The libstdc++ debug mode includes some overzealous runtime checks (O(n)
checks for O(log(n)) operation), which basically ruin our debug_renumber
test. We have to resort to something very ugly to fix that.

Add explicit noexec stack markers in asm files

See merge request !88

The cache mechanism outsmarts us, and recycles and outdated cache...

This gets rid of a linker warning.

Resolve "factor base cache not available with your libstdc++ library, sorry."

Closes #30044

See merge request !87

The fbc file is architecture-specific anyway. Presently, we've
encountered 4k and 16k page sizes. Alphas and itaniums are no longer to
be seen, but did use 8k page sizes IIRC.

cmake config tweaks, mostly for x-compiling

See merge request !85

This is somewhat hacky, because we have to provide for the situation
where the docker passes its docker socket to the container, *and* fix
the permissions ofthe user in the container as well.

CMAKE_BINARY_DIR does not always in a slash, and anyway the
cmake-generated files should be below CMAKE_FILES_DIRECTORY. The pattern
added by this commit is the one that is encountered elsewhere in cmake
prose.

Prepare container images at each commit.

See merge request !84

Container images are not super large. Actually most of the size comes
from python. This is still WIP, in the sense that more functionality
would be desired (dlog, automatic selection based on hardware, etc)

Resolve "gcovr-5.1 failure [Job Failed #2065565]"

Closes #30043

See merge request !83

I tend to agree that the data that gets produced is borked anyway. It's
not really gcovr's fault. Well, at least 5.0 seems to be content with
this data as it is. Let's use it for the time being.

This will certainly resolve over time, but in the meantime, we do either
that, or stick to an old version (which will anyway break someday if our
base image no longer has it)

FreeBSD 13.1 released today. Do our tests work?

See merge request !82