@gkdr, I made this^^ ready for review now.

The pre-merge GitHub Action logs can be viewed at https://github.com/hartwork/lurch/actions .

Moving CodeCov integration to GitHub Actions will need a new secret on the repository. We can do that migration before or after merge of this pull request.

PS: AppVeyor CI seems to also need disabling on their website, deleting the file alone doesn't make it stop.

it looks like the windows result is not loadable yet. report from a friend using windows, for both versions, all the zip contents unpacked into the plugins dir:

(21:41:43) plugins: probing C:\Users\USER\AppData\Roaming\.purple\plugins\lurch.dll
(21:41:43) plugins: C:\Users\USER\AppData\Roaming\.purple\plugins\lurch.dll is not loadable: `C:\Users\USER\AppData\Roaming\.purple\plugins\lurch.dll': The specified module could not be found.

@gkdr I notice now that for Windows test test/test_lurch_loadable.c passes in (SHARED=OFF) but fails in (SHARED=ON), with the same error. We ignore tests failures on Windows and that's why I didn't notice earlier. It could be that we need to copy the DLL closure into the build folder before running tests, but then that would not explain the problem on your friend's machine, provided they copied all DLLs in there. I'm also wondering if maybe the jabber plugin needs to be in the same folder or so.

It could be that we need to copy the DLL closure into the build folder before running tests

@gkdr it turned out that doing that^^ fixed the issue in the CI. It also implies that your friend did not have all needed files at the right places. My guess is still the jabber plugin DLL.

is it maybe related to setting the rpath like here?

is it maybe related to setting the rpath like here?

lurch/Makefile

Line 90 in 71c806b

LDFLAGS += -ldl -lm $(PKGCFG_L) $(LJABBER) -Wl,-rpath,$(PURPLE_PLUGIN_DIR)

@gkdr excellent point, let me look into that more.

is it maybe related to setting the rpath like here?

lurch/Makefile

Line 90 in 71c806b

LDFLAGS += -ldl -lm $(PKGCFG_L) $(LJABBER) -Wl,-rpath,$(PURPLE_PLUGIN_DIR)

@gkdr excellent point, let me look into that more.

@gkdr thanks for bringing RPATH to my attention, this is important and was not working as needed. What I found out is that:

• RPATH is a topic on Linux but not Windows.
• By default, CMake builts with an explicit RPATH but the resets it in the binaries at install time.
• As a result, current make install would install a lurch.so that is not loadable by Pidgin while our own make install-home is loadable by Pidgin just fine (because it still has the RPATH set (or not yet reset)).
• If we make CMake use the installation-time RPATH during build, the test suite will fail.
• If we make CMake NOT use the installation-time RPATH during build, then make-install home will install binaries with the wrong rpath.

Because of all that, what I have done now is this:

• Make make install-home install files libaxc.* and libomemo.* (when using a bundled copy in shared library mode) as well
• Split off target lurch_home just for make install-home with the right RPATH set at build time.

Does that approach work for you?

@gkdr you keep running away from #pidgin on librachat before you see my replies :)

It looks like the lurch dll's are trying to load libxml2-2.dll weirdly, trying to load DllMain() as a funtion, rather than linking to the functions in libxml:

I'm using https://github.com/lucasg/Dependencies and https://www.dependencywalker.com to scan for dependencies

Compare to other dlls that get loaded which look like

Also the plugin is trying to load in libsqlite3-0.dll, but Pidgin ships with sqlite3.dll, so might be superflous?

thanks a lot, @EionRobb! that does indeed look a bit weird and is a very good starting point :D

Also, (and its probably unrelated) just running something else through GDB and saw:

(21:56:04) plugins: probing C:\Program Files (x86)\Pidgin\plugins\lurch_semi_shared.dll
Dwarf Error: wrong version in compilation unit header (is 5, should be 2, 3, or 4) [in module C:\Program Files (x86)\Pidgin\plugins\lurch_semi_shared.dll]
Dwarf Error: wrong version in compilation unit header (is 5, should be 2, 3, or 4) [in module C:\Program Files (x86)\Pidgin\libsqlite3-0.dll]
(21:56:05) plugins: C:\Program Files (x86)\Pidgin\plugins\lurch_semi_shared.dll is not loadable: `C:\Program Files (x86)\Pidgin\plugins\lurch_semi_shared.dll': The specified procedure could not be found.
Dwarf Error: wrong version in compilation unit header (is 5, should be 2, 3, or 4) [in module C:\Program Files (x86)\Pidgin\plugins\lurch_semi_shared.dll]
Dwarf Error: wrong version in compilation unit header (is 5, should be 2, 3, or 4) [in module C:\Program Files (x86)\Pidgin\libsqlite3-0.dll]

@gkdr Can you merge this PR?

@gkdr, @xvitaly: Have you looked the last changes from @hartwork?

@gkdr: Any news on this old important PR?

Hi @Neustradamus! At this point, as far as I know, this does not work, as in the resulting plugin won't load on Windows. Neither of us has a Windows computer to test this, so I don't know how to develop it.

@gkdr as long as a re-run gets green CI, we could merge it and ask for pull requests from people actually using Windows. What do you think?

Hi @Neustradamus! At this point, as far as I know, this does not work, as in the resulting plugin won't load on Windows. Neither of us has a Windows computer to test this, so I don't know how to develop it.

I can take another look.

I couldn't get it to build on Windows, as there's no mingw-w64-i686-pidgin package in pacman any more; pacman gives an error "error: target not found: mingw-w64-i686-pidgin". Looks like there's only 64-bit packages now? https://packages.msys2.org/base/mingw-w64-pidgin

Tested by building on Windows with act -P windows-2019=-self-hosted --pull=false -W .github/workflows/windows.yml (using https://github.com/nektos/act for local action runners)

@EionRobb thanks for trying and the hint, I have adjusted for 64bit now and the CI is now back to green.

@hartwork What value is added if this does not add Windows support?

@EionRobb Thank you!

@hartwork What value is added if this does not add Windows support?

@gkdr

• the pull request also replaces zombie AppVeyor CI by GitHub Actions
• the pull request also migrates to a CMake build system e.g. to
  • pave the way for compilation for Windows
  • get out of the hard-to-get-right hand-written makefile zone into a more high level language
  • add high-level switches to the build system for end users and packagers
  • add proper error reporting for missing dependencies whereas the makefile lets you run right into compile errors
• the pull request also adds test test_lurch_loadable.c
• building for Windows is a precondition for running on Windows, not the other way around; with no Windows support in current dev, adding a Windows build covered by CI is a big step forward for first class Windows support, even if it is not working for everyone yet in practice.

Would you like to continue with the current Makefile on dev, instead?

@gkdr: Have you seen the latest @hartwork comment?