Extracted the duplicated conformance plumbing into a single driver:
- lib/guest/conformance.sx — two helper fns that emit (gc-result NAME P F T)
lines for the bash side to grep: gc-dict-result for runners returning
a {:passed :failed :total} dict, and gc-counters-result for guests that
bump a global pass/fail counter from a test file load.
- lib/guest/conformance.sh — config-driven bash driver. Sources a per-lang
conf, locates sx_server, runs sx_server in either single-session "dict"
mode (one preload + many suite evals) or per-suite "counters" mode
(fresh sx_server per suite, with shared preloads). Aggregates and writes
scoreboard.{json,md} via per-lang emit_scoreboard_* functions.
- Ported lib/prolog/conformance.sh and lib/haskell/conformance.sh down to
one-line wrappers that exec the shared driver against their .conf file.
Verification:
- Prolog: 590/590 — diff vs baseline is timestamp-only.
- Haskell: 156/156 — significantly higher than the 0/18 in baseline. The
old conformance.sh was buggy (its `(ok-len 3 ...)` grep never matched,
defaulting every program to 0 pass / 1 fail). Updated baseline to the
true count; no actual test regressed. Plan baseline cell updated.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Verified all 4 hs-upstream-core/sourceInfo tests now pass without any parser
changes. The parser already had `link-next-cmds` (sets `:next` on each command
in a CommandList when hs-span-mode is true) and `:true-branch` extraction in
`parse-cmd` for if statements; Step 3's `:end`/`:line` token fields were the
only missing pieces.
Probed via sx_eval against the parser:
(hs-line-at "if true\n log 'it was true'\n log 'it was true'"
(list :true-branch :next))
returns " log 'it was true'" — matches the expected upstream behaviour.
Test runner output:
PASS: hs-upstream-core/sourceInfo > debug
PASS: hs-upstream-core/sourceInfo > get line works for statements
PASS: hs-upstream-core/sourceInfo > get source works for expressions
PASS: hs-upstream-core/sourceInfo > get source works for statements
Extend hs-make-token to (type value pos &rest extras) producing dicts
{:pos :end :line :value :type}. End defaults to pos+len(value); line
defaults to 1. Both tokenize loops now track current-line via newline
counting in advance!. hs-emit! and t-emit! pass the right end and
start-line to the constructor; redundant dict-set! after construction
removed.
Mirror copied to shared/static/wasm/sx/hs-tokenizer.sx (byte-identical).
Verify: (hs-make-token "NUMBER" "1" 0) returns
{:pos 0 :end 1 :line 1 :value "1" :type "NUMBER"}.
OCaml suite: 4529 pass, 1339 pre-existing failures (baseline). All
4/4 hs-upstream-core/sourceInfo tests now pass (was 2/4 — closes E38).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
In `resume_vm`'s `restore_reuse`, the saved sp captured by
`call_closure_reuse` was ignored when restoring the caller frame after the
async callback finished. The suspended callee's locals/temps stayed on the
value stack above saved_sp, so subsequent LOCAL_GET/SET in the caller
frame (e.g. letrec sibling bindings waiting on the suspending call) read
stale callee data instead of their own slots. Sibling bindings appeared
nil after a perform/resume cycle on the JIT path used by the WASM
browser kernel.
Fix: after popping the callback result and restoring saved_frames, reset
`vm.sp <- saved_sp` (when sp is above), then push the callback result.
Mirrors the OP_RETURN+sp-reset discipline that sync `call_closure_reuse`
already follows.
New tests in `spec/tests/test-letrec-resume.sx` cover single binding,
sibling bindings, mutual recursion siblings, and nested letrec —
all four pass. Full OCaml run_tests: 4529/5868 (was 4525/5864), zero
regressions.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
In `call_closure_reuse`, the success path used a bare `pop vm` that relied on
OP_RETURN having left the stack at exactly `saved_sp + 1`. When the callee
returns a closure (or hits the bytecode-exhausted fallback path), `vm.sp` can
end up inconsistent with the parent frame's expected layout, corrupting
intermediate values such as parser combinator state in `parse-bind`/`many`/
`seq`.
Fix: read the result at the expected slot, then explicitly reset
`vm.sp <- saved_sp` before returning so the parent frame sees a clean stack
regardless of what the callee left behind.
