Re: Windows cross-build of libzip 1.4.0 fails

[Thomas Klausner <tk%giga.or.at@localhost>]

Hello Benjamin!


On Sun, Jan 07, 2018 at 12:46:00AM -0800, Benjamin Gilbert wrote:
> libzip 1.4.0 has several problems cross-building for Windows from Linux
> with MinGW-w64:
> 
> 1. regress/malloc.c #includes dlfcn.h, which doesn't exist on Windows.  The
> Autotools infrastructure did not attempt to build malloc.c on Windows, but
> regress/CMakeLists.txt does not appear to contain any such exclusion.

We're not currently using this, so I've unhooked it from the build completely.

> 2. ziptool is built for the target system, but "make" attempts to run it on
> the build system to create test data.  Autotools did not do this if "make
> check" was not run.

That was fixed shortly after 1.4.0 already:
https://github.com/nih-at/libzip/commit/e9244b207c0fc883ff84c1fc46602feb1e2b0347

> 3. nonrandomopentest is once again being built, and this time failing with
> a different error:
> 
> CMakeFiles/nonrandomopentest.dir/objects.a(nonrandomopentest.c.obj): In
> function `main':
> [...]/libzip-1.4.0/regress/nonrandomopentest.c:45: undefined reference to
> `zip_random'
> 
> That makes sense, as zip_random is not declared dllexport.

And it shouldn't be exported either in the general case, but it's
needed for this test to try out if zip_random is used from the
replacement DLL, or if the original one is used.

Do you have a suggestion on how to solve this? Or should we just skip
this and the non-random AES encrytion tests on Windows?
 Thomas