🐛 Allow passing extra_libs to CheckLibWithHeader

[rdbisme]

Contributor Checklist:

• I have created a new test or updated the unit tests to cover the new/changed functionality.
• I have updated CHANGES.txt and RELEASE.txt (and read the README.rst).
• I have updated the appropriate documentation

[mwichmann]

The concept seems good... not sure why that argument was left out of the upper layer when it's present in the lower layer (Conftest.py). Could ask the same question about SConf/CheckLib.

[rdbisme]

First time I contribute to scons. Do I need to add a specific test for this?

[bdbaddog]

First time I contribute to scons. Do I need to add a specific test for this?

Yes. That's why we have the checklist in the initial decription.
You can edit it and add an X next to each item once you've added it.

[bdbaddog]

And add this change to the docs as well.
We'll help you through those if you need help.
Also asking on the discord server may be useful.

[rdbisme]

Yes. That's why we have the checklist in the initial decription.
You can edit it and add an X next to each item once you've added it.

I mean, my remark was more like that test wouldn't test much? CheckLib already has that argument. I'm just exposing it in the interface of CheckLibWithHeadere, but I'll give a look to the tests. :)

[rdbisme]

I should have added all things required. I'm not 100% sure about the doc (XML, really?! :)).

[mwichmann]

Yes, xml, really :-) (this project dates back just a few years, like 20+). We can help with docs, there's not really much to do there for this kind of change.

The configure stuff is a weird two-layer thing, SConf is the "API layer" and Conftest is the implementation layer. Except, they reach into each other and aren't as separated as they could be. Not making excuses.

[bdbaddog]

@mwichmann - what's the blurb to indicate this is added in NEXT_RELEASE?

[mwichmann]

Since the xml files are not processed by Sphinx, it needs the hand-crafted version:

<para>
<emphasis>Changed in version NEXT_RELEASE:</emphasis>
Added the <parameter>extra_libs</parameter> parameter.
</para>

It's a little unclear whether an existing method getting a new parameter is considered "added" or "changed". I'm tending to follow what CPython does, which is to consider that "changed" (though I admit that's not been consistent).

The corresponding markup in source code (aka the docstring) is:

.. versionchanged:: NEXT_RELEASE
    Added the  *extra_libs* parameter.

[mwichmann]

Still think this should be added to SCons.SConf.CheckLib as well.

[mwichmann]

If I can figure out how to do it, I'm going to push an update to this that

• also enables extra_libs for the CheckLib API,
• adds the version-added rst to the docstrings
• adds the version-added xml and the `extra_libs addition to the manpage, plus tweaks a little bit (trying to resist doing more this time!)

[mwichmann]

Hmmm, I just spotted that this isn't passing on the Windows CI:

 213/1250 (17.04%) C:\hostedtoolcache\windows\Python\3.12.8\x64\python.exe test\Configure\CheckLibWithHeader_extra_libs.py
D:\a\scons\scons\scripts\scons.py returned 2
STDOUT =========================================================================
scons: Reading SConscript files ...
scons: done reading SConscript files.
scons: Building targets ...
cl /FolibB.obj /c libB.c /nologo /IC:\Users\RUNNER~1\AppData\Local\Temp\scons\testcmd.5912.50_ghoo6\libA
libB.c
link /nologo /dll /out:libB.dll /implib:libB.lib /LIBPATH:C:\Users\RUNNER~1\AppData\Local\Temp\scons\testcmd.5912.50_ghoo6\libA A.lib libB.obj
LINK : fatal error LNK1181: cannot open input file 'A.lib'
scons: building terminated because of errors.

STDERR =========================================================================
scons: *** [libB.dll] Error 1181

Will investigate soon if someone doesn't get there first...

[mwichmann]

Need some guidance here... on Windows you have to link against the .lib version, which doesn't even build... not sure how this is supposed to work.

[jcbrill]

@mwichmann My guess is that A.dll does not export any symbols; therefore, an import library is not generated.

[mwichmann]

@mwichmann My guess is that A.dll does not export any symbols; therefore, an import library is not generated.

That's certainly true for the test file in the PR. I've added those in a worktree here, and it still doesn't, so I think I must be doing something wrong with that.

[jcbrill]

The following builds libA with MSVC from the command-line outside of the test. It correctly produces the A.lib file. I haven't checked to see if it breaks gnu/gcc yet. It shouldn't but one never knows.

libA/libA.h:

#ifdef _MSC_VER
  #ifdef BUILDINGSHAREDLIB
    #define MYAPI __declspec(dllexport)
  #else
    #define MYAPI __declspec(dllimport)
  #endif
#else
  #define MYAPI
#endif

MYAPI void libA();

libA/libA.c:

#include <stdio.h>
#include "libA.h"

MYAPI void libA() {
    printf("libA\\n");
}

libA/SConstruct:

SharedLibrary('A', source=['libA.c'], CPPDEFINES="BUILDINGSHAREDLIB")

I updated the test and it failed though.

P.S.: pretty sure the extra backslash in the print statement is erroneous.

[mwichmann]

@mwichmann My guess is that A.dll does not export any symbols; therefore, an import library is not generated.

That's certainly true for the test file in the PR. I've added those in a worktree here, and it still doesn't, so I think I must be doing something wrong with that.

I have an alternate independently developed example which does work, so later I'll reconcile the two and hopefully find what errors I made adjusting the original.

[mwichmann]

The following builds libA with MSVC from the command-line outside of the test. It correctly produces the A.lib file. I haven't checked to see if it breaks gnu/gcc yet. It shouldn't but one never knows.

libA/libA.h:

#ifdef _MSC_VER
  #ifdef BUILDINGSHAREDLIB
    #define MYAPI __declspec(dllexport)
  #else
    #define MYAPI __declspec(dllimport)
  #endif
#else
  #define MYAPI
#endif

MYAPI void libA();

libA/libA.c:

#include <stdio.h>
#include "libA.h"

MYAPI void libA() {
    printf("libA\\n");
}

libA/SConstruct:

SharedLibrary('A', source=['libA.c'], CPPDEFINES="BUILDINGSHAREDLIB")

I updated the test and it failed though.

P.S.: pretty sure the extra backslash in the print statement is erroneous.

yes, that looks like what it should want. And the backslash is left over from when it was a string-print in the testcase, once listed as raw-string, then not, then I think put back. Anyway, this should be enough to wrap it up.

[jcbrill]

Attached is a version of test/Configure/CheckLibWithHeader_extra_libs.py which appears to pass on Windows using msvc and WSL using gcc.

CheckLibWithHeader_extra_libs.zip

In hindsight, the double backslash may be necessary in the test since it is not inside a raw string (e.g., r""").

[mwichmann]

Attached is a version of test/Configure/CheckLibWithHeader_extra_libs.py which appears to pass on Windows using msvc and WSL using gcc.

CheckLibWithHeader_extra_libs.zip

In hindsight, the double backslash may be necessary in the test since it is not inside a raw string (e.g., r""").

thanks. I have one working here, but will look over.

[mwichmann]

Okay, this should be working now. Thanks to @jcbrill for confirming that what I was figuring out was on a workable path.

[rdbisme]

Thanks for bringing this forward :)

[bdbaddog]

I think you should just be able to do this once as
test.dir_fixture(['fixture', 'checklib_extra'])

[mwichmann]

I suppose. If you think it's worth doing yet another iteration...