[many ports] improvements for linux/wsl

[cenit]

No description provided.

[cenit]

finally fixed the clapack very bad regression on linux introduced in #6542

[cbezault]

Could we have the libusb changes as its own PR.
Could we also get all the changes associated with BLAS/LAPACK in its own PR.

[cenit]

I am still fixing things up, I will start thinking about splitting commits later, if ever.
To be honest, now it's a moment of crysis with vcpkg because I am loosing months of work fixing things up (I think recent PRs of mine are here to demonstrate it) and it seems there's no end, also because broken PRs are still creeping in breaking stuff (lapack/openblas are totally broken now on master.............).
Finally CI has been introduced but it is still VERY insufficient (this tragedy of build_tests always disabled means that we don't even check library downstream usability even with self-included tests.... we only check if it builds but you can understand it's a VERY broken assumption).

Just let me finish this work, then do whatever you want to merge what you need.
I was fixing unnecessary/broken paths here and there while waiting for builds to finish during the night, IMHO just a thorough human read and analysis should be enough to verify it and ease a lot of work to disentangle commits that were meant to make things work...

[cenit]

I'd like to consider this work as done, in order to be ready for OpenCV 4...

[cbezault]

And I don't have the heart to pick the changes out into different prs. I'll audit the changes and merge once we get a green check.

[cenit]

@cbezault ok, seems reasonable.
CI on osx is requiring an unexpected extremely long time, don't know why. Apart from that, Linux is ok and windows has a regression for ffmpeg on x64-uwp which looks like a CI problem: MSYS just melts down at the beginning, being unable to install some packages required to work... also, I can't reproduce it locally. I already tried many times to trigger a new build, the scary thing is that it always fails at the MSYS initial setup?!??

[cenit]

@Rastaban maybe you know better what's going on here on windows.
Also, I found out that the logic behind CI is not so clear. I was sure that if B depended on A, modifying A would trigger a rebuild of B, but this is not always the case, I had to manually trigger a rebuild of B to have its regression fixed thanks to the fix in A

[cbezault]

The Pacman error is a known issue that we're trying to fix permanently. It has to do with a lock file not getting deleted properly.

Modifying A should retrigger a build of B unless you changed A back to a version that has already existed in the past. Anything else is a bug.

[cenit]

you can just see the last two commits... just triggering rebuild of ports seems enough to fix them... I modified their dependencies to fix them (the bug was not in the ports themselves but in their dependencies), but it seems that they were not rebuilt and I had to do it manually.
Without documentation i also realized that reasonable rule, but somehow it does not work always

[cbezault]

I'll look into it once I'm in today. It's concerning.

[cbezault]

Also don't bother retriggeing uwp.

[cenit]

green mark? :)

Please let me know any other review request you might have. Also because #5169 is ready, so I'd like to close this work

[cbezault]

Sorry to slow this down.

[cenit]

No problem. Better to fix problems now before it’s too late. Just wanted to avoid wasted time, but review requests are never wasted time