On 07/08/2018 04:39 PM, Dave Reisner wrote:
On Sun, Jul 08, 2018 at 06:10:09PM +0200, Andreas Baumann wrote:
Hi,
Sorry, if this message is a little bit out of order, I only just subscribed to the pacman-dev group.
I want to raise the issue of bootstrapping: - automake, autoconf, libtool and friends need m4 and perl and bash - meson needs python
Bootstrapping m4 and perl is comparatively easy to bootstrapping python.
So then build a minimal python as a "stage 1". You only need libffi and expat, which themselves have no build dependencies beyond glibc.
Upstream has heard this argument before, thus:
http://mesonbuild.com/FAQ.html#why-is-meson-implemented-in-python-rather-tha... http://mesonbuild.com/Use-of-Python.html
Porting to other architectures is easier if there are not too many dependencies.
See also my experiments here:
https://git.archlinux32.org/archlinux32/bootstrap32
and some porting work from oaken-source (Parabola):
I'm not sure exactly what I'm supposed to get out of these. The reality is that I don't really buy your argument. If you're bootstrapping Arch Linux, you need to build Python anyways. You'll have to build Meson in order to build systemd. You'll have to build Python in order to run pacman's unit tests (you should absolutely care about these for bootstrapping). Changing your bootstrap order to accomodate this doesn't really seem that onerous.
Anyways I pointed out that meson has cross-compilation support, and cross-compilation is already in play, so this should not be onerous. He agreed that this makes bootstrapping no longer an issue. I'd actually forgotten that the testsuite was in python :D but bootstrapping generally means building a minimal system, often with tests disabled, sufficient to cleanly rebuild everything with tests re-enabled. So I'm not sure that itself was a blocker. -- Eli Schwartz Bug Wrangler and Trusted User