OCaml run_tests baseline: 4525/5864 unchanged. WASM kernel tests: 24/29
unchanged. No regressions.
Parser, layout, desugar, lazy eval, ADTs, HM inference, typeclasses
(Eq/Ord/Show/Num/Functor/Monad), real IO monad, full Prelude. 775/775
green across 13 program suites.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
String=[Char] via pure-SX views, show, error, numeric tower,
Data.Map, Data.Set, records, IORef, exceptions. Briefing updated
to point at new plan; old phases 1-6 plan untouched.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Rewrote the coroutine implementation to use lib/fiber.sx (make-fiber,
fiber-resume, fiber-done?) instead of eagerly running the proc body and
collecting all yields into a list. Each coroutine is now a live fiber —
calls to the coro command invoke fiber-resume, yield suspends via call/cc.
- make-tcl-interp: remove :coroutines/:in-coro/:coro-yields, add :coro-yield-fn nil
- tcl-cmd-yield: calls :coro-yield-fn (fiber's yield fn) to truly suspend
- tcl-cmd-yieldto: same pattern, yields "" to resumer
- make-coro-cmd: takes fiber (not coro-name), calls fiber-resume on each invoke
- tcl-cmd-coroutine: creates a fiber whose body runs the proc with :coro-yield-fn set
- tcl-call-proc result merge: drop :coro-yields/:coroutines propagation
- test.sh: load lib/fiber.sx before lib/tcl/runtime.sx in epoch 4
All 337/337 tests pass including all 20 coro tests.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
parse-int "2.0" returns nil in SX (strict integer parse); fixed by adding
tcl-num-float? (char scan for ./e/E) and tcl-parse-num (routes to
parse-float when float-shaped). Applied in tcl-apply-binop (all arith +
comparisons), tcl-apply-func (parse-float for all math args), unary minus,
and tcl-expr-parse-power (**). Real sqrt/floor/ceil/round/pow/sin/cos/tan/
exp/log now used instead of integer stubs. Integer division still truncates
when both operands are integer-shaped. 329/329 tests green.
Parser: limit `from SOURCE` to parse-collection/cmp/arith/poss/atom
(stops before parse-logical so `or` is not consumed as binary op),
then collect `or EVENT from SOURCE` pairs via recursive collect-ors!.
Adds :or-sources key to the on-feature parts list.
Compiler: scan-on gains or-sources param (11th); new :or-sources cond
clause extracts the list; terminal `true` branch wraps on-call in
(do on-call (hs-on target event handler) ...) for each extra source.
Test: "can handle an or after a from clause" moved from skip-list to
MANUAL_TEST_BODIES and now passes (1478/1496).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When `from #doesntExist` resolves to nil, hs-on silently skips
listener registration instead of crashing on dom-listen nil.
Removes "can ignore when target doesn't exist" from skip-list.
Also adds host-make-js-thrower native utility (plain JS throwing
function, no K.callFn re-entry) — investigated for the js-exceptions
catch test but that test stays skipped: native JS throws from host
calls escape OCaml WASM try-with guards.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Parser: bracket-open in obj-collect key cond → (computed-key expr).
Compiler: detect computed-key list at object-literal pair key and compile
the inner expression instead of emitting a literal string.
Generator: special case for 'expressions work in object literal field names'
using eval-hs-locals with host-callback so hs-win-call can find the fn.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
JIT compilation on first call to many functions incurs a step cost of
200–600k CEK steps. The 200k default was silently failing ~70 tests
across suites like hs-upstream-default, hs-upstream-on, comparisonOp,
and others that work correctly but need JIT warmup headroom. Raising
to 1M reveals all of these as passing. The hypertrace/repeat-forever
tests that are genuinely unbounded remain in _NO_STEP_LIMIT.
Full suite scan (all ranges) now shows 1475/1496 (21 pre-existing
SKIP/untranslated failures, 0 actual failures).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Parser parses optional deriving clause; only appended to AST when non-empty.
hk-bind-decls! data arm generates dictShow_Con / dictEq_Con per constructor.
hk-binop == and /= now deep-force both sides (SX dict equality is by
reference — two thunks wrapping the same value compared as not-equal without
this). Three token-type fixes in the deriving parser (lparen/rparen/comma,
not "special").
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
