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

Some non-FHS operating systems like NixOS are heavily reliant on
DT_RUNPATH.

Unlike the old DT_RPATH, DT_RUNPATH is relatively simple to implement:
if an object libdependent depends on another object libdependency,
only the DT_RUNPATH of libdependent is used to search for libdependency
(the DT_RUNPATHs of "larger" libraries or the executable are ignored).

DT_RPATH is not implemented here, because it's more difficult: we would
have to do what ld.so does, which is to build up a stack with the
object that loaded the object that loaded (etc.) the object that loaded
the one we are currently considering, and walk all the way up the stack
looking for RPATHs.

Helps: https://github.com/ValveSoftware/steam-runtime/issues/511


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>

The flag `CAPTURE_FLAG_EVEN_IF_OLDER` is not supposed to be considered
pertinent if we are checking a library dependency.

For this reason when we evaluate if a library dependency is a glibc
sub-lib, we should skip checking if `CAPTURE_FLAG_EVEN_IF_OLDER` is set.

This fixes an issue when capturing libraries with the pattern `gl:`,
while using an host system with glibc <= 2.33 and a container with glibc
>= 2.34. In this situation we were wrongly choosing `libpthread.so.0`
from the provider and `libc.so.6` from the container.

Signed-off-by: Ludovico de Nittis <ludovico.denittis@collabora.com>
Co-authored-by: Simon McVittie <smcv@collabora.com>

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

Ludovico de Nittis authored Sep 06, 2021 and Simon McVittie committed Sep 06, 2021

If a filter pattern has an '@' in it, it's safe to assume that it is
expected to be compared against the `symbol@version` fully-qualified
name.
Instead if there isn't an '@', like before, we just compare it against
only the symbol.

Signed-off-by: Ludovico de Nittis <ludovico.denittis@collabora.com>

Ludovico de Nittis authored Sep 03, 2021 and Simon McVittie committed Sep 06, 2021

Starting from glibc 2.34, libc.so.6 is a regular file instead of a
symlink. Also all the symbols from its small libraries like libpthread,
librt etc., are now moved into libc.so.6.

This leads to two issues:
1. The default comparison by name doesn't work anymore for libc.so.6
2. Counting symbols or symbol-versions for the small libraries of glibc
is not going to work.

For this reason, if we are capturing a small library of glibc, we first
try libc.so.6, using an hard-coded comparison by versions, symbols and
name. After that we will take the requested library from the same place
where we also took libc.so.6.

Signed-off-by: Ludovico de Nittis <ludovico.denittis@collabora.com>

Ludovico de Nittis authored Sep 02, 2021 and Simon McVittie committed Sep 06, 2021

Signed-off-by: Ludovico de Nittis <ludovico.denittis@collabora.com>

mallinfo() has been deprecated since glibc 2.33 and has been replaced
with mallinfo2().

Building libcapsule with glibc 2.33 fails with the following error:
```
capsule/capsule-wrappers.c: In function ‘address_within_main_heap’:
capsule/capsule-wrappers.c:323:20: error: ‘mallinfo’ is deprecated
[-Werror=deprecated-declarations]
  323 |             struct mallinfo mi = mallinfo();
      |                    ^~~~~~~~
In file included from capsule/capsule-wrappers.c:16:
/usr/include/malloc.h:118:24: note: declared here
118 | extern struct mallinfo mallinfo (void) __THROW
__MALLOC_DEPRECATED;
      |                        ^~~~~~~~
cc1: all warnings being treated as errors
```

To fix that we try to use the newer mallinfo2(), if it is available.

Signed-off-by: Ludovico de Nittis <ludovico.denittis@collabora.com>

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

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

Reference: https://reuse.software/


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

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

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

To select SONAMES that match a glob pattern, we want to use the
soname-match mode, not the soname mode.

Spotted by @LuigiPiucco on
<https://github.com/NixOS/nixpkgs/issues/100655

>.

Thanks: Luigi Sartor Piucco
Signed-off-by: Simon McVittie <smcv@collabora.com>

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

The Debian 11 toolchain produces them like this.

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

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

Clear Linux patches glibc to put ld.so.cache in /var/cache/ldconfig/
to meet their goal of /etc being initially empty and reserved for the
sysadmin. We can handle this just like the conventional path: it
doesn't need to be architecture-specific.

Partially addresses
https://github.com/ValveSoftware/steam-runtime/issues/345



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

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

The File::Glob module in Ubuntu 12.04 'precise', and therefore also
Steam Runtime 1 'scout', doesn't support named keywords like this.

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

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

I'm using Exherbo as an example of a host OS with an unusual layout that
breaks libcapsule's assumptions. Tested in an x86_64 Docker container, but
i386 is symmetrical and should also work.  This also puts the framework
in place to deal with other distributions' architecture-specific ld.so
caches.

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

This is mostly just a simplification; it also improves the error
handling a little.

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

Exherbo uses the traditional-cross-compilation-style prefix,
/usr/<tuple>, which broke the tests' assumptions.

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

This avoids worrying about which SONAME we're dealing with, and is
maybe more common in smallish containers (in particular Exherbo's
Docker container has it, which was useful for testing).

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

I'm using Exherbo as an example of a host OS with an unusual layout
that breaks libcapsule's assumptions, in order to get the framework
in place for being able to receive patches from other weirder host OSs.

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

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

The debug symbols for libcapsule-tools-relocatable would be the same as
for libcapsule-tools, making them non-co-installable. Use the environment
variable instead of --no-automatic-dbgsym for compatibility with
Steam Runtime 1 'scout', which is too old to have that option.

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

If you don't want to change the link target for all the symlinks that
will be created, instead of using `--link-target`, it is now possible to
use the new `--remap-link-prefix` and just list the directories that
should be located under a different target.

Signed-off-by: Ludovico de Nittis <ludovico.denittis@collabora.com>

This can be used when we want to test some particular cases that we
expect should not succeed.

Signed-off-by: Ludovico de Nittis <ludovico.denittis@collabora.com>

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

If we have the CAPTURE_FLAG_IF_EXISTS option flag, we should not
propagate an error if we are not able to find some dependencies of a
library. Instead we should just skip it.

This fixes an error that some users reported because they had some
leftovers unused libraries in their system and libcapsule reported an
error while trying to capture their (missing) dependencies.

Signed-off-by: Ludovico de Nittis <ludovico.denittis@collabora.com>

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

If we `make dist` with gtk-doc-tools 1.28 from Debian 10 'buster' (the
current stable release), it generates a libcapsule-docs.xml in which
generation of tree_index.sgml is mandatory. This is incompatible with
building this non-GObject library with gtk-doc-tools >= 1.30, in which
tree_index.sgml is only generated if the library contains at least one
GObject type.

Most projects treat the libcapsule-docs.xml generated by gtk-doc as
a template and commit it to git, but in this project we don't
particularly want to maintain it and would prefer to keep regenerating
it during build.

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

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

Signed-off-by: Ludovico de Nittis <ludovico.denittis@collabora.com>