Yes, if you can create a debian-zig mailing list and perhaps a
debian-zig IRC chat with a Matrix mirror that would be great! Let's
get off debian-devel, this is a general list. We can post here again
of course if we need general guidance from the project.

I am not a Debian project member but yes, I have packaged a couple of
things (as did you). Let's coordinate.

Regarding C++: that's okay; we weren't on the same page, but that
happens. I will check with Jesse if the Zig test suite covers Zig with
C++. As long as we have some test coverage, we can see how things go.

n.b.: [email protected] invited me to lobste.rs, I will answer Andrew Kelley


On Fri, Nov 13, 2026 at 2:37 AM Edmund Lodewijks <[email protected]> wrote:
>
> Hi Ivan,
>
> On 2026/11/12 22:24, Ivan Carvalho wrote:
> > Can you invite me to lobste.rs so I can post a reply? Invite
> > [email protected]; that's my personal email (although the work
> > one would be fine).
>
> Sorry, can't do (don't have the option).
>
> > I will post a more detailed reply later, but of course we will ship a
> > compiler that supports C++. The idea of not being able to compile
> > Ghostty is absurd.
>
> Unless, of course, the deps can be pulled from the archive, which is the
> whole idea behind devendoring.
>
> Also, while Ghostty is cool, Zig is more than getting Ghostty into Debian.
>
> > Unvendoring and shipping a minimal Zig doesn't mean the result will be
> > botched. For libc++, we'd try to use libc++abi-dev, then try a
> > separate package like zig-libc++, then vendor referencing a precedent.
> > I need to check what Emscripten does today, they are a C++ compiler
> > targeting WASM that we should study.
>
> I guess I did not explain things clearly enough. Currently, Zig will not
> look at whatever system-C++ you have installed. The code path in Zig has
> not been written yet (it does work for command-line invocations, but not
> when using build.zig).
>
> > Unvendoring doesn't mean stripping everything, it just means trying to
> > reuse as much as possible from Debian itself to avoid duplicate work.
>
> I'm not sure that I understand / that that sounds correct.
>
> > No one wants to apply the same fix to thousands of packages, the
> > Debian release and security teams want to fix  libc++abi-dev once and
> > have it propagate everywhere.
>
> Again, I must have not explained the issue correctly. See above.
>
> @Maytham: Perhaps a Zig mailing list on Debian would be useful to keep
> these discussion open and public, and that they can be re-read and used
> again later. I think that much of what is written now will be lost for
> the future, which is a shame in terms of findings made. For now, I am
> Cc-ing debian-devel again.
>
> Another reason: I was not aware of there being any current effort in
> trying to get Zig into Debian when I started looking into it a few weeks
> ago. It caused a bunch of double work.
>
> Some coordination by a Debian DD (not sure if you want to be that, but I
> have been including you since I was told that you "own" the zig-team
> Salsa repo) of the attempt to get Zig into Debian would be very welcome,
> too.
>
> Kind regards,
>
>      Edmund
>
>
>
>
> == Ivan replied to: ==
>
> On 2026/11/12 22:05, Edmund Lodewijks wrote:
>  > Hi Ivan,
>  >
>  > Thank you for your email, and for adding me to the zig repo!
>  >
>  > I apologise, I didn't have much time today to work on Zig-in-Debian (or
>  > my emails) at all.
>  >
>  > On 2026/11/11 21:25, Ivan Carvalho wrote:
>  >> If you send an MR to zig-wasm-bootstrap extending the chain I can
>  >> review it. However, in my opinion, the most useful work you can do is:
>  >
>  > Okay.
>  >
>  >> 1. Try to shorten the number of hops from 0.16 -> 0.17. I haven't
>  >> looked at your transitions, but small patches that save a hop save us
>  >> build time. Saving a hop save us build time.
>  >
>  > I'll have a look, but not sure that can be done.
>  >
>  >> 2. Check if the bootstrap works in aarch64. Both 0.10 -> 0.16 and 0.16
>  >> -> 0.17. Worst case use QEMU. I haven't had time to test it, but this
>  >> is valuable data.
>  >
>  > Happy to do that!
>  >
>  >> Going forward, getting Zig into Debian means recognizing that most
>  >> tasks will be boring. Everyone wants to upload the package, few want
