Segmentation fault when importing kaldi_pybind #3836
Comments
|
please try to get a stack trace somehow. gdb or valgrind or pdb3 might
help.
…On Fri, Jan 10, 2020, 9:46 PM mohan-li ***@***.***> wrote:
Hello, I have just compiled kaldi with brand pybind11 by
configure —shared —mathlib=ATLAS —cudatk-dir=somewhere/cud’s-10.0,
with the gcc version as 5.3.1, and I installed pybind11 for python by pip.
Compiling the regular kaldi src seemed to be normal, but when I tried to
compile pybind package, the make did succeed, as all .o files were
generated normally, however when the makefile tested on kaldi_pybind in
python3 there was a segmentation fault error.
I was wondering if this is because of the mathlib version or something
else?
Please help, thanks a lot!
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#3836?email_source=notifications&email_token=AAZFLO4LTBPUCXGDB5GZKPDQ5B33PA5CNFSM4KFIAIW2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IFLFL3Q>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAZFLO4Z3E3PNVA5AISIRT3Q5B33PANCNFSM4KFIAIWQ>
.
|
Thanks for your reply, and I tried to use gdb to debug python -c 'import kaldi_pybind', the backtrace command gave as: #0 std::string::_Rep::_M_is_leaked (this=this@entry=0x956a5a94ffe9) I suspect the error arised form the conda environment as the path /home/conda/feedstock_root doesn't exist at all... |
|
It seems that the error is from OpenFST
Maybe you can exclude OpenFST from kaldi pybind and recompile it again to see if
is the source path of the people who compiled Python and it is embedded into the binary. |
Yes, I have recompliled the tools in kaldi including openfst, and it worked. Possibly it would be better for the developers to clarify the required gcc version for this branch. Presumably gcc has to be newer than 4.9, as I tried that 4.8.5 didn’t work but 5.3.1 did. Thanks a lot! |
|
This kind of error is due to an ABI mismatch.
E.g. one part of the codebase was compiled with the C++11 ABI and one with
the older ABI.
We should definitely figure out how to detect the ABI mismatch earlier and
issue an informative error message.
…On Sat, Jan 11, 2020 at 6:09 AM mohan-li ***@***.***> wrote:
Closed #3836 <#3836>.
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#3836?email_source=notifications&email_token=AAZFLO6CWDQF5ZC3CDECOM3Q5DWZVA5CNFSM4KFIAIW2YY3PNVWWK3TUL52HS4DFWZEXG43VMVCXMZLOORHG65DJMZUWGYLUNFXW5KTDN5WW2ZLOORPWSZGOV46R4LY#event-2940018223>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAZFLOYY23YE5ERVI3QX7TDQ5DWZVANCNFSM4KFIAIWQ>
.
|
Hello, I have just compiled kaldi with brand pybind11 by
configure —shared —mathlib=ATLAS —cudatk-dir=somewhere/cud’s-10.0,
with the gcc version as 5.3.1, and I installed pybind11 for python by pip.
Compiling the regular kaldi src seemed to be normal, but when I tried to compile pybind package, the make did succeed, as all .o files were generated normally, however when the makefile tested on kaldi_pybind in python3 there was a segmentation fault error.
I was wondering if this is because of the mathlib version or something else?
Please help, thanks a lot!
The text was updated successfully, but these errors were encountered: