On 2024/05/30 07:35, Fabien ROMANO wrote:
> Haaa ok, thx for explanation, thus the port needs either.
> 
> ONLY_FOR_ARCHS = aarch64 arm amd64 i386 powerpc

No need for that: if luajit is not available, the dependency cannot
be satisfied, so it won't be built anyway.

> I have issues with googlesource, I understand your point. Github use a big cache
> to avoid generating new tarballs or they just care about timestamp ? Crazy but
> handy CI/testing (not for releases, I agree !).

They have caches, and they mostly try to avoid changes in the software
stack that would result in differing tarballs, but don't guarantee it.
We've certainly had times when people fetching from github in different
parts of the world got different files.

The archives can vary depending on commit activity too. Either later
commits on a tag, or from embedded 'short' commit hashes (there are
options to allow including them in git-archive files) which can get
longer if a later commit means that an old short tag is now ambiguous.

> This port is using a release, not a commit id, which is a fixed tarballs from
> github, this is perfectly fine ?

Releases by default just have links to on-the-fly/cached git-archive
files (the "source code .tar.gz" links) but project owners can also
upload "release artefacts" - the named files you'll see in some projects
- the latter are stable.


On 2024/05/30 00:42, Anthony J. Bentley wrote:
> Switching away from Lua is necessary because newer minetest decided to
> stop supporting system installations of Lua. So the choice is to either
> build the Lua embedded in minetest, or switch to system luajit.

ah, gotcha. Probably not worth the complexity in this case - now that
luajit's updated to the 2.1 branch it covers archs where minetest might
be useful.