Before, native glibc and dynamic linker detection attempted to use the
executable's own binary if it was dynamically linked to answer both the
C ABI question and the dynamic linker question. However, this could be
problematic on a system that uses a RUNPATH for the compiler binary,
locking it to an older glibc version, while system binaries such as
/usr/bin/env use a newer glibc version. The problem is that libc.so.6
glibc version will match that of the system while the dynamic linker
will match that of the compiler binary. Executables with these versions
mismatching will fail to run.
Therefore, this commit changes the logic to be the same regardless of
whether the compiler binary is dynamically or statically linked. It
inspects `/usr/bin/env` as an ELF file to find the answer to these
questions, or if there is a shebang line, then it chases the referenced
file recursively. If that does not provide the answer, then the function
falls back to defaults.
This commit also solves a TODO to remove an Allocator parameter to the
detect() function.
Previously, this code would fail to detect glibc version because it
relied on libc.so.6 being a symlink which revealed the answer. On modern
distros, this is no longer the case.
This new strategy finds the path to libc.so.6 from /usr/bin/env, then
inspects the .dynstr section of libc.so.6, looking for symbols that
start with "GLIBC_2.". It then parses those as semantic versions and
takes the maximum value as the system-native glibc version.
closes#6469
see #11137closes#12567
This adds the following for passthrough to lld:
- `--print-gc-sections`
- `--print-icf-sections`
- `--print-map`
I am not adding these to the cache manifest, since it does not change
the produced artifacts.
Tested with an example from #11398: it successfully prints the resulting
map and the GC'd sections.
If there are zerofill sections, the loader may copy the contents of
the physical space in file directly into memory and attach that to
the zerofill section. This is a performance optimisation in the loader
but requires us, the linker, to properly zero-out any space between
__DATA and __LINKEDIT segments in file. This is of course completely
skipped if there are no zerofill sections present.
This is a temporary workaround to an unclear platform-dependence
behavior we have in libstd for `std.fs.File` abstraction. See
https://github.com/ziglang/zig/issues/12783 for more information.
Previously if a decl failed its capture scope would be deallocated and
set to undefined which would then lead to invalid dereference in
`zirClosureGet`. To avoid this set the capture scope to a special
failed state and fail the current decl with dependency failure if
the failed state is encountered in `zirClosureGet`.
Closes#12433Closes#12530Closes#12593
This relieves register pressure, and reduce generated code size
(since now we can use the same index register for both `mov_scale_src`
and `mov_scale_dst` MIR instructions).
Fix lowering of ModRM + SIB encodings where index register is extended
- previously, we would carelessly ignore the fact generating incorrect
encodings.
As far as I can see, unlike with MachO, we don't have any stubs
helper routines available and need to load a bound pointer into
a register to then call it.
