René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Match CPython borrow and fast-local superinstruction selection","articleBody":"Summary\n\n  Many remaining diffs are no longer about control-flow or constants, but about the exact local-variable opcode\n  forms RustPython chooses.\n\n  Evidence\n\n  Recurring mismatches include:\n\n  - `LOAD_FAST` vs `LOAD_FAST_BORROW`\n  - `LOAD_FAST_LOAD_FAST` vs `LOAD_FAST_BORROW_LOAD_FAST_BORROW`\n  - `STORE_FAST` vs `STORE_FAST_STORE_FAST`\n  - `STORE_FAST_LOAD_FAST` where CPython keeps separate instructions\n\n  This pattern appears across:\n\n  - `Lib/test/test_buffer.py`\n  - `Lib/test/test_math.py`\n  - `Lib/test/test_tarfile.py`\n  - `Lib/dataclasses.py`\n  - `Lib/test/test_dis.py`\n\n  Expected direction\n\n  Borrow opcode and fused local opcode selection should follow CPython’s optimizer decisions, not a RustPython-\n  specific heuristic. This should be driven by the same structural conditions CPython uses.\n\n  Likely implementation areas\n\n  - `crates/codegen/src/ir.rs`\n  - local-op fusion and borrow/deopt logic after CFG shaping\n\n  Done when\n\n  Representative functions in the files above stop drifting on borrow/fused local opcodes and match CPython’s\n  chosen opcode forms.","author":{"url":"https://github.com/youknowone","@type":"Person","name":"youknowone"},"datePublished":"2026-04-20T15:27:23.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":2},"url":"https://github.com/27/RustPython/issues/27"}