pythonPackages.kaldi-active-grammar: fix build #203485
Conversation
ckiee
added
0.kind: ZHF Fixes
ckiee
changed the title
|
|
||
| # commit f07a0615ea82351bbddc0799364774cb350ff759 "openblas: 0.3.20 -> 0.3.21" | ||
| # ... breaks header compatibility with the outdated kaldi-active-grammar tree. | ||
| old-openblas = openblas.overrideAttrs (prev: rec { |
There was a problem hiding this comment.
Weird things might happen if this package were imported alongside numpy
There was a problem hiding this comment.
Action recommended? It's contained to just this derivation so I don't anticipate much fallout.
There was a problem hiding this comment.
Took a bit of digging but revealed opencv having similar issues:
* Use LAPACK_xxxx to transparently (via predefined lapack macros) deal with pre and post 3.9.1 versions.
* LAPACK 3.9.1 introduces LAPACK_FORTRAN_STRLEN_END and modifies (through preprocessing) the declarations of the following functions used in opencv
* sposv_, dposv_, spotrf_, dpotrf_, sgesdd_, dgesdd_, sgels_, dgels_
* which end up with an extra parameter.
* So we also need to preprocess the function calls in opencv coding by prefixing them with LAPACK_.
* The good news is the preprocessing works fine whatever netlib's LAPACK version.
There was a problem hiding this comment.
Adding
substituteInPlace src/matrix/cblas-wrappers.h \
--replace stptri_ LAPACK_stptri \
--replace dtptri_ LAPACK_dtptri \
--replace sgetrf_ LAPACK_sgetrf \
--replace dgetrf_ LAPACK_dgetrf \
--replace sgetri_ LAPACK_sgetri \
--replace dgetri_ LAPACK_dgetri \
--replace sgesvd_ LAPACK_sgesvd \
--replace dgesvd_ LAPACK_dgesvd \
--replace ssptri_ LAPACK_ssptri \
--replace dsptri_ LAPACK_dsptri \
--replace ssptrf_ LAPACK_ssptrf \
--replace dsptrf_ LAPACK_dsptrf
to postPatch allows it to build successfully with 0.3.21
There was a problem hiding this comment.
Same with openfst. We should not merge this in the current state.
There was a problem hiding this comment.
The openfst one isn't new and I think it's much less likely to produce a collision - numpy is everywhere on the other hand.
There was a problem hiding this comment.
Pushed @risicle's suggestion; @SuperSandro2000: KAG uses a openfst fork, so it's not the same.
(I tried typing this with the working build and it does work, but effort. Spent like two minutes figuring out I need to say "location <500ms> say <200ms> super <500ms> tab" or a longer string)
There was a problem hiding this comment.
KAG uses a openfst fork, so it's not the same.
has it the same import path? If so they are considered the same.
There was a problem hiding this comment.
I also tried it before saying that and it does not work. The fork is based on 1.7, we are on 1.8 - I am not interested in debugging this, and it's out of the scope of my changes: cleaning [openfst] up.
|
Might be worth suggesting the openblas fix to upstream. |
|
Successfully created backport PR #204282 for |
Description of changes
Fixes #203089; cc @Lykos153 @mweinelt.
Snuck
backports-zoneinfoin as it's an indirect dependency and something I ran into because my test shell hadpython38. Can drop.Things done
sandbox = trueset innix.conf? (See Nix manual)nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage./result/bin/)nixos/doc/manual/md-to-db.shto update generated release notesZHF: #199919