Created attachment 889573 [details] emerge --info during IPA pass: inline scipy/stats/_biasedurn.cpython-311-x86_64-linux-gnu.so.p/_biasedurn.cpp: In function '__pyx_f_5scipy_5stats_10_biasedurn_next_normal': scipy/stats/_biasedurn.cpython-311-x86_64-linux-gnu.so.p/_biasedurn.cpp:6910:26: internal compiler error: original not compressed with zstd 6910 | __pyx_r = random_normal(__pyx_v_5scipy_5stats_10_biasedurn__glob_rng, __pyx_v_m, __pyx_v_s); | ^ 0x1b8994f internal_error(char const*, ...) ???:0 0xa2eb84 lto_end_uncompression(lto_compression_stream*, lto_compression) ???:0 0xa28e1d lto_get_section_data(lto_file_decl_data*, lto_section_type, char const*, int, unsigned long*, bool) ???:0 0x73f48c cgraph_node::get_untransformed_body() ???:0 0xc298ab optimize_inline_calls(tree_node*) ???:0 0x958c1b inline_transform(cgraph_node*) ???:0 0xaf3f7f execute_all_ipa_transforms(bool) ???:0 0x74e733 cgraph_node::expand() ???:0 0x6b6f7a lto_main() ???:0
Created attachment 889574 [details] full build log
Error persists also without LTO.
The same error also on old i5 computer.
With new gcc-14 it does not build with -flto but builds OK without it.
It's not a regression at all. It's a weird GCC bug wrt compression format for LTO.
Pretty sure this is just the usual "I upgraded the compiler and now my static libraries are invalid"? Unless you have an unrelated bug that isn't the original "original not compressed with zstd". (In this case it's not compiler bytecode revisions being cranked I think -- it's an issue with flipping USE=zstd.)
> Pretty sure this is just the usual "I upgraded the compiler and now my static libraries are invalid"? It failed for me before the gcc upgrade
(In reply to Andrew Nowa Ammerlaan from comment #7) > > Pretty sure this is just the usual "I upgraded the compiler and now my static libraries are invalid"? > > It failed for me before the gcc upgrade It's the variation of it with 23.0 which enabled zstd by default. We can't do much about it, Eli filed a bug upstream already.
