These are chat archives for dropbox/pyston

24th
Nov 2015
Kevin Modzelewski
@kmod
Nov 24 2015 00:05
what kinds of AST_stmt's don't end up as external calls?
Marius Wachtler
@undingen
Nov 24 2015 00:59
it's more meant in the other direct... an AST_stmt will often produce several calls to external functions (at least in the baseline jit) and this number will be different then the llvm JITed one which may lead to differences if we check after each call site if a signal got executed
Kevin Modzelewski
@kmod
Nov 24 2015 01:01
oh interesting
what's the downside then to checking after the AST_stmt?
Marius Wachtler
@undingen
Nov 24 2015 01:09
I think we would check less often the flag than cpython does
but I don't think / I'm unsure if it matters
maybe I should just add the check at the statement layer so that we have the same behavior between all tiers and if this ever shows up as a problem we can add more checks if neccessary...
Kevin Modzelewski
@kmod
Nov 24 2015 01:19
I think our AST_stmt's correspond pretty closely to CPython's bytecodes
like, any time we do something different it has a good chance of being a bug
(I guess except for callattrs)
Kevin Modzelewski
@kmod
Nov 24 2015 01:30
hmm I guess there are some small differences
like, we won't emit the equivalent of a LOAD_CONST
I would guess that the differences don't make a difference in what is user-visible at bytecode boundaries, but it's hard to know for sure :/
Marius Wachtler
@undingen
Nov 24 2015 01:35
Yeah I'm probably overracting and there are not many bytecodes which we don't generate a statement for.