From: tianon@debian.org
On Wed, 18 Feb 2026 at 10:59, Dr. Tobias Quathamer wrote:
> So, if you don't have any objections, I plan to switch the default
> golang compiler to 1.26 in the next few days.
It's a tiny bit tangential, so apologies for that, but I think it
would probably be prudent to finally do something about gccgo too,
while we're opening the hood on src:golang-defaults.
For a long time now, gccgo has been stuck at an implementation of Go
1.18, but even then, it's only a *partial* implementation of 1.18
(notably missing generics).
As noted and discussed in the CC'd https://bugs.debian.org/961916,
bin:gccgo-go's version is misleading at best and actively harmful at
worst, because it *looks* like it's 1.24, and this is where I think
the overlap with the proposal to update to 1.26 appears - that number
drifts even further from any truth.
In ideating on solutions, I was thinking about doing something really
goofy, like having bin:gccgo-go have a Version: that's 1.18 or
something, but that ship has sailed (we can't downgrade in the
archive), and it would be really poor behavior for one source package
to do that anyhow.
So concretely, what I'd propose for src:golang-defaults 1.26+ is that
we remove all traces of gccgo entirely instead. This is somewhat
disruptive, but I don't think it's unreasonable. Every interesting
architecture is currently supported by upstream golang-go, gccgo
hasn't been a suitable GOROOT_BOOTSTRAP for Go itself for several
versions now, and gccgo is a very outdated implementation. On top of
all that, I strongly doubt there are many (if any) meaningful Go
packages left in the archive that really successfully build with
gccgo.
I am going to work on an even more concrete MR for review after
sending this email. 🙇
♥,
- Tianon
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
