idna: prevent allocations (mainly) in Punycode decoding

[djc]

No description provided.

[codecov-commenter]

Codecov Report

Merging #635 into master will increase coverage by 0.21%.
The diff coverage is 80.24%.

@@            Coverage Diff             @@
##           master     #635      +/-   ##
==========================================
+ Coverage   81.09%   81.31%   +0.21%     
==========================================
  Files          22       22              
  Lines        3512     3537      +25     
==========================================
+ Hits         2848     2876      +28     
+ Misses        664      661       -3     

Impacted Files	Coverage Δ
url/src/lib.rs	67.13% <0.00%> (-0.58%)	⬇️
idna/src/uts46.rs	94.87% <75.00%> (+3.13%)	⬆️
idna/tests/unit.rs	100.00% <100.00%> (ø)
url/tests/unit.rs	99.65% <100.00%> (+<0.01%)	⬆️
idna/src/punycode.rs	91.05% <0.00%> (+0.81%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8dedda4...5c91e06. Read the comment docs.

[djc]

@valenting can you review the API introduced in 5c91e06? (Feel free to look at the other stuff too, but it's a bit gnarly and only amounts to a modest performance improvement for more complex invocations of to_unicode().)

[valenting]

@valenting can you review the API introduced in 5c91e06? (Feel free to look at the other stuff too, but it's a bit gnarly and only amounts to a modest performance improvement for more complex invocations of to_unicode().)

I think it's generally OK. As a side note, do we use Idna.normalized for anything? Or can it be just a local buffer?

[djc]

What do you mean by "local buffer"? processing() needs a local buffer. Keeping it in Idna amortizes the cost of the allocation over all calls to Idna::to_ascii() and Idna::to_unicode().

[valenting]

What do you mean by "local buffer"? processing() needs a local buffer. Keeping it in Idna amortizes the cost of the allocation over all calls to Idna::to_ascii() and Idna::to_unicode().

In the sense that it doesn't seem to be used outside of processing(). But maybe I'm missing something.

[djc]

No, that is correct. I just though amortizing the cost of allocating that buffer would be better for performance than allocating it locally for every call to processing().

[bors-servo]

☔ The latest upstream changes (presumably 3cc7e69) made this pull request unmergeable. Please resolve the merge conflicts.

[djc]

Newer version in #645.