snarkmaster wrote: @ChuanqiXu9,
(1) I'm not actually changing `libc++`. I just added an end-to-end test for this compiler feature. It's backwards- and forwards-compatible, so I don't think there's any risk to bundling it with this PR. (2) My patch isn't a fully general solution to https://github.com/llvm/llvm-project/issues/148380. Rather, it's a (probably) better solution to the more narrow class of problems of short-circuiting suspends. To use my thing, two things have to be true: - You actually have to opt into the new attribute and define `await_suspend_destroy` to get the perf benefits. - You can only do so if your `await_suspend` would destroy the coro **unconditionally**. When both of those are true (opt-in & unconditional deestroy), it results in a much simpler compilation setup -- HALO, less IR, less optimized code, better perf. https://github.com/llvm/llvm-project/pull/152623 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
