Hi,

bad.pyd has a strange section .voltbl at the beginning. good.pyd looks as
expected.
(pyd-files are DLLs as used by CPython)

Carl

Am Sa., 22. Jan. 2022 um 15:11 Uhr schrieb Matthew Brett <
[email protected]>:

> Hi,
>
> On Sat, Jan 22, 2022 at 12:40 PM Martin Storsjö <[email protected]> wrote:
> >
> > Hi,
> >
> > On Sat, 22 Jan 2022, Matthew Brett wrote:
> >
> > > The Scipy build links against a static library `npymath.lib`, until
> > > recently built with VS2017 v141 toolset.
> >
> > Mixing static libraries or object files between mingw and msvc is not
> > supported, and not expected to work, in general. You might have been
> lucky
> > and your library might have been simple enough not to hit any problematic
> > case though.
> >
> > Or are the python pyd files linked with msvc tools - what parts are based
> > on mingw in this constellation?
>
> We link the pyd files with mingw-w64.    Specifically:
>
> Numpy build process:
>
> * Compiling npymath obj files from .c files - MSVC
> * Assembling obj files into npymath.lib file - MSVC
>
> Scipy build process:
>
> * Linking npymath.lib file to pyd file - mingw-w64
>
> I guess we were lucky, with the v141 build of npymath.lib, and less
> so, for the v142 build.
>
> > > This build worked fine, until recently when the build for
> > > `npymath.lib` switched to using the VS2019 v142 toolset.   Our build
> > > now runs to completion without error, but the generated dynamic link
> > > libraries, _that link to the npymath.lib file_ all generate errors on
> > > loading, of form:
> > >
> > >
> C:\repos\scipy\installdir\Lib\site-packages\scipy\spatial\_distance_wrap.cp39-win_amd64.pyd
> > >   Error 193:
> C:\repos\scipy\installdir\Lib\site-packages\scipy\spatial\_distance_wrap.cp39-win_amd64.pyd
> > > is not a valid Win32 application.
> > >
> > > This is ERROR_BAD_EXE_FORMAT (193, 0xC1)
> >
> > So while this isn't a setup that is supported, this error doesn't sound
> > like one of those that you'd expect to run into if mixing in msvc static
> > libraries in a mingw environment. It sounds like some odd mixup of
> > architectures or other details.
> >
> > I don't have any other clues offhand, but you'd have to sit down and
> > inspect the generated dll (pyd) with tools like objdump, llvm-readobj
> etc,
> > to pinpoint what's wrong in it.
>
> Do you have any hints where we should be looking in the objdump etc
> output?   This repo has the pyd files with their respective:
>
> objdump -x
> llvm-readobj --all
> dlldiag trace
>
> outputs.
>
> https://github.com/matthew-brett/dll_investigation
>
> Error we see when importing the pyd file in Python replicated at:
>
>
> https://github.com/matthew-brett/dll_investigation/blob/main/dlldiag_bad.txt#L47
>
> in case there was anything very obvious to a quick glance.
>
> Cheers,
>
> Matthew
>
>
> _______________________________________________
> Mingw-w64-public mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
>

_______________________________________________
Mingw-w64-public mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mingw-w64-public

Reply via email to