(Follow-up for bug 1402247.) For conversions like UTF-16 to UTF-8, which expand the number of code units in the non-ASCII case, avoid trying to size the buffer for the ASCII case when: 1) the target string is an autostring and the worst case fits within the inline buffer. 2) when the target string is not an autostring but the input string is 64 or fewer code units in length.

Comment on attachment 8990253 [details] Bug 1473337 - Omit ASCII-case length calculation for short strings when performing an encoding conversion that expands non-ASCII. https://reviewboard.mozilla.org/r/255290/#review264138 ::: servo/support/gecko/nsstring/src/conversions.rs:124 (Diff revision 1) > let mut handle = unsafe { > self.bulk_write(old_len.checked_add(needed).ok_or(())?, old_len, false)? > }; > let written = $convert(other, &mut handle.as_mut_slice()[old_len..]); > - Ok(handle.finish(old_len + written, true)) > + let new_len = old_len + written; > + Ok(handle.finish(new_len, new_len > 64)) is this "64" supposed to be the cache line? can we use a constant here? ::: servo/support/gecko/nsstring/src/conversions.rs:333 (Diff revision 1) > + None > + } > + } else { > + None > + }; > + let (filled, num_ascii, mut handle) = if worst_case_needed.is_none() && long_string_stars_with_basic_latin(other) { nit: long line this might actually be clearer with uninitialized let statements above, and assignments into them rather than the tuple return values...

Comment on attachment 8990253 [details] Bug 1473337 - Omit ASCII-case length calculation for short strings when performing an encoding conversion that expands non-ASCII. https://reviewboard.mozilla.org/r/255290/#review264138 Thanks! > is this "64" supposed to be the cache line? can we use a constant here? Used constant. > nit: long line > > this might actually be clearer with uninitialized let statements above, and assignments into them rather than the tuple return values... Wrapped.

Try run after rebase: https://treeherder.mozilla.org/#/jobs?repo=try&revision=d1d1dcf2071f76b782a792c8e7b0c5de9571ce07

Pushed by hsivonen@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/1e0dffecbfb0 Omit ASCII-case length calculation for short strings when performing an encoding conversion that expands non-ASCII. r=Nika