Tomasz Andrzejak (Jun 15 2026 at 12:27): Tomasz Andrzejak (Jun 15 2026 at 12:27):Hey :wave:
What's the recommended way of flushing stdout when using wasi/p3? In p2 we have explicit functions like blocking-write-and-flush. The stdout writes are line buffered by default and I don't see anything in wasmtime p3 implementation that would actually call flush, not even on drop or stream close.
https://github.com/bytecodealliance/wasmtime/blob/main/crates/wasi/src/p3/cli/host.rs#L92
Writer's poll_flush exists but it's never called. Am I missing something?
I could workaround this by using https://docs.rs/wasmtime-wasi/45.0.1/wasmtime_wasi/struct.WasiCtxBuilder.html#method.stdout and provide my flushing stream implementation but I doubt that this is the intended way of solving the issue?
Of course, I'm happy to prepare a patch if we think that's an actual gap, thanks!
Alex Crichton (Jun 15 2026 at 14:23):WASI-wise there's no flush operation on streams, bytes have either been acknowledged-to-be-sent or not. Implementation-wise though this is something we need to navigate, and are you seeing data not get printed to stdout?
Tomasz Andrzejak (Jun 15 2026 at 15:08):Yes, let me create a repro -- it is tricky doing with wasmtime p3 test harness because it captures stdout on process exit which flushes and masks the behavior. I need something more "interactive".
Tomasz Andrzejak (Jun 15 2026 at 17:05):A test with a potential fix:
https://github.com/bytecodealliance/wasmtime/pull/13654
Joel Dice (Jun 15 2026 at 17:21):I recall some informal discussion of adding a stream.lull intrinsic to the CM which would mean "there's probably not more data coming for a while, so this would be a good time to flush if applicable". Luke may be able to describe that in more detail when he's back from his vacation.
Alex Crichton (Jun 15 2026 at 17:24):for now though I believe this patch/fix are correct, it's tokio buffering things we didn't realize
Last updated: Jul 29 2026 at 05:03 UTC