Henri Sivonen
Henri Sivonen
It would be nice if cargo fuzz had subcommand to run the most recent failure case in gdb or rr such that one wouldn't need to first copy and paste...
Steps to reproduce: 1) Navigate to https://developer.microsoft.com/en-us/microsoft-edge/testdrive/demos/eme/ in Firefox. 2) In the list of videos on the left, click any video that's not the currently-selected one. Actual results: A message...
Currently, the ASCII acceleration code manually reinterprets slice memory as wider SIMD or ALU types. This code predates the `align_to` and `align_to_mut` methods on slices. This code should be rewritten...
Pernosco helpfully has special knowledge to render `CharacterAndClass` in a more usable way that generic DWARF would allow. Adding comments to warn against breaking this integration.
Add aligned vs. unaligned benches with two slices where index into the second is read from the first
As a follow-up to #1391, it would be relevant to have a benchmark that more closely tests the access pattern that is relevant to `CodePointTrie`: Having two slices where the...
[`norm_bench`](https://github.com/hsivonen/norm_bench) compares the normalization performance of ICU4X against other implementations, including ICU4C. The test strings are relative long (multiple memory pages). To avoid testing `rust_icu` buffer sizing logic, you need...
`CollationElements` should hold the new type instead of holding references to `CollationDataV1`. See #2554.
`Decomposition::delegate_next_no_pending()` checks the character against `decomposition_passthrough_bound`. Measure if this is actually an optimization in the absence of a supplement: I.e. measure if the trie bails out quickly on its own.
#2378 optimizes away trie lookups when NFC passthrough succeeds. However, that's only with contiguous buffers. With the iterator formulation, characters that are equal or above `decomposition_passthrough_bound` but below `composition_passthrough_bound` could...