[Arfrever] has quit [Read error: Connection reset by peer]
[Arfrever] has joined #pypy
nexussfan has quit [Quit: Konversation terminated!]
derpydoo has quit [Quit: derpydoo]
BarrensZeppelin has joined #pypy
<cfbolz>
BarrensZeppelin: I should hopefully time to find give your PRs a proper read this weekend or next week at the latest
<BarrensZeppelin>
Perfect, thanks!
jcea has joined #pypy
<cfbolz>
BarrensZeppelin: fwiw, I like the idea of keeping _pypy_typing.py around. it's reasonable looking python code, and we have precedence for keeping such things in pure python (eg generic aliases)
<BarrensZeppelin>
Right. Then we'd tolerate differences that we cannot fix in pure python (immutability & introspection things)?
<BarrensZeppelin>
Like: In CPython, type(TypeVar.__bound__) is <class 'getset_descriptor'>, but in PyPy it's currently <class '_pypy_typing._LazyEvaluator'>.
<cfbolz>
BarrensZeppelin: yes, I think we would tolerate these. Porting all this stuff to RPython is a pain, for imo very little advantage
<BarrensZeppelin>
cfbolz: Alright. Then I'll put some work into polishing _pypy_typing. 🙂
<cfbolz>
you could look at _pypy_generic_alias.py for comparison
<cfbolz>
BarrensZeppelin: reading through the code a bit, one of the things I am somewhat uncomfortable about is the patching of extra info into node._type_params_node in symtable.py. how does cpython do this?
<BarrensZeppelin>
Yeah, I also don't like that part. CPython uses the id of AST nodes and type parameter lists as keys.
<BarrensZeppelin>
It's less type safe, and I'm unsure how RPython handles id(...).
<BarrensZeppelin>
Another thing I considered was switching to pairs of (ast node, boolean flag indicating whether the scope is for a type parameter list).
BarrensZeppelin has quit [Remote host closed the connection]
<korvo>
Is there an RPython version of the subprocess module? I didn't find one but maybe I didn't look hard enough.