aboutsummaryrefslogtreecommitdiffstats
path: root/testing/rust/musl-fix-linux_musl_base.patch
diff options
context:
space:
mode:
Diffstat (limited to 'testing/rust/musl-fix-linux_musl_base.patch')
-rw-r--r--testing/rust/musl-fix-linux_musl_base.patch63
1 files changed, 63 insertions, 0 deletions
diff --git a/testing/rust/musl-fix-linux_musl_base.patch b/testing/rust/musl-fix-linux_musl_base.patch
new file mode 100644
index 0000000000..38194290cb
--- /dev/null
+++ b/testing/rust/musl-fix-linux_musl_base.patch
@@ -0,0 +1,63 @@
+From: Jakub Jirutka <jakub@jirutka.cz>
+Date: Sat, 08 Aug 2016 15:06:00 +0200
+Subject: [PATCH] Fix linux_musl_base for native musl host
+
+See https://github.com/rust-lang/rust/pull/40113
+
+--- a/src/librustc_back/target/linux_musl_base.rs
++++ b/src/librustc_back/target/linux_musl_base.rs
+@@ -13,54 +13,14 @@
+ pub fn opts() -> TargetOptions {
+ let mut base = super::linux_base::opts();
+
+- // Make sure that the linker/gcc really don't pull in anything, including
+- // default objects, libs, etc.
+- base.pre_link_args.push("-nostdlib".to_string());
+-
+ // At least when this was tested, the linker would not add the
+ // `GNU_EH_FRAME` program header to executables generated, which is required
+ // when unwinding to locate the unwinding information. I'm not sure why this
+ // argument is *not* necessary for normal builds, but it can't hurt!
+ base.pre_link_args.push("-Wl,--eh-frame-hdr".to_string());
+
+- // There's a whole bunch of circular dependencies when dealing with MUSL
+- // unfortunately. To put this in perspective libc is statically linked to
+- // liblibc and libunwind is statically linked to libstd:
+- //
+- // * libcore depends on `fmod` which is in libc (transitively in liblibc).
+- // liblibc, however, depends on libcore.
+- // * compiler-rt has personality symbols that depend on libunwind, but
+- // libunwind is in libstd which depends on compiler-rt.
+- //
+- // Recall that linkers discard libraries and object files as much as
+- // possible, and with all the static linking and archives flying around with
+- // MUSL the linker is super aggressively stripping out objects. For example
+- // the first case has fmod stripped from liblibc (it's in its own object
+- // file) so it's not there when libcore needs it. In the second example all
+- // the unused symbols from libunwind are stripped (each is in its own object
+- // file in libstd) before we end up linking compiler-rt which depends on
+- // those symbols.
+- //
+- // To deal with these circular dependencies we just force the compiler to
+- // link everything as a group, not stripping anything out until everything
+- // is processed. The linker will still perform a pass to strip out object
+- // files but it won't do so until all objects/archives have been processed.
+- base.pre_link_args.push("-Wl,-(".to_string());
+- base.post_link_args.push("-Wl,-)".to_string());
+-
+- // When generating a statically linked executable there's generally some
+- // small setup needed which is listed in these files. These are provided by
+- // a musl toolchain and are linked by default by the `musl-gcc` script. Note
+- // that `gcc` also does this by default, it just uses some different files.
+- //
+- // Each target directory for musl has these object files included in it so
+- // they'll be included from there.
+- base.pre_link_objects_exe.push("crt1.o".to_string());
+- base.pre_link_objects_exe.push("crti.o".to_string());
+- base.post_link_objects.push("crtn.o".to_string());
+-
+ // Except for on MIPS, these targets statically link libc by default.
+ base.crt_static_default = true;
+
+ base
+ }