Re: rust 1.94

[Tobias Nygren <tnn%NetBSD.org@localhost>]

On Sat, 28 Mar 2026 20:15:43 +0100
Thomas Klausner <wiz%netbsd.org@localhost> wrote:

> Hi!
> 
> I'm planning to update lang/rust to 1.94.1 using the package in
> wip/rust194 (from he@ with help from adam@ and pin@).
> 
> bulk-test-rust is fine.
> 
> This will fix a CVE in the tar crate and help with some rust packages
> that already need rust 1.92 or newer.
> 
> I'm currently waiting for wip/rust194-bin to provide the same version,
> and for Illumos fixes for wip/rust194, so this is a heads-up for the
> commit that will follow when these are in.

I gave it a try by building www/chromium and it failed like this:

   Compiling swc_config v0.1.15
fatal runtime error: out of TLS keys, aborting
error: could not compile `sourcemap` (lib)

The build continued if I set PTHREAD_KEYS_MAX=1024 in the environment.
Not sure why it needs to do so many pthread_key_create(3) calls or
if this is a new regression as I have not tried to build chromium
before. (I'm adding aarch64 support to chromium).

The default value of PTHREAD_KEYS_MAX looks to be 128. Annoyingly there
is no way to bump it from rust runtime code after the pthread
initializers have ran.

Open questions right now:

- Is this a new problem?
- Does on only happen on many-core systems?
- Should we bump the default in NetBSD?
- Do we need to add PTHREAD_KEYS_MAX to the pkgsrc environment for
  packages that run rustc?

Side note, getconf(1) does not seem to know about _POSIX_THREAD_KEYS_MAX
if someone is up to fix that.

Kind regards,
-Tobias