[Numpy-discussion] numpy.distutils, windows dll vs lib

Robert Kern robert.kern@gmail....
Thu Mar 15 23:40:36 CDT 2007

David Cournapeau wrote:
> Hi,
>     recently, I got some problems detecting a dynamic library (dll) with 
> numpy.distutils. Basically, I have a package which uses a class derived 
> from system_info from numpy.distutils to detect a dll I use through ctypes.
>     If only the dll is present, my numpy.distutils.system_info derived 
> class does not find the library; if the .lib is present too, then it is 
> detected. Why is that ? Can I modify my class do detecting the dll is 
> enough ? I don't know how windows dynamic linking and dynamically loaded 
> libraries work, and I am kind of confused by this (I thought .lib was 
> .a, and .dll was .so, and that symbols were not exported by default on 
> windows contrary to Unix, but the difference seems more subtle than 
> this....).

Generally, you don't use .dll files to link against. With MSVC, you need a .lib
file corresponding to your target .dll file which has the symbols. mingw and
cygwin use similar .a files. This information is coded in the
system_info.library_extensions() method, which you can override if you need
something different. Of course, since you don't mention what methods on
system_info that you are using, I can't quite be sure this will satisfy your needs.

Note that so_ext here is obtained from the distutils configuration for your
platform and is the file extension for extension modules, thus '.pyd' on
Windows, not '.dll'.

    def library_extensions(self):
        static_exts = ['.a']
        if sys.platform == 'win32':
            static_exts.append('.lib')  # .lib is used by MSVC
        if self.search_static_first:
            exts = static_exts + [so_ext]
            exts = [so_ext] + static_exts
        if sys.platform == 'cygwin':
        if sys.platform == 'darwin':
        return exts

Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
 that is made terrible by our own mad attempt to interpret it as though it had
 an underlying truth."
  -- Umberto Eco

More information about the Numpy-discussion mailing list