Failure of the latest Windows builds

[illi-kun]

Jenkins builds fail with the following console output:

...
obj/visitable_remove.o:visitable_remove.cpp:(.text+0x1894): undefined reference to `item::in_its_container()'
collect2: error: ld returned 1 exit status
make[1]: *** [cata_test] Error 1
make[1]: Leaving directory `/var/lib/jenkins/cataclysm-reference-repo/Windows-Curses/tests'
make: *** [tests] Error 2
...
Build step 'Execute shell' marked build as failure

[BorkBorkGoesTheCode]

Build 5081 may have been the first

[Coolthulhu]

By the way, why does Jenkins care about tests? I don't think they're supplied with the executable.

We should probably have a make switch that disables tests. And if we do have one already, Jenkins should invoke it.

[illi-kun]

Jenkins cares about that for bringing developers' attention to (potencial) issues highlighted by the failed tests, at least I guess so.

[Coolthulhu]

It should recover from mistakes when it's possible. It can announce a failed build status on all PRs, but it should still produce the regular executable anyway.

[illi-kun]

It sounds reasonable.

[Zireael07]

I need my Cataclysm fix. Any fix for this coming?

[Coolthulhu]

I don't understand how does this bug happen.
Could it be that the compiler inlines in_its_container and removes it from the object file?
The function is very simple (literally just return in_container( type->default_container );).

If this is the case, guaranteeing that tests always build may be a pain, so we really need to make Jenkins understand "partial failure" better.

[Coolthulhu]

I didn't fix the builds, but at least I changed the error. Now it's one that can't be fixed with PRs alone.

w64-mingw32.static-g++  -DRELEASE -DCROSS_LINUX -DGIT_VERSION -DLOCALIZE -ffast-math -Os  -Wall -Wextra   -Werror -std=c++11 -MMD  -DLUA -c src/version.cpp -o objwin/version.o
make: /home/narc/mxe/usr/bin/i686-w64-mingw32.static-ccache: Command not found

I'm pretty sure only @narc0tiq can fix it.

[Rivet-the-Zombie]

Has... has our mingw32 compiler gone AWOL?

[kevingranade]

It should recover from mistakes when it's possible. It can announce a failed build status on all PRs, but it should still produce the regular executable anyway.

Nope, that removes any pressure whatsoever from keeping the tests building, then they bitrot and become useless. If nobody cares enough to keep tests building on a platform, we're evidently not supporting that platform anymore.

This particular error is quite odd, it's trying to link the tests before doing anything else.
I suspect this might be the actual problem: http://ci.narc.ro/job/Cataclysm-Matrix/Graphics=Curses,Platform=Windows_x64/5082/console

ccache /home/narc/mxe/usr/bin/x86_64-w64-mingw32.static-g++ -DRELEASE -DCROSS_LINUX -DGIT_VERSION -DLOCALIZE -ffast-math -Os  -Wall -Wextra   -Werror -std=c++11 -MMD  -DLUA -I../src -Wno-unused-variable -Wno-sign-compare -Wno-unknown-pragmas -Wno-parentheses -c iuse_test.cpp -o obj/iuse_test.o
objwin/cursesport.o:cursesport.cpp:(.text+0xbc1): undefined reference to `handle_additional_window_clear(WINDOW*)'
collect2: error: ld returned 1 exit status
make: *** [cataclysm.exe] Error 1
make: *** Waiting for unfinished jobs....

[Coolthulhu]

Nope, that removes any pressure whatsoever from keeping the tests building, then they bitrot and become useless.

They aren't trivial to run on any supported platform. They're quite bittrotten already.
Linux x64 here, trying to run cata_tests just segfaults instantly.

[kevingranade]

So fix it or at least report it with some supporting information, I'd rather start dropping platforms than drop the tests. Seriously, if we don't get some momentum behind unit tests, this project WILL implode under its own weight.

[Rivet-the-Zombie]

Looks like it's fixed as of build 5136, and I'm not entirely sure why. I'm guessing narc did something on his end.

[narc0tiq]

I'm guessing narc did something on his end.

Nope, I've done nothing.

[narc0tiq]

Weird, I definitely recall there being a make clean in the mainline-build script. Apparently, that went missing at some point. It's why 5135 wasn't marked as fixing the issue -- for some reason, make decided not to rebuild tests/visitable_remove.cpp until the later build.

I've put the make clean back in where it should be.