Ekaitz's tech blog:
I make stuff at ElenQ Technology and I talk about it

During the latest posts I described that we managed to build GCC using our bootstrapping chain, and also that we built a modern GCC using our bootstrapped one, but now, some connection work has been done. It’s time to report on that.

Bootstrapping chain discovered

In my commencement.scm project, I started working on a bootstrapping path following Andrius Štikonas’ steps in live-bootstrap. His work is what is driving everything for several reasons: he knows how to do it well and Guix has a handful of peculiarities that make this way more complex than in live-bootstrap.

First of all, live-bootstrap uses custom Makefiles for most of the projects, that allows it to build things without relying on the feature detection that the configure scripts normally use. We could do the same thing on Guix but that would mean rewriting many packages from scratch and Guix package definitions would become way more complex than what they are right now.

Second, live-bootstrap is launched in some kind of a chrooted environment that is shared by all steps. This means all the packages in it share the / folder, and they can easily find other programs in /bin and libraries and stuff in the folders expected by the FHS. In Guix we do not support the FHS, and our packages are isolated, so we need to tweak the environment variables of the packages in order to make them find the dependencies. This sometimes requires patching the source of some packages, and here problems appear.

The most heavily patched package in this regard is actually GCC. It needs to be able to work well with our environment and that’s not easy to do. This makes my job harder.

The good news is Andrius managed to build GCC 9.5 with the bootstrapping chain in live-bootstrap and that means we discovered a full-source bootstrapping path for RISC-V in 64 bit that is working right now.

Time to make it reach a distro

I managed to build GCC 9.5 with C support, using our bootstrapping path, and that’s more than nothing, but we are working with Efraim Flashner (one of the Guix maintainers) to add full support for GCC 9.5 and also Janneke is working on top of my commencement.scm file to merge it with the support they already did for x86 (and was already done when I joined this effort).

So, the efforts to include our work in Guix are currently happening and I don’t think there’s much Andrius and I can do to help in this direction. The project is on the hands of the people that know the best.

Issues

During the project we found some small inconveniences that we could not fix, and those must be improved but they’ll take long time, as they come from some related projects. I’d love to say I have time, energy and knowledge to fix those, but most of them are not under my control.

First we have the problem we detected with Gash, that produces hangs in the very first steps of the bootstrapping process. This is really hard to fix but Timothy Sample, the project maintainer, is working on the issue.

Second, some of the projects we use in the bootstrapping process require some packaging skills I lack at this very moment, but they have been bootstrapped in live-bootstrap so this can be done. The projects are flex and bison, and maybe a couple more, but they are out of the scope of the project so I just used the non-bootstrapped ones and left them as to-dos in the commencement.scm file¹.

What now, then?

I think what to do next is pretty obvious now: handover to Guix.

I’ll open an issue in Guix in the following days where we’ll discuss the inclusion of this bootstrapping path in Guix in the following months, as the other related issues are fixed and more steps are included in the chain.

This handover process will take time because the bootstrapping path doesn’t end in a modern GCC, but in a proper GCC, GLibC, and some other packages that are required for almost everything. Our discovery is enough to continue further, but we didn’t do that as our goal was to provide the RISC-V support in the places were it wasn’t ready. We have already shown that we did that. Now I guess it’s time for the distros to catch up.

Of course, with our help. As always.