Closed Mark-Simulacrum closed 2 years ago
Have we looped in the tokio/mio people already?
I think this is all from old versions and yes, but it seems like a good idea to re-check-in, especially since bug reports directed at them aren't implausible.
We've discussed this pretty extensively in libs-api, and we're comfortable with this change; we do need a release note here, though.
The PR is labeled appropriately.
Filing this mostly so that we're aware of scope of impact per Crater run and can potentially use this as a landing ground for complaints/concerns (probably should have some issue for that even if not this one, maybe pinned). The scope of impact is pretty depressing, but I don't think the decision by libs-api will change since this isn't new data, so we'll need to clearly communicate it (I believe already planned into release post + notes). It seems likely that we might want to invest in mechanisms to stop future breakage like this (e.g., more strict linting/errors on transmutes of std types).
introduced by PR #78802
Root (compilation) regressions:
Test failures, which are probably caused by broken pointer casting in older tokio/mio/etc (haven't chased down specifically though), leading to InvalidInput / invalid argument from the kernel: