source-academy / py-slang / 31163103731

86%

push

github

Py2JS: a module-level name falls back to its builtin before assignment (#420)

py-slang#415: `print(1)` followed later by `def print(x): return x`
raised a NameError on the *first* `print(1)` call -- even though it
runs before the redefinition ever executes -- instead of using the
real builtin, as CPython (and the substitution stepper) both do.
CPython resolves a module-level name dynamically at the point of each
read: globals() first, then builtins(), never decided statically
ahead of time.

compileProgram's builtin preamble (compiler.ts) already skips
generating a hoisted `const $name = __py.builtins[name]` binding for
any name the chunk itself assigns anywhere at module level, so a
later read (once that assignment has actually run) correctly sees the
local binding. But an *earlier* read -- before the assignment runs --
had nothing left to fall back to: runtime.ts's gref (REPL mode) and
the inline `programGlobals` read compiler.ts emits (program mode)
both went straight to NameError the moment the chunk's own binding
was still undefined, with no builtin fallback at all.

gref now falls back to `this.builtins[name]`, mirroring CPython's own
lookup order. The program-mode path needed the identical fallback
somewhere else, since its "current value" is a hoisted `let` read
inline rather than an indexed globals-table lookup -- new `pgref`
factors out the shared fallback logic and emitName's programGlobals
branch now calls it instead of nameErr directly.


Claude-Session: https://claude.ai/code/session_019GKx36me2L9FrkDhqvHJ6X

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

4803 of 6008 branches covered (79.94%)

Branch coverage included in aggregate %.

5 of 7 new or added lines in 2 files covered. (71.43%)

10725 of 12002 relevant lines covered (89.36%)

164452.11 hits per line