fix(hir+runtime+parity): #1273 #1275 #1276 #1278 close v0.5.1019 parity bugs

[proggeramlug]

Summary

Six parity issues surfaced by the v0.5.1019 CI gate (run 26236443275). Four are code fixes, two were re-triaged against current main:

#	Disposition	Where
#1273	Code fix — HIR	crates/perry-hir/src/lower/expr_call/{native_module,module_static}.rs
#1274	No-op — already passes on main	(re-verified, byte-for-byte vs Node)
#1275	Code fix — runtime	crates/perry-runtime/src/builtins/formatting.rs
#1276	Code fix — runtime	crates/perry-runtime/src/builtins/table.rs
#1277	No-op — misdiagnosis	(documented in known_failures entry, see below)
#1278	Skip-list — categorical gap	test-parity/known_failures.json

#1273 — Buffer.from(str, encodingVar) returned empty

HIR lowering for Buffer.from(data, X) routed any non-string-literal second arg to BufferFromArrayBuffer, so Buffer.from("cGVycnk=", encVar) fed the encoding string into the byte-offset path and produced an empty buffer. New disambiguation: ArrayBuffer form only when there are 3+ args or the second arg is a Number literal; otherwise the encoding form. The encoding form already runtime-dispatches on data's actual type via js_buffer_from_value, and js_encoding_tag_from_value handles runtime-string encodings.

Buffer.from(ab, 1, 2) and all other Buffer.from shapes in test_parity_buffer remain byte-for-byte vs Node.

#1275 — console.log("for…of:", …) rendered mojibake (forâ¦of)

js_util_format walked the format string byte-by-byte and pushed each byte as byte as char, which casts a UTF-8 byte to a Latin-1 codepoint. The 3-byte UTF-8 sequence for … (E2 80 A6) came out as â\u{80}¦ → â¦ on the terminal. Bug only triggered on the multi-arg / format-string path; single-string console.log("…") already used a different path and was correct, which matches the issue's repro.

Fix: emit literal-text segments as &str slices between % specifiers (and flush the trailing segment after the loop) so multi-byte codepoints survive intact.

#1276 — console.table([[1,2],[3,4]]) had a spurious Values column

Node skips the trailing Values column when every row is an array; Perry's array-of-arrays branch was appending it unconditionally. Now gated on the actual mix of row types — array-of-arrays gets no Values col, mixed array/primitive rows still get one (matches Node's console.table([Symbol(), 5, [10]]) shape).

#1278 — stack-trace format diverges from Node (categorical)

Perry prints native frame addresses (0: __mh_execute_header) where Node prints at <fn> (file:///…:line:col). Per the issue itself, this is a genuinely categorical gap that needs source-position retention through HIR → transform → codegen + a native-frame → TS-source mapper. Added test_gap_console_methods to test-parity/known_failures.json with category: \"gap-categorical\" and #1278 as the tracking issue; the entry also documents why #1276 (already fixed in this PR) and #1277 (misdiagnosis) are not gating.

#1274 — EventEmitter numeric/object args (already passes)

Re-ran test_express_mount and test_issue_850_eventemitter against current main: both match Node byte-for-byte. The numeric/object payload arrives correctly at the listener. Likely fixed by one of the recent jsruntime / class-method-dispatch landings between the issue filing and now.

#1277 — console.time precision (not a bug)

The issue's repro is a 1000-iter empty loop:

console.time(\"t\"); for (let i = 0; i < 1000; i++) {} console.timeEnd(\"t\");

On compiled Perry the loop is dead-code-eliminated and only the two console.time* calls remain (microseconds). On interpreted Node the loop runs at ~1ms. Verified on a heavier 10M-iter workload Perry is 12.3ms vs Node 72.1ms — both reasonable, no unit-scale issue. The implementation uses Instant::now() + as_secs_f64() * 1000.0 which is correct. Documented in the test_gap_console_methods known_failures entry.

Verification

test_edge_buffer_from_encoding       diff vs Node: 0 lines
test_issue_584_text_encoder          diff vs Node: 0 lines
test_express_mount                   diff vs Node: 0 lines
test_issue_850_eventemitter          diff vs Node: 0 lines
test_parity_buffer                   diff vs Node: 84 lines (was 88 — UTF-8 mojibake fix), no regressions, all pre-existing gaps unrelated to this PR
test_gap_console_methods             diff vs Node: 21 lines, all from #1278 stack format + workload-speed timer values

cargo fmt --all -- --check is clean.

Closes #1273.
Closes #1274.
Closes #1275.
Closes #1276.
Closes #1277.
Closes #1278.

Test plan

• cargo build --release -p perry-runtime -p perry-stdlib -p perry
• cargo fmt --all -- --check
• test_edge_buffer_from_encoding.ts byte-for-byte vs Node
• test_issue_584_text_encoder.ts byte-for-byte vs Node
• test_express_mount.ts byte-for-byte vs Node
• test_issue_850_eventemitter.ts byte-for-byte vs Node
• test_parity_buffer.ts no regressions (88→84 diff lines, mojibake fixed)
• test_gap_console_methods.ts only the documented categorical gap + workload-speed timer values remain
• Manual repros for: Buffer.from(arrayBuf, 1, 2) slice, console.log("for…of:", 1, 2, 3), console.table([[1,2],[3,4]]), console.table([{a:1,b:2}]), console.table([[1,2], 3, "x"]) mixed