This doesn't seem to have ever worked on any BE platform. ppc64 BE fails, ppc64 LE succeeds. Further, tested on ppc64 + ppc and got the exact same results, indicating it's not a bitness problem. sparc is even more hopeless with over 60% of the test suite failing. 92% tests passed, 74 tests failed out of 896 Total Test time (real) = 28.53 sec The following tests FAILED: 1 - test_plugin_test_ndlz (Failed) 2 - test_plugin_test_zfp_acc_float (Failed) 3 - test_plugin_test_zfp_prec_float (Failed) 4 - test_plugin_test_zfp_rate_float (Failed) 5 - test_plugin_test_zfp_rate_getitem (Failed) 6 - test_plugin_test_ndcell (Failed) 7 - test_plugin_ndmean_repart (Failed) 8 - test_plugin_ndmean_mean (Failed) 9 - test_plugin_bytedelta (Failed) 10 - test_plugin_int_trunc (Failed) 11 - test_b2nd_append (SEGFAULT) 12 - test_b2nd_copy (Failed) 14 - test_b2nd_delete (SEGFAULT) 15 - test_b2nd_full (SEGFAULT) 16 - test_b2nd_get_slice (Failed) 17 - test_b2nd_get_slice_buffer (Failed) 18 - test_b2nd_insert (SEGFAULT) 19 - test_b2nd_metalayers (Failed) 20 - test_b2nd_open_offset (Failed) 21 - test_b2nd_persistency (SEGFAULT) 22 - test_b2nd_resize (SEGFAULT) 23 - test_b2nd_roundtrip (Failed) 24 - test_b2nd_save (Failed) 25 - test_b2nd_serialize (Failed) 26 - test_b2nd_set_slice_buffer (SEGFAULT) 27 - test_b2nd_squeeze (SEGFAULT) 28 - test_b2nd_squeeze_index (SEGFAULT) 29 - test_b2nd_uninit (SEGFAULT) 30 - test_b2nd_zeros (SEGFAULT) 31 - generate_inputs_corpus (SEGFAULT) 304 - test_copy (SEGFAULT) 305 - test_delete_chunk (Failed) 308 - test_dict_schunk (Failed) 310 - test_fill_special (SEGFAULT) 312 - test_frame (Failed) 314 - test_frame_offset (SEGFAULT) 315 - test_get_slice_buffer (Failed) 717 - test_insert_chunk (Failed) 718 - test_lazychunk (Failed) 727 - test_reorder_offsets (Failed) 729 - test_schunk_frame (SEGFAULT) 730 - test_schunk_header (SEGFAULT) 731 - test_set_slice_buffer (Failed) 732 - test_sframe (Failed) 733 - test_sframe_lazychunk (Failed) 867 - test_small_chunks (Failed) 868 - test_udio (SEGFAULT) 869 - test_update_chunk (Failed) 870 - test_urcodecs (SEGFAULT) 872 - test_zero_runlen (Failed) 873 - test_compat_blosc-1.11.1-blosclz.cdata (Failed) 874 - test_compat_blosc-1.11.1-lz4.cdata (Failed) 875 - test_compat_blosc-1.11.1-lz4hc.cdata (Failed) 876 - test_compat_blosc-1.11.1-zlib.cdata (Failed) 877 - test_compat_blosc-1.11.1-zstd.cdata (Failed) 878 - test_compat_blosc-1.14.0-blosclz.cdata (Failed) 879 - test_compat_blosc-1.14.0-lz4.cdata (Failed) 880 - test_compat_blosc-1.14.0-lz4hc.cdata (Failed) 881 - test_compat_blosc-1.14.0-zlib.cdata (Failed) 882 - test_compat_blosc-1.14.0-zstd.cdata (Failed) 883 - test_compat_blosc-1.17.1-lz4-bitshuffle4-memcpy.cdata (Failed) 884 - test_compat_blosc-1.17.1-lz4-bitshuffle8-nomemcpy.cdata (Failed) 885 - test_compat_blosc-1.18.0-lz4-bitshuffle4-memcpy.cdata (Failed) 886 - test_compat_blosc-1.18.0-lz4-bitshuffle8-nomemcpy.cdata (Failed) 887 - test_compat_blosc-1.3.0-blosclz.cdata (Failed) 888 - test_compat_blosc-1.3.0-lz4.cdata (Failed) 889 - test_compat_blosc-1.3.0-lz4hc.cdata (Failed) 890 - test_compat_blosc-1.3.0-zlib.cdata (Failed) 891 - test_compat_blosc-1.7.0-blosclz.cdata (Failed) 892 - test_compat_blosc-1.7.0-lz4.cdata (Failed) 893 - test_compat_blosc-1.7.0-lz4hc.cdata (Failed) 894 - test_compat_blosc-1.7.0-zlib.cdata (Failed) 895 - test_compat_blosc-2.0.0-lz4-bitshuffle4-memcpy.cdata (Failed) 896 - test_compat_blosc-2.0.0-lz4-bitshuffle8-nomemcpy.cdata (Failed)
Created attachment 894610 [details] build.log and emerge --info (ppc)
Created attachment 894611 [details] build.log and emerge --info (ppc64)
Created attachment 894612 [details] build.log and emerge --info (sparc)
It's complicated and related to the real fix for bug 923423.
If the depgraph can be easily kept satisfied with masking on BE profiles, we should do that.
(In reply to Sam James from comment #5) > If the depgraph can be easily kept satisfied with masking on BE profiles, we > should do that. The bus error stuff on sparc is, but ppc64/ppc has no such restriction, so isn't the BE stuff orthogonal?
(In reply to matoro from comment #6) > (In reply to Sam James from comment #5) > > If the depgraph can be easily kept satisfied with masking on BE profiles, we > > should do that. > > The bus error stuff on sparc is, but ppc64/ppc has no such restriction, so > isn't the BE stuff orthogonal? No, because the same code needs fixing. It does unaligned access and type punning hacks.
There are known endian problems with multiple revdeps, clearing this out would get rid of a bunch of bugs. Casualties would be: dev-python/tables - bug 922235 dev-python/pandas - bug 916502 dev-python/xarray - not strictly a BE problem but has troubles on 32-bit and ppc, bug 890549, bug 927711 dev-python/pyarrow - could be saved with restricting tests, but this has a uname whitelist, bug 923968