This fills the corresponding entries in the cast table with a Dead-Code
type so that when the contract-generation pass calls the contract-def
thunk, it finds that in the table.
* call compute-constraints instead of sc->constraints in get-max-contract-kind
* test cast on an intersection type involving Rec
* remove memory limit on sandboxed-unsafe-ops test
Instead of looking at the size of the type, check to see
how many times each type is referenced in other types.
Only lift a type out as a definition if it reaches a
threshold (currently set to 5 refs).
This reduces the zo size of typed/private/framework-types
by roughly 1MB (more than half).
Also move more of the env code generation into the
init-envs.rkt file itself.
Instead of storing a parameter with a box, use a
module-level variable with a hash table and a separate
queue. This separates memoization and the generation
of definitions, which is cleaner.
Avoids using mzlib/pconvert in favor of a few big match
clauses. This lets us cut out a package dependency and
makes the code easier to understand.
This commit also removes the use of mzlib/pconvert
in the debug printer in favor of just using the type
serialization performed in init-envs.rkt.
In addition, a few optimizations for type serialization
were implemented that cut a few percent off of zo sizes.
Note that this commit regresses for zo sizes for modules
that heavily use GUI classes, but that is fixed in a
future commit.
TR sometimes inlines contracts instead of defining them
separately in order to cooperate with the contract system's
optimizations. In some cases, this caused TR to compile
duplicated contract definitions. This commit eliminates
this inefficiency.
* Give correct type to `in-port` when used with custom reader.
Currently, `in-port` returns `(Sequenceof Any)` unconditionally,
which is correct if the given read function is `read` (default
value). However, `(in-port read-line)`, `(in-port read-char)`,
etc. should have more specific types.
* For `in-port`, strip out EOF from the sequence type.
Correctly restrict the struct predicate's filter type when
a parent struct is mutable but the child is not and they both
have polymorphic type variables.
See the discussion in GH issue #205
since 'restrict' will now create intersections when there is
a complex relationship between the two types, calling it
'intersect' makes a lot more sense.
Adds intersection types as a better way to handle the the case
when restrict cannot structurally intersect two types (e.g. when
you learn within a polymorphic function a variable x of type A
is also an Integer, but we dont know how A relates to Integer).
This allows for non-lossy refinements of type info while typechecking.
This reverts commit 3f889bcf8c.
Unfortunately, this change is not backwards compatible, as seen in
both DrDr failures and errors reported by Ben Greenman.
