race condition with uv.new_work()

[FelipeLema]

Hi, there

I'm using luv within a neovim plugin; using new_work() has a race condition that ends up with an abort() in the separate thread.

Some users have been able to reproduce it here using neovim: https://codeberg.org/FelipeLema/cmp-async-path/issues/29

In that issue you can check further info (optimized stacktrace, "it is not the use of vim module in separate thread")

I dunno if it helps, but... can consecutive calls to new_work() be done without problems? (this is the first problem I can think of right now, but no evidence that points to it).

[truemedian]

The stacktrace points at https://github.com/neovim/neovim/blob/master/src/nvim/event/loop.c#L44 being the abort called, I'm not sure luv is unrelated, but I'd check with neovim that you're not doing something that isn't allowed, and that they're also handling worker threads correctly.

new_work just allocates state, it doesn't even call into libuv, so its plenty fine to call as many times as you want. queue_work should be equally safe, as it just queues the work context.