13:50 <enebo> lopex: my branch so far is largely zero sum. I have not pushed the bytelists through into the runtime at all

13:50 <enebo> lopex: they are made back into strings

13:51 <enebo> lopex: but there are two ideas we talked about (one of which I think you wanted): 1) ByteList is the common type for all ID lookup (ID from MRI conceptually) 2) We actually just use RubySymbol instances everywhere

13:53 <nirvdrum> enebo: I'd have to double-check, but I think your Symbols suffer from the same incorrect encoding issue ours does.

13:54 <enebo> nirvdrum: this is what started all of this

13:54 <nirvdrum> Ahh.

13:54 <enebo> nirvdrum: so I have a branch which pushes bytelist all the way up to IRBuilder

13:55 <enebo> 9.2 or 10 we will make a big change in how we identify stuff but the real reason is to finally completely fix our bytes being wrong issue

13:56 <enebo> nirvdrum: one reason it is a big change is Java integration and Java native extensions will need to pass in proper byte/symbol for accessing methods/constants as opposed to j.l.String

13:56 <enebo> nirvdrum: in most cases the string will be good enough if you are living in a 7-bit or UTF-8 world

13:57 <nirvdrum> You're going to require an exact match?

13:57 <nirvdrum> Ruby allows compatible encodings to match.

14:09 <enebo> nirvdrum: well it will be the same

14:09 <enebo> nirvdrum: so long as they pass equality test

14:10 <enebo> nirvdrum: object identity will not be used

14:10 <nirvdrum> Okay. It sounds like you have a handle on it.

14:20 <enebo> nirvdrum: most of our object identity stuff was done in Java 1.4 days when it made a significant difference

14:20 <enebo> nirvdrum: in todays JVM it makes zero warm perf difference and not measurable even when cold

14:25 <nirvdrum> enebo: Well, this was less about object identity. It's that ByteList#equals doesn't take its encoding into account. And so if you have an entry in the symbol table, it's always used regardless of the encoding of the lookup object.

14:26 <enebo> nirvdrum: I was just speaking to notion that if it is compatible bytes it will still match

14:26 <nirvdrum> Okay.

14:27 <enebo> nirvdrum: vs us intern'ing strings and using object identity

14:27 <nirvdrum> So, you're removing the explicit interning from the lexer then?

14:41 <enebo> nirvdrum: yeah

14:42 <enebo> nirvdrum: it was expensive too