Fix typo in TLS initialization code

[LemonBoy]

Come on guys, did you even try to run anything on ARM32?

[andrewrk]

I tested building it, which is how I found a7fd140. I don't personally have any hardware for testing arm32 currently. I believe @Snektron did plenty of testing for #3152 but made some typos when I asked for last-minute modifications before merging the PR.

I'm definitely open to suggestions on how we can get CI testing hooked up for ARM32.

[andrewrk]

In related news, I did order a PineBook Pro, which is expected to arrive in October. That's arm64 though.

[marler8997]

Does qemu not work for arm32 testing?

[andrewrk]

It works but the slowdown is something like 20x. Full zig test suite - including testing optimized builds - takes ~48 hours. A more custom CI solution would be needed for this, as most services have much lower limits. The most generous one is Azure at 6 hours.

See here for attempt to use qemu on sr.ht: #2978 (comment)

[andrewrk]

I would still be interested in setting something like this up. We could have a dedicated instance that runs the latest master branch commit, however long it takes. When that finishes, instead of trying the next commit, skips to the latest. So at least we will find out results periodically, rather than never. With this strategy we can probably take advantage of qemu and greatly increase architecture/OS coverage.

[LemonBoy]

It works but the slowdown is something like 20x.

I'm using QEMU's user-mode emulation to run the cross-compiled test artifacts.

[andrewrk]

That's good to know. I need to learn how to use that.

[andrewrk]

Ideally we could add it as a build option, e.g. -Denable-qemu-testing, and it will run the applicable cross-compiled test artifacts. Combined with -Denable-wine-testing I think we'd be able to run everything except for macOS.