Re: [Buildroot] [PATCH 1/2] support/dependencies: introduce BR2_HOST_CMAKE_AT_LEAST

[Peter Korsgaard <peter@korsgaard.com>]

>>>>> "Romain" == Romain Naour <romain.naour@gmail.com> writes:

 > Some packages (e.g. libjxl) requires a quite recent cmake version,
 > that is not yet available in most distributions, especially those
 > LTS versions.

 > Currently, when we bump the minum cmake version we require, it gets
 > bumped for all packages, regardless of their own minimum required
 > version, which means that a given configuration will trigger the
 > build of our host-cmake even if the packages that require it are not
 > enabled and those that are would be content with the system-provided
 > cmake.

 > Since host-cmake can take quite some time to build, this can get a
 > bit annoying to pay the price of a host-cmake build that would
 > otherwise not be needed.

 > Some packages even use an alternative build system when available
 > since they requires a more recent version of cmake than the our
 > minum cmake version
 > (wpewebkit use Ninja: 78d499409f71d8a22b0632c8ebc06f67ee6ae6dd).

 > We introduce config options that packages can select to indicate
 > what minimal cmake version they require, and use that version as the
 > required minimal version required by the current configuration [0].

 > We would like to ensure that the currently selected minimum cmake
 > version is indeed lower (or equal) to the cmake version we package,
 > but that is not possible: dependencies.mk is parsed before we parse
 > packages, so we do not yet know the cmake version we have, and we
 > can't invert the parsing order as we need to know the requires
 > dependencies before we parse packages (so that we can build their
 > dependency rules in Makefile). So we can only add comments in both
 > places, that refer to the other location.

 > [0] note that this is yet not optimal, as in such a case, host-cmake
 > would be in the dependency chain of all cmake-based packages, even
 > for those packages that do not require it. The optimum would be for
 > each package to gain such a dependency on an as-needed basis, but
 > this is by far more complex to achieve, and would only speed up
 > cases where a single package is built from scratch (e.g. with:
 > make clean; make foo), which is not worth optimising (yet?)

 > Signed-off-by: Romain Naour <romain.naour@gmail.com>
 > Cc: Julien Olivain <ju.o@free.fr>
 > Cc: Arnout Vandecappelle <arnout@mind.be>
 > Cc: Peter Korsgaard <peter@korsgaard.com>
 > Cc: Yann E. MORIN <yann.morin.1998@free.fr>
 > Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
 > ---
 > v2: Use the commit log suggested by Yann
 >     http://lists.busybox.net/pipermail/buildroot/2023-June/668312.html

Committed, thanks.

-- 
Bye, Peter Korsgaard
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot