Henri Sivonen

I have tried distro-provided ICU4C 70, which is presumably compiled with GCC, ICU 72.1 with clang 14 from `mach bootstrap`, and ICU 72.1 with clang 15 from apt.llvm.org. In all...

ICU4C build system used `-O3` with clang. Last I checked, which was a long time ago, `opt_level=3` was the cargo default, but I haven't checked whether cargo criterion overrides it.

It sure looks a lot like the Rust code should have gotten compiled with `opt_level=3` as well.

> However, for Hanzi, ICU4X should make the decision in the first branch of the branchy code. It would be worthwhile to check that the branchy code is compiled with...

It turns out that I was running benchmarks without Rust-side LTO. LTO improves perf for the other Rust crates, generally improves ICU4X perf in cases where the trie returns non-default...

Updated the results to use Rust-side LTO. Moved and linked the old results.

If I'm reading the sources right, the semantics of the decomposition callback appear to be that singleton decomposition reports the second character of a decomposition as U+0000.

I intend to write a draft for this and then move this to help wanted for end-to-end polish.

"unstable" can mean many very different things of different scariness. If we mean semver-unstable, let's have "semver" in some form in the name.

[`window-reuse-in-nested-browsing-contexts.tentative.html`](https://wpt.fyi/results/html?label=master&label=experimental&aligned&view=subtest&q=window-reuse-in-nested-browsing-contexts.tentative.html) appears to be testing the topic of this issue. I'm trying to make Gecko's about:blank behavior more Web-compatible. The major breakthrough so far has been no longer pursuing for...