source-academy / frontend / 35598234767

46%

Build Type

push

github

Committed by web-flow

Commit Message

fix(cse): render Source's empty list and global frame correctly (#4396)

The CSE snapshot adapter was written against py-slang's shape. js-slang's
Conductor CSE evaluators (source-academy/js-slang#2061) expose two places
where that shape does not fit Source.

1. The empty list. The adapter maps the labels nonetype/none/null to a
   `{ toReplString: () => 'None' }` stand-in, deliberately avoiding the
   ArrayNullUnit visual — right for Python, wrong for Source, where the
   empty list *is* that visual and terminates every box-and-pointer
   diagram. js-slang therefore labels Source's null 'empty_list', which
   this maps to a real JS null.

2. The global frame. py-slang's global frame is genuinely empty (its
   builtins live in nativeStorage), so the adapter injects a
   ':::pre-declared names::' sentinel to stand in for them. js-slang puts
   every builtin in the global environment's own head, so that frame
   arrives full and the sentinel is just a spurious extra row. It is now
   injected only when the frame has no bindings of its own.

Dropping the sentinel exposed a latent bug in Frame: it moves the
sentinel to the front of the global frame's bindings, but when the
sentinel is absent findIndex returns -1 and `splice(-1, 1)` reads that as
"the last entry" — silently promoting an unrelated binding to the front.
Demonstrated by mutation: without the guard, a global frame bound
first/middle/last renders as last/first/middle. This is not specific to
Conductor; js-slang's global frame has never carried the sentinel, so the
legacy Source CSE tab hits it too.

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

Coverage Stats

3174 of 7890 branches covered (40.23%)

Branch coverage included in aggregate %.

5 of 6 new or added lines in 2 files covered. (83.33%)

3 existing lines in 1 file now uncovered.

7179 of 14678 relevant lines covered (48.91%)

31.6 hits per line