Signed-off-by: Simon McVittie <smcv@collabora.com>

We have to do this in two stages, because passing ${ORIGIN} through
Autoconf, Automake and Make is a massive pain: first write in a
dummy rpath that starts with /_ORIGIN_ (implemented here), then
rewrite it to start with ${ORIGIN} with chrpath (to be done
externally).
Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

This should make it a little less confusing to work out what we're
looking at.
Signed-off-by: Simon McVittie <smcv@collabora.com>

This simplifies the situation by only needing one extra directory in
the LD_LIBRARY_PATH.
Signed-off-by: Simon McVittie <smcv@collabora.com>

If we mount the standalone GL stack at /run/host/$host_path and
make the library filenames be symlinks to there, then it's
immediately obvious from looking at the symlink which GL stack we're
using.
Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

If we capture the NVIDIA driver and all its dependencies, then we
have to use the equivalent of --libc-provider=auto, because its
dependencies include libc but ld.so has to be in lockstep with libc.
Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

In particular this is enough to capture the dependencies of the
Mesa DRI drivers:

    $ capsule-capture-libs --dest=lib \
      only-dependencies:path-match:"/usr/lib/x86_64-linux-gnu/dri/*.so"
    $ ls lib
    libbsd.so.0
    libc.so.6
    libdl.so.2
    libdrm_amdgpu.so.1    # etc.

and the drivers themselves:

    $ capsule-capture-libs --dest=mesa-drivers \
      no-dependencies:path-match:"/usr/lib/x86_64-linux-gnu/dri/*.so"
    $ ls mesa-drivers
    i915_dri.so   # etc.

in separate batches.
Signed-off-by: Simon McVittie <smcv@collabora.com>

For Mesa DRI drivers, we will need to capture the dependencies but
not the driver itself, because we load the driver itself by setting
LIBGL_DRIVERS_PATH instead.

I don't have a concrete use-case for no-dependencies:, which is
implemented for symmetry.
Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

We still do this for unadorned wildcards, but explicit is better than
implicit.
Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

We can set needed[0] to the absolute path to a loadable module, and
in fact we will need to do that when collecting the dependencies
of a Mesa DRI driver.
Signed-off-by: Simon McVittie <smcv@collabora.com>

Previously, we just used strrchr(), but that didn't actually return
the basename as in basename(1): we were comparing strings like
"/libc-2.17.so" or "/libglib-2.0.so.0.5600.0".
Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

The Steam Runtime SDK runtime currently includes the D-Bus machine ID.
This is a bug, but we can easily be more robust here.

Exclude a few other /var/lib subdirectories while I'm here.
Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

We also need to wrap realloc() to dispatch pointers to the correct
free/alloc implementation.

glibc voodoo moved to its own header to make it easy to drop once
we don't need it any more

malloc/free cluster of functions moved into their own sub-struct
pointer so they don't perturb the main capsule namespace struct
layout if/when they change (as we expect them to).

abstract out the arcane memory checks so that the new realloc
wrapper can share them with the free wrapper

malloc/calloc/realloc/posix_memalign are simply replaced by
their external counterparts.

free() gets a wrapper which route the pointer to the hopefully
correct free() implementation.

~/.local/share/Steam/ubuntu12_32/steam has a symbol with a weird
binding type part way through its symbol table:

46 [OBJECT   LOOS/GNU_UNIQUE] size: 0 (0)_ZNSs4_Rep20_S_empty_rep_storageE 0000 (nil)

but we can't just stop there, because there are symbols after it that
need relocating.

I blame C++.
Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

These are the same ones special-cased by Flatpak, although Flatpak
would generate a minimal /etc/passwd and /etc/group rather than using
the ones from the host system.
Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>

Signed-off-by: Simon McVittie <smcv@collabora.com>