When a container (list, set, tuple, or dict) literal expression is
used as an argument to an overloaded function it will get repeatedly
typechecked. This becomes particularly problematic when the expression
is somewhat large, as seen in python#9427

To avoid repeated work, add a new cache in ExprChecker, mapping the AST
node to the resolved type of the expression. Right now the cache is
only used in the fast path, although it could conceivably be leveraged
for the slow path as well in a follow-up commit.

To further reduce duplicate work, when the fast-path doesn't work, we
use the cache to make a note of that, to avoid repeatedly attempting to
take the fast path.

Fixes python#9427

JukkaL pushed a commit that referenced this pull request

When a container (list, set, tuple, or dict) literal expression is
used as an argument to an overloaded function it will get repeatedly
typechecked. This becomes particularly problematic when the expression
is somewhat large, as seen in #9427

To avoid repeated work, add a new cache in ExprChecker, mapping the AST
node to the resolved type of the expression. Right now the cache is
only used in the fast path, although it could conceivably be leveraged
for the slow path as well in a follow-up commit.

To further reduce duplicate work, when the fast-path doesn't work, we
use the cache to make a note of that, to avoid repeatedly attempting to
take the fast path.

Fixes #9427

huguesb deleted the pr-fast-container-type-cache branch