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
So far creating the sandbox has been a setup hook, however it's a bit
ugly how we gathered the needed paths for the chroot file system by
recursively searching for store paths.
While I'd like to have the sandbox being built within the main
derivation, it really isn't very practical when the build takes longer
than 10 minutes.
With this implementation however the sandbox builds really fast and we
can also modify the sandbox without needing to rebuild a particular
game.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is basically to make sure various games can't write to whatever
they want in the file system, so it's not a complete sandboxing
solution.
Currently there's a drawback in that we can't easily determine the
runtime dependencies while building a particular game, so we need to
recursively dig through all referenced store paths to look them up.
A better solution for this would be to gather the build time reference
graph prior to building so that we can limit searching for these
references within only the actual build outputs instead of churning
through all inputs.
In addition to that, we currently mount the namespaced root file system
on top of /tmp, which makes the real /tmp unavailable to us. While in
theory this shouldn't be a problem, it actually turns out it is indeed a
problem if the application wants to connect to the X server socket,
which is at something like /tmp/.X11-unix/X0 for display :0.
Apart from these drawbacks we have a working solution for simple
applications (not games, because they usually require X), which now get
its own chroot with only the paths accessible that are strictly
necessary.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
| |
Most older games don't cope very well with PulseAudio, so let's add ALSA
so that a fallback is available. This is also very useful if
withPulseAudio is set to false, because this would actually mean "no
audio at all".
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
|
| |
A lot of games come as a Zip archive, so let's add it to
nativeBuildInputs by default so we no longer need to deal with that for
every single game.
This also removes that line from Invisigun Heroes, so the package now is
pretty much minimal and very clean :-)
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
| |
A lot of games (especially Unity games) come as a tarbomb, so in order
to avoid having those files all over the place we now create an
additional directory before unpack and set the sourceRoot to that if
it's a tarbomb and if that's not the case set it to the single file
within that directory.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
|
| |
Now we no longer need to invoke patchelf by ourselves and all of the
dependencies are patched automatically.
I've added gtk2-x11, gdk_pixbuf and glib to buildUnity as well, because
after inspecting other Unity games they seem to require the same
libraries as well.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
| |
This allows us to add libraries to the RPATH despite being required by
the respective game. By default there is only PulseAudio at the moment.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The main functionality for this function is to gather missing
dependencies in ELF executables and shared libraries and using patchelf
to set the right RPATH.
All of the dependencies are searched based on what we have in one of the
buildInputs variables, so all we need to do is list them in there.
One thing that's still left to solve is adding libraries to the RPATH
which are only required at runtime. An example for this would be the
pulseaudio library.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
|
|
|
|
| |
This automatically creates a desktop entry based on the given fullName
and the standard path for the icon (Resources/UnityPlayer.png).
Well, at least I hope that the icon path is standard, but we'll see once
we use buildUnity for more games.
Tested using Plasma 5 for Invisigun Heroes.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
|
|
|
|
|
|
| |
The executable of Unity games at the moment is unable to find its own
path if it can't do a stat() on argv[0]. So wrapping the program helps
here for now.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|
|
There are a lot of Unity games out there, so it's a good idea to have a
generic builder just for Unity, which we then can apply to a lot more
games.
Right now this is a pretty basic abstraction which I've only tested with
Invisigun Heroes.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
|