Linux support, from SlimeVR-OpenVR-Driver

[kitlith]

Shit, CI is failing on linux because of the version embedding code.

Error: /home/runner/work/SlimeVR-Feeder-App/SlimeVR-Feeder-App/src/main.cpp:21:10: fatal error: version.h: No such file or directory

i don't know why. this is kinda important though because I want all the releases to come from CI.

EDIT: I can reproduce on WSL, so it's not limited to CI.

[funnbot]

-- version in git: doesn't actually print the version, so a tag isnt checked out in that case or something? Happens on windows and ubuntu builds.

[kitlith]

-- version in git: doesn't actually print the version, so a tag isnt checked out in that case or something? Happens on windows and ubuntu builds.

that should still result in version.h being generated, just with a blank version. but in my local reproduction (where it is parsing the version correctly, even), version.h just... isn't generated at all. no error or anything, it just isn't there.

[kitlith]

-- version in git: is output from the custom command, so it's getting run, but configure_file isn't actually producing anything i guess?

re: your latest commit: i'm ahead of you there, the directories seem to be set correctly

[funnbot]

Windows side prints this warning...

C:\Program Files\Microsoft Visual Studio\2022\Enterprise\MSBuild\Microsoft\VC\v170\Microsoft.CppCommon.targets(247,5): warning MSB8065: Custom build for item "D:\a\SlimeVR-Feeder-App\SlimeVR-Feeder-App\build\CMakeFiles\e9797dbf0013314fb2d386b44164446b\version.h.rule" succeeded, but specified output "d:\a\slimevr-feeder-app\slimevr-feeder-app\build\always_run" has not been created. This may cause incremental build to work incorrectly. [D:\a\SlimeVR-Feeder-App\SlimeVR-Feeder-App\build\SlimeVR-Feeder-App.vcxproj]

OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/version.h" always_run Ah, so the file is never generated, which ensures it always tries to generate the version.h?

[kitlith]

That's the idea, yeah. Was done that way to ensure that locally, a new commit always causes the version to include it on build.

There may be a better way to handle that part of the cmake setup (i.e. it doesn't seem to work on ubuntu, which I know because i tried doing touch build/version.h; chmod 000 build/version.h to at least see if i could get configure_file to error and instead it just skipped it)

[kitlith]

okay, if I manually run the custom command, version.h does appear. so why doesn't it appear during the build...?

i grabbed inotifywait and this is what it says happens:

build/ CREATE version.h.tmp
build/ OPEN version.h.tmp
build/ MODIFY version.h.tmp
build/ CLOSE_WRITE,CLOSE version.h.tmp
build/ OPEN version.h.tmp
build/ CREATE version.h
build/ OPEN version.h
build/ CLOSE_NOWRITE,CLOSE version.h.tmp
build/ CLOSE_WRITE,CLOSE version.h
build/ OPEN version.h.tmp
build/ DELETE version.h
build/ CREATE version.h
build/ OPEN version.h
build/ ACCESS version.h.tmp
build/ MODIFY version.h
build/ CLOSE_NOWRITE,CLOSE version.h.tmp
build/ CLOSE_WRITE,CLOSE version.h
build/ ATTRIB version.h
build/ DELETE version.h.tmp
build/ DELETE version.h

so it appearently does get created (at some point), it just gets deleted again afterwards i guess? perhaps it's time to re-do how cmake handles the versioning bit.

[funnbot]

Oh because the makefile generator deletes OUTPUT, I think thats just for clean though...

[kitlith]

yeah okay good that fixed that part of CI, but there's a bit of an error in the C++ that only shows up on linux: error: cannot bind non-const lvalue reference of type ‘simdjson::fallback::ondemand::object&’ to an rvalue of type ‘simdjson::fallback::ondemand::object’

[kitlith]

@funnbot okay. can you please test this now that it compiles? keep in mind that the fix for #6 isn't in this branch, but if you rebase off of the main branch you should be good.

[ImUrX]

should have pinged me, I need to add the socket in the server lol

[kitlith]

@ImUrX ah.

This also has one more thing that needs to be done that I apparently forgot to make a comment on the PR about.

We need to set RPATH so that we actually use the copy of openvr that we bundle with the feeder. In the meantime, setting LD_LIBRARY_PATH at runtime should work as a workaround for testing.

Heh, it probably also needs an update to use XDG_RUNTIME_DIR as well instead of hardcoding /tmp

[ImUrX]

yeah, i mentioned funnbot about that, I can commit it myself tbh

[kitlith]

Okay, that should actually be everything, for real this time. Just needs testing (have some artifacts: https://github.com/funnbot/SlimeVR-Feeder-App/suites/10935047338/artifacts/553084572) and then I'll merge it in.

Artifacts for the two platforms are getting merged in CI, I should probably fix that at some point, but it's okay for now, so I'm gonna punt on that. Not a blocker for this PR.

[kitlith]

facepalm i've been saying this is finally ready to test, except it isn't, because it still doesn't include the driver rename fix. I'll just do that rebase now.

[ImUrX]

Tested it and it works!