We crash on shutdown if we atomize a string containing non-BMP chars, because HashCodeAsUTF8 does: 347 U = PRUint32( (W1&0x03FF)<<10 | (W&0x3FFF) ); which is wrong -- it needs to add 0x10000. This has apparently been wrong as long as the code has been in our tree... While I was here I updated the code to the more recent UTF8 RFC (which only allows sequences of up to 4 bytes).
Created attachment 203149 [details] [diff] [review] Like so
12 years ago
Comment on attachment 203149 [details] [diff] [review] Like so sr=jst