Hi Ivan,

On 2026/11/17 21:03, Ivan Carvalho wrote:
Maytham wants version numbers, which I agree with. Ivan suggests also
renaming the package to zig-core, because... I don't know the reason
why?

This was suggested in debian-devel because functionalities are being
stripped. Upstream is very worried about people running "apt install
zig" and having a suboptimal experience.

There is no precedence that I could find in the archive for doing this kind of renaming, but I am happy to be corrected. I did but a cursory search.

What is missing in terms of functionality, that I can see:
 - cross-compilation
 - static musl-linking (I intend to ITP the musl headers if Zig makes it
   into the achive)
 - glibc version pinning. This is because of missing abilists and
   start.S. I think we can get this back in, it's free code after all.
 - ThreadSanitiser (libsan). I haven't really looked into this yet.

So yeah, Zig is not fulfilling its promises of building static libraries that run everywhere.

You mentioned nim and swiftlang below. Neither of them renames their packages to swiftlang-core or so; and I am guessing that swiftlang is missing a lot more compared to its upstream, than zig to zig's upstream!

Still, nim and swiftlang also don't rename their packages because of missing functionality.

May I suggest the following to deal with this issue:

(1) At first, we do NOT upload the package 'zig-defaults'.

(That package is just a tiny meta-package with Depends: zig-0.17, plus a /usr/bin/zig pointing at /usr/bin/zig-0.17.)

(2) At first, we ONLY upload the package 'zig-0.17', which installs zig-0.17 and zig-0.17-src.

Now, when you install zig (apt install zig), you get an error, because that package does not exist. If someone really wants to install zig, they have to type apt install zig-0.17 (which meant they went looking for it to find it), and the Description will say clearly that there is no glibc in that version, and what that costs.

At the same time, ghostty can be built against zig=0.17 (it requires 0.17 anyway). We will possibly end up with a couple of versions of Zig in the archive, as long as it's not backwards compatible?

What do you think of this proposal? Would that work for you all?

I would like to upload the zig package to the zig-team Salsa repo.

Please refrain for now, I also want to make it clear to Maytham: we do
not have a working Zig upload for the compiler itself today.
https://salsa.debian.org/jmrx/zig has the furthest work on
devendoring, https://salsa.debian.org/ejjl/zig seems to have the most
work in autopkgtest but neither is uploadable as of today.

I apologise, I have been working on and off on this, but I see I didn't push much of anything to Salsa. I am merging two branches currently (devendor work, plus the one that uses zig-wasm-bootstrap) and will upload that as zig-0.17 to my branch. You can then have a look.

I did send an email earlier that I had successfully built zig with zig-wasm-bootstrap, then rebuilt zig with that zig, rebuilt the unvendored zig with that one, ran the full upstream test suite against it... I think we do have a working version of zig, and I will upload it so that you all can poke at it :) You might want to grill me about some choices made?

What do you think? Does this make sense?

Please send an MR to https://salsa.debian.org/zig-team/book with your
package layout proposal. We need to show the community that we are
organized to be taken seriously.

Will do. Funny reasoning, though.

I don't object to numbering the package, but while you highlighted
cases of package numbering remember that we are closer to
https://tracker.debian.org/pkg/nim and
https://tracker.debian.org/pkg/swiftlang of being a niche programming
language.

I suggest that we follow mostly what Go does, except for that they don't follow DEP-14 as much, presumably because Go predates it ;-) We can start with a clean bill, and laying the best foundations prevents changes later. I'll lay something out in the Book tomorrow, I hope, but Golang + DEP-14 is where I took my inspiration.

zig-0.17-libc also requires significant further discussion. If you
naively copy https://codeberg.org/ziglang/zig/src/branch/master/lib/libc
I strongly think it risks rejection in the new queue due to macOS, BSD
and MinGW headers. ...

Naively...? Anyway, here's how I see it:
 - macOS: Correct, non-free license. That's two files.
 - BSD: Incorrect. All DFSG free.
 - MinGW: Incorrect. ZPL-2.1 with BSD and LGPL parts, explicitly noted
   as FSF-GPL-compatible. Debian already ships those same upstream
   sources in its own mingw-w64 source package

So, it's mostly untangling the copyright notices, which is time consuming, and a bit mind-numbing. It's not that it can't be done. So, that is why I propose that that package comes a bit later, when the tedious work has been done. Am I missing anything?

> ... Move the discussion to Salsa so we can have a
> community record of decisions.

We're waiting for that mailing list. In the mean time, I don't understand why you want to be off the debian-devel list so badly. I think that is exactly the place for this, as long as there isn't a zig-lang list. We had some very useful input and feedback from others there, too!

Kind regards,

    Edmund


--
Edmund Lodewijks <[email protected]>
TZ: UTC+2 / GMT+2

Attachment: OpenPGP_signature.asc
Description: OpenPGP digital signature

Reply via email to