| 22 Dec 2024 |
 jade_ | In reply to @k900:0upti.me
This is in libcxx somewhere let me look in libstdcxx since it is linux | 04:51:10 |
| jade_ | and uhhh our support in nixpkgs for building with real libcxx on linux is 🤡 for basically no reason | 04:51:43 |
| jade_ | actually this might be rust | 04:54:56 |
| jade_ | i don't know why it is in rust though | 04:55:02 |
| jade_ | UMMMM https://github.com/llvm/llvm-project/blob/8df4e60945fa9efdeab1a2fa2a5cdf22ea9c9112/lld/test/ELF/gc-sections-strip-debug.s#L1-L17 | 04:57:19 |
| jade_ | okay so this is in the llvm testsuite, what would possibly cause this to be bork? | 04:57:53 |
| jade_ | * okay so this is in the llvm testsuite, what would possibly cause this to be bork and allow llvm to release it? | 04:58:01 |
| jade_ | okay i am pretttty sure this is llvm's bug | 05:00:18 |
| jade_ | being as they are the only ones with the "'%<type>'" string in errors | 05:00:34 |
| jade_ | In reply to @k900:0upti.me
https://github.com/llvm/llvm-project/issues/24438 this is almost certainly not true since there is a popsection implementation. however, i have filed a bug against clang because that diagnostic is an atrocity: https://github.com/llvm/llvm-project/issues/120870 | 05:12:21 |
| jade_ | uhhhhhhhhhhhhh | 05:19:00 |
| jade_ | clang 19 regression? | 05:19:03 |
| jade_ | currently downloading clang 19 to try to minimal repro againstg | 05:19:16 |
| jade_ | likely it would be fixed for us if we used -no-integrated-as btw | 05:19:48 |
| jade_ | ze fuck, still works | 05:20:17 |
| jade_ | https://github.com/termux/termux-packages/blob/d65e2cdc22c3ccb9735a402d7854bf3cd4504659/packages/valgrind/build.sh#L29-L35 hmmmmmmmmmm | 05:22:08 |
| jade_ | Yureka (she/her): this looks like a smoking gun that this is an arm issue | 05:22:21 |
| jade_ | as for what caused it to blow up in your face, idk | 05:22:30 |
| jade_ | ~ » armv7l-unknown-linux-gnueabihf-cc test.c
<inline asm>:1:41: error: expected '%<type>' or "<type>"
1 | .pushsection ".debug_gdb_scripts", "MS",@progbits,1
| ^
<inline asm>:4:12: error: .popsection without corresponding .pushsection
4 | .popsection
| ^
2 errors generated.
~ » armv7l-unknown-linux-gnueabihf-cc -no-integrated-as test.c
bingo
| 05:23:23 |
| jade_ | filing a llvm bug | 05:32:42 |
| jade_ | https://github.com/llvm/llvm-project/issues/120871 alright, here's bug number two | 05:46:34 |
| jade_ | i think this should be a satisfactory amount of complaining ^^ | 05:46:48 |
| jade_ | still wondering what the actual cause is tbh | 05:47:25 |
| jade_ | like, who is putting that code into our build in the first place? | 05:47:35 |
| jade_ | aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaand the wheel spins and it lands on BOOST https://github.com/boostorg/json/blob/7f0bceb81280df758c7b4a7cacf8779351ebcc53/include/boost/json/detail/config.hpp#L236-L238 | 05:55:47 |
| jade_ | (the way i found this was by github search, not by doing anything to actually find out with strict correctness: https://github.com/search?q=%22.pushsection%22+AND+%22.debug_gdb_scripts%22&type=code) | 05:58:33 |
| jade_ | (okay it was probably boost outcome? https://github.com/boostorg/outcome/blob/a3686af224a86dfcf5f12aeb039502bae82d2a6b/include/boost/outcome/experimental/status-code/status_code.hpp#L725) | 05:59:37 |
| jade_ | complaints have been fired in hopefully enough directions for someone to care about fixing this properly; our interim solution should be -no-integrated-as on arm platforms, i think | 06:11:13 |
| jade_ | or we could patch boost in nixpkgs using this methodology that fixed it elsewhere: https://github.com/ned14/outcome/pull/308/files | 06:25:25 |
| jade_ | either way | 06:25:26 |
