forked from M-Labs/artiq
compiler: rein in overzealous cast monomorphization.
This commit is contained in:
parent
218720cfa7
commit
f68e4ae519
|
@ -4,7 +4,7 @@ expressions of undetermined integer type to either 32 or 64 bits.
|
|||
"""
|
||||
|
||||
from pythonparser import algorithm, diagnostic
|
||||
from .. import types, builtins
|
||||
from .. import types, builtins, asttyped
|
||||
|
||||
class CastMonomorphizer(algorithm.Visitor):
|
||||
def __init__(self, engine):
|
||||
|
@ -20,5 +20,12 @@ class CastMonomorphizer(algorithm.Visitor):
|
|||
if (not types.is_var(typ["width"]) and
|
||||
builtins.is_int(node.args[0].type) and
|
||||
types.is_var(node.args[0].type.find()["width"])):
|
||||
if isinstance(node.args[0], asttyped.BinOpT):
|
||||
# Binary operations are a bit special: they can widen, and so their
|
||||
# return type is indeterminate until both argument types are fully known.
|
||||
# In case we first monomorphize the return type, and then some argument type,
|
||||
# and argument type is wider than return type, we'll introduce a conflict.
|
||||
return
|
||||
|
||||
node.args[0].type.unify(typ)
|
||||
|
||||
|
|
Loading…
Reference in New Issue