source-academy / py-slang / 29571170123

87%

push

github

Add Module Loader  (#217)

* feat(cse): add data handler

* feat: add module interop

* fix: linting issues

* fix: arity calculation

* feat: add tests

* fix: make Gemini's changes

* fix: allow it to work with cse machine

* fix: use opaqueidentifier in opaquemap

* fix: wrap body in lambda

* fix: Gemini's changes

* fix: ptm(mtp(closure)) returns the same closure

* chore: update module loader package

* Fix module interop list/pair round-tripping and closure identity

pythonToModule's "list" case only ever built a DataType.ARRAY, so a
PAIR returned by one module call (e.g. a binary tree node) came back
as an ARRAY when passed into another module call, breaking anything
checking value.type against PAIR/LIST at every level of a chain.

Fixed by tagging genuine Python list literals at construction time
(InstrType.LIST, and the *args rest-parameter collection in
environment.ts - both build flat, arbitrary-length collections the
same way) in a listLiteralValues WeakSet, rather than trying to guess
from shape alone. Tagged values always reconstruct as a proper
PAIR/EMPTY_LIST chain regardless of length. Untagged values (from
pair()/llist(), or a module PAIR round-tripped through moduleToPython)
reconstruct as a single pair_make(head, tail) without requiring the
chain to terminate in None, since a module PAIR need not be a proper
list (e.g. sound's Sound is (wavesPair, duration), a dotted pair whose
second element is a number, not another list link). This closes the
exact gap raised in review: ptm(mtp(pairObject)) is now a fixpoint
regardless of whether the pair happens to be a proper list.

Also fix PyCseEvaluator's closureMap, which was keyed by the
TypedValue<DataType.CLOSURE> object itself rather than its
ClosureIdentifier. A closure that round-trips through Python gets a
freshly-constructed TypedValue with the same id, so a reference-keyed
map missed it silently (via optional chaining), returning undefined
where an AsyncGenerator was expected - su... (continued)

3361 of 4465 branches covered (75.27%)

Branch coverage included in aggregate %.

80 of 146 new or added lines in 10 files covered. (54.79%)

7356 of 8839 relevant lines covered (83.22%)

79349.24 hits per line