Improve computation of offset in EscapeUnicode
#31253
Conversation
|
r? @brson (rust_highfive has picked a reviewer for you, use r? to override) |
| // digit should be printed and (which is the same) avoids the | ||
| // (31 - 32) underflow | ||
| let msb = 31 - (c | 1).leading_zeros(); | ||
| let msdigit = msb / 4; |
There was a problem hiding this comment.
This code is pretty confusing (I had to re-interpret it several times), so bear with me…
Should this variable be named nibbles/hex_digits (to-be-output) or something along the lines?
There was a problem hiding this comment.
I am afraid that could be misleading, because it is not the number of (hex)digits, but the index of the most significant one.
I hoped that the parallel with MSB would guide the reader and I thought that that presenting it as number of hex_digits would make them believe that an hex_digits = 0 would not be acceptable, while it is the correct offset for any c between 0 and 0xf.
The `offset` value was computed both in `next` and in `size_hint`; computing it in a single place ensures consistency and makes it easier to apply improvements. The value is now computed as soon as the iterator is constructed. This means that the time to compute it is spent immediately and cannot be avoided, but it also guarantees that it is only spent once.
Instead of iteratively scanning the bits, use `leading_zeros`.
ranma42
force-pushed
the
improve-unicode-iter-offset
branch
from
283f3b8 to
79dfa25
Compare
|
I tried to improve the naming ( |
|
I don't feel qualified to review this. Anybody know who is? @SimonSapin maybe? |
|
I believe this code is equivalent to the one before, and it does look nicer. I haven’t benchmarked it to check if it’s faster. |
|
@bors r+ |
|
📌 Commit 8984242 has been approved by |
Improve computation of offset in `EscapeUnicode` Unify the computation of `offset` and use `leading_zeros` instead of manually scanning the bits. This PR removes some duplicated code and makes it a little simpler . The computation of `offset` is also faster, but it is unlikely to have an impact on actual code. (split from #31049)
Unify the computation of
offsetand useleading_zerosinstead of manually scanning the bits.This PR removes some duplicated code and makes it a little simpler .
The computation of
offsetis also faster, but it is unlikely to have an impact on actual code.(split from #31049)