Integrate utf8 hPutStr to standard hPutStr #589
Conversation
BebeSparkelSparkel
force-pushed
the
directly-write-utf8
branch
from
47987f8
to
603dffc
Compare
src/Data/Text/IO.hs
Outdated
| wantWritableHandle "hPutStr" h $ \h_ -> do | ||
| bmode <- getSpareBuffer h_ | ||
| return (bmode, haOutputNL h_) | ||
| let isUTF8 = maybe False ((== "UTF-8") . textEncodingName) $ haCodec h_ |
There was a problem hiding this comment.
Let's use textEncodingName GHC.IO.Encoding.utf8 instead of hard-coded "UTF-8".
Do we have any idea of performance impact? Comparing strings is not cheap. Maybe we can use GHC.Exts.unsafePtrEquality# GHC.IO.Encoding.utf8 (haCodec h_). After all, it's just an optimization.
There was a problem hiding this comment.
Thanks for finding this the string comparison was hurting performance a lot when I benchmarked this.
There was a problem hiding this comment.
One problem is it is possible to make your own utf-8 encoding https://hackage.haskell.org/package/base-4.20.0.0/docs/GHC-IO-Encoding.html#t:TextEncoding
There was a problem hiding this comment.
But who does that?
There was a problem hiding this comment.
If you crafted your own Unicode TextEncoding, you would not benefit from optimized code path, but that's fine.
It's actually safer to check GHC.Exts.unsafePtrEquality# GHC.IO.Encoding.utf8 (haCodec h_) than == "UTF-8", because everyone can set textEncodingName to whatever they like.
There was a problem hiding this comment.
I have tried implementing this but I am unfamiliar with pointer equality. If I implemented this correctly, please resolve. Otherwise, I'll try and resolve the errors.
Data.Text.IO.Utf8.hPutStrwas implemented in #503 and has much better performance thanData.Text.IO.hPutStrwhen the encoding is"UTF-8"and the neline isLF, so I added it.Question: Is there a faster way to check for the encoding without checking for string equality?