| 16 Jun 2025 |
 alexfmpe | John Ericson s1341 : non-prebuilt android builds fails early on with compiler-rt not finding pthreads.h
$ nix-build -A pkgsCross.aarch64-android.hello
...
/build/compiler-rt-src-19.1.7/compiler-rt/lib/builtins/os_version_check.c:287:10: fatal error: 'pthread.h' file not found
287 | #include <pthread.h>
| ^~~~~~~~~~~
been this way for several months, any clue what that's about?
| 00:51:09 |
| alexfmpe | feels pretty similar to https://github.com/NixOS/nixpkgs/pull/344079 | 00:51:33 |
| alexfmpe | * John Ericson s1341 : non-prebuilt android builds fails early on with compiler-rt not finding pthreads.h
$ nix-build -A pkgsCross.aarch64-android.hello
...
/build/compiler-rt-src-19.1.7/compiler-rt/lib/builtins/os_version_check.c:287:10: fatal error: 'pthread.h' file not found
287 | #include <pthread.h>
| ^~~~~~~~~~~
been this way for several months, any clue what that's about?
| 00:51:47 |
| alexfmpe | $ nix-build -A pkgsCross.mingwW64.abseil-cpp
...
/build/source/absl/base/internal/thread_identity.cc:18:10: fatal error: pthread.h: No such file or directory
18 | #include <pthread.h>
| 00:53:51 |
| alexfmpe | whereas aarch64-android-prebuilt works just fine for both hello and abseil-cpp
hmm all places consuming useAndroidPrebuilt seem to basically be doing gcc vs clang shenanigans | 00:59:49 |
| alexfmpe | * John Ericson s1341 : non-prebuilt android builds fails early on with compiler-rt not finding pthreads.h
$ nix-build -A pkgsCross.aarch64-android.hello
...
/build/compiler-rt-src-19.1.7/compiler-rt/lib/builtins/os_version_check.c:287:10: fatal error: 'pthread.h' file not found
287 | #include <pthread.h>
been this way for several months, any clue what that's about?
| 02:16:51 |
 Alyssa Ross | In reply to @genericnerdyusername:matrix.org
ig i mustve mixed up versions the bug is there in 25.05 but not 24.11 Sounds bisectable? | 04:54:54 |
 GenericNerdyUsername | I did, breaking commit is 6b9c4bbe, which updated binutils from 2.43.1 -> 2.44 | 08:35:01 |
| GenericNerdyUsername | the most relevant changelog entry is that a bunch of new extensions are supported now | 08:36:17 |
| GenericNerdyUsername | Download flake.nix | 09:11:28 |
| GenericNerdyUsername | example | 09:11:33 |
| GenericNerdyUsername | new contains c.addi, old does not | 09:12:08 |
| GenericNerdyUsername | * the most relevant changelog entry is that a bunch of new riscv extensions are supported now | 09:17:26 |
 emily | I believe our wrapper strips -march by default? | 09:38:37 |
| emily | I could be wrong | 09:38:40 |
| emily | you're really meant to specify that stuff in the Nixpkgs system structure | 09:38:58 |
| emily | using an unwrapped compiler may also be an option | 09:39:22 |
| GenericNerdyUsername | what would i have to do use the unwrapped compiler? is it as simple as replacing gcc with gccUnwrapped or is there more to it? | 16:31:36 |
| GenericNerdyUsername | also why would a binutils update mean that -march getting stripped suddenly matters? | 16:34:15 |
 K900 | -march should not be stripped unless it's =native IIRC | 16:34:30 |
| GenericNerdyUsername | how would i tell if this is a binutils regression? | 16:42:42 |
| K900 | Revert binutils and try it, I guess? | 16:43:02 |
| GenericNerdyUsername | in that case it definitely is | 16:44:01 |
| GenericNerdyUsername | wait a minute whats binutils with gold? | 16:44:52 |
| K900 | Configured to use the gold linker | 16:45:22 |
| K900 | Instead of the default bfd one | 16:45:26 |
| K900 | gold was an early attempt at writing a fast linker | 16:46:00 |
| K900 | It wasn't very fast and was never very finished | 16:46:08 |
| GenericNerdyUsername | then why did we change to the binutils-with-gold source tarball? | 16:47:14 |
| K900 | Because that just includes the sources of gold | 16:47:31 |
