Nowadays, LibGPR2 does not depend on Libadalang nor Langkit directly
(the dependency on Langkit is vendored), so unconditionally rebuilding
everything that depends on the `eng/gpr/gpr` repository through anod's
`--smart-rebuild` option serves no purpose in the context of our CB, and
takes a lot of time. Do not override GPR2 sources to get rid of this.
Nowadays, GNATformat is part of GNAT, so just having an explicit build
of GNAT with the `--smart-rebuild` anod option and `gnatformat-edge` in
`anod vcs` is enough to trigger its build.
Note that since GNATformat has been stabilized, its build on the safe
track does *not* use the Libadalang wavefront, making the current
duplicated build useless.
Either the generic_anod_ci script changed where it performs repository
checkouts recently, either the CB never worked since the transition to
generic_anod_ci...
The default cloning branch was recently switched from "master" to
"next", this breaking everything that depends on the "master" branch
expecting it to be the default, such as our CB.
Since gnatformat build failures prevent Libdalang test machines from
booting, it makes sense to check that Libadalang/Langkit commits do not
break gnatformat's build.
We need it to build ada_language_server for aarch64-linux. There is
no GitHub Action runner for aarch64, so we will use the cross toolchain.
Refs eng/ide/ada_language_server#1251
Refs eng/ide/ada_language_server#1249 for removing `edge` branch of `gpr`.
The purpose of the only scheduled pipeline we have for now is to create
a pre-built Langkit package used to speed up the Libadalang CI. There is
no need to run the CB in this pipeline.
The CB currently requests two builds: gnat and libadalang_for_customers,
each one with its own "anod build" shell command. With enough bad luck,
new components can be uploaded to Cathod in between, making the CB
build during the second step components that were already built during
the first step, making the CB timeout (the CB is already close to the
timeout limit).
Switch to an electrolyt plan instead, so that this kind of issue is less
likely to happen (the Anod scheduler/Cathod requests run only once).
