runtime/object-model: implement strict and sloppy function this binding

[andrewtdiz]

Summary

Sloppy calls box primitive/nullish this as required and strict calls preserve exact this values; direct and .call forms agree with ECMAScript.

Affected Test262 rows

• Windows: 5 styling FFI stubs awaiting custom paint pass (#185 follow-up) #210 language/function-code/10.4.3-1-1-s.js (runtime-fail: Uncaught exception: bar.call(1) Expected SameValue(«"number"», «"object"») to be true)
• Class method body cannot capture enclosing-function locals #212 language/function-code/10.4.3-1-100gs.js (runtime-fail: Uncaught exception: 'this' had incorrect value!)
• String-array element access: arr[i] SIGSEGVs and indexOf returns -1 #214 language/function-code/10.4.3-1-102gs.js (runtime-fail: Uncaught exception: 'this' had incorrect value!)
• doc-tests: wasm_snippets.ts banner should constrain to // targets: wasm — currently fails to link on every native target #219 language/function-code/10.4.3-1-13-s.js (runtime-fail: Uncaