Bill Spitzak
Bill Spitzak
I also think any transcoding should be based on the UTF-8 bytes, not on decoding the UTF-8 into Unicode code points. This is so it can deal with invalid UTF-8...
One possibility is to "decode" invalid bytes in the UTF-8 to U+DC80...U+DCFF, which is what Python is doing. You can then apply Punycode to the result. The UTF-8 encoding of...
I'm rather suprised that UTF-16 and UTF-32 are supported. Is this necessary, or can maybe they be deprecated and dropped as well?
It seems like the best approach would be to just leak all of these. Can't be an enormous use of memory. Certainly every version I have seen of this type...
That's what I am suggesting. From the bug report it sounds like there is some code in USD that is throwing away old entries in the table. I recommend instead...