ccache: update to 4.0 (alternative approach for chroot-cmake)

[sgn]

Alternative for #25724

[pullmoll]

Hmm.. depending on chroot-cmake if $CHROOT_READY seems the wrong way round, no?

[sgn]

I was thinking about stage1 and stage2.
Either put inside chroot_ready or set it unconditionally. I don't really know. I haven't put much thought into that. 🤷

[pullmoll]

As a rule of thumb you usually do not depend on anything chroot-... because it is, well, in the chroot anyway.
That's why I thought that perhaps you meant

if [ "$CHROOT_READY" -a "$CROSS_BUILD" ]; then
        hostmakedepends+=" cmake"
fi

We need to test that anyway... I'll close my PR, your's is better.

[sgn]

This chroot-cmake is very basic cmake, it lacks a lot of cmake functionality (but enough to build ccache and real cmake).
I think we should treat it the same with pkgconf and pkg-config. IOW, don't add it into base-chroot but let's package add it into hostmakedepends.

[pullmoll]

Ok, I see, then this makes sense.
Let's hope that ccache doesn't become too fancy with its cmake scripts in the next updates :)

[pullmoll]

If we don't add it to base-chroot then a better name then would be something like cmake-minimal or cmake-bootstrap.

[sgn]

OK, I'll change it into cmake-bootstrap

[sgn]

Looking around, I think if ccache don't use those fancy ExternalProject, we should be fine.

The whole curl, libarchive, rhash, expat and nghttp2 was added for fetching tarball from external resource, verify checksum and build them.

[pullmoll]

I like your approach. If we don't need more than what a cmake bootstrap supplies that would be very space saving.

[pullmoll]

Running ./xbps-src bootstrap now behaves odd.
It seems to enter an endless loop trying to build ncurses.
When I break with ^C and redo the command it succeeds.
Not sure what kind of dependency loop (?) is causing this.

Did you try a bootstrap?

[pullmoll]

In srcpkgs/ccache/template add the following lines to fix cross builds:

if [ "$XBPS_TARGET_NO_ATOMIC8" ]; then
        makedepends+=" libatomic-devel"
fi

pre_configure() {
        if [ "$XBPS_TARGET_NO_ATOMIC8" ]; then
                vsed -i src/CMakeLists.txt \
                        -e "/add_library/a target_link_libraries(ccache_lib PRIVATE atomic)"
        fi
}

[q66]

bad short_desc, cmake is not a "cross make" or anything of the sort

[sgn]

Heh, they self-claimed "cross-platform make" and I was lazy. I'll be more lazy to just write "CMake"

[pullmoll]

On a 2nd try the ./xbps-src zap and ./xbps-src bootstrap succeeded to build stage0. No idea what was causing the loop in the 1st try.

[sgn]

Running ./xbps-src bootstrap now behaves odd.
It seems to enter an endless loop trying to build ncurses.
When I break with ^C and redo the command it succeeds.
Not sure what kind of dependency loop (?) is causing this.

Did you try a bootstrap?

Still running, my main machine is going down, so I need to bootstrap on real crappy machine now.

[pullmoll]

I've been testing it and already successfully built x86_64 bootstrap stage0 + stage1, cross base-chroot for i686 and x86_64-musl, and several thousand x86_64 packages without issues so far. IMO this PR can be merged.

[sgn]

I'll let someone else press the merge button.

[ericonr]

why is this commented?

[sgn]

Originally, I wanted to remove all 3rd party libraries except libuv since I know that libuv is required. I found out jsoncpp was required later on.

[ericonr]

shouldn't this be "full cmake" ? It's still real in that it can build packages.

[sgn]

I used that because CMake uses real real cmake to talk about cmake, I'll change description to "this is half baked cmake"

[sgn]

Ready again, now!

[Piraty]

why all the pain?
why not remove ccache from bootstrap packages, i don't see the necessity to have it that early really.
With ccache being bootstrap=yes we can't

1. make it build its manpages, as they need asciidoc and perl.
2. run tests with do_check() as it requires ctest which is part of cmake, but not cmake-bootstrap

[sgn]

For future reference, the discussion is on #26034