>  >> to solve the mundane tasks. Let Jesse unvendor things, he already
>  >> started. Focus on translating your hops to work with the Magefiles. I
>  >> will manage the repositories and secure upload sponsorship.
>  >
>  > Yep, very aware of the mundane efforts of getting this done. Happy to
>  > put in my share, to help Zig into Debian.
>  >
>  > I did already do a bunch of unvendoring in the last week -- with a
>  > script to test what works and what breaks when I remove stuff.
>  >
>  > Also, have a look at what AndrewK, Zig creator, writes on lobste.rs. I
>  > think you and Jesse might find this interesting:
>  >
>  > https://lobste.rs/s/0ouxwf/libghostty_1_0#c_rzalas
>  >
>  > I have been testing a little on this unvendoring thing based on what
>  > Andrew wrote. I can unvendor everything, but if I unvendor the C++
>  > stuff, things break. Zig can't compile C++ anymore, and Zig doesn't have
>  > the code yet to use the system's compiler even if it is installed (I
>  > found a TODO comment in the Zig code about that).
>  >
>  > So, the zig-team will have to choose between two options, it seems to me:
>  >
>  > 1. Whether we unvendor everything, including C++. That means that zig
>  > can't compile C++ on Debian until this is sorted upstream (and patched
>  > in Debian). Ghostty has deps written in C++, so... Bye bye Ghostty? Not
>  > necessarily, because we can just use those C++ deps from Debian straight
>  > (provided the versions match..). For example, harfbuzz is a C++ dep, and
>  > Zig would compile it for Ghostty with its C++ compiler. If we declare
>  > harfbuzz as a dependency of Ghostty, zig might not find it necessary to
>  > compile it from source (I have not tested this). Alternatively:
>  >
>  > 2. Whether we unvendor everything, except C++. This was my initial
>  > unvendor build and resulted in:
>  >
>  > Removing lib/libc, lib/include and lib/libtsan, and keeping lib/libcxx,
>  > lib/libcxxabi and lib/libunwind. That takes the source from 21,961 files
>  > to 4,758, and it removes the licences that are the actual obstacle:
>  > APSL-1.1, which is not DFSG-free, plus BSD-4-clause, ZPL-2.1 and a
>  > header that has IETF RFC terms in it.
>  >
>  > It's a big win all the same. The licenses (checked via `licensecheck`)
>  > look okay. By far the most files are Apache-2.0 with LLVM excepteion.
>  > Then the rest are Apache-2.0 and/or Boost, Apache-2.0 and/or Unicode;
>  > and 3 unknown (haven't looked further).
>  >
>  > On lobste.rs, @rcalixte pm'd me and shared two links of his, in which he
>  > had worked around the fact that Zig can't currently use C++. I haven't
>  > looked much into it, but considering what he patched, that would mean
>  > that we would have to patch EVERY software written in Zig with C++ deps,
>  > which seems to me to not weigh up against leaving C++ dirs with DFSG
>  > okay licenses in Zig for now.
>  >
>  > Anyway, just wanted to share this. Please do share it with Jesse as well
>  > if you like; I simply haven't found the time yet to reach out and say hi.
>  >
>  >> n.b.: I am not a zig-team admin, ask Mayham for membership.
>  >> n.b 2:  have already talked with DDs who are willing to sponsor this
>  >> package once things are suitable.
>  >
>  > I am sure DD's won't be an issue to find if we have a good proposal.
>  > What we mostly need now is coordination and focus to get something into
>  > NEW soon, so that it can be adjudicated, and that we have time to fix
>  > things if it isn't accepted at once.
>  >
>  > Happy to hear your thought. I haven't reached out yet to DD Maytham yet
>  > about joining Zig team, either. Will come in time. In fact, let me just
>  > Cc him, if he's heading the Zig-team. Hello @Maytham!
>  >
>  > This email got much longer than intended... I just wanted to say thanks,
>  > and let you know that I will find time again later.
>  >
>  > Kind regards,
>  >
>  >      Edmund
>  >
>  >
>  > P.S.: After writing this, and considering that it's lights-out where I
>  > live now, I decided to just Cc Maytham and Jesse. I hope you don't mind!
>  > Hi! Sorry for the somewhat rocky introduction.. o/
>
>
>
> --
> Edmund Lodewijks <[email protected]>
> TZ: UTC+2 / GMT+2
>

Reply via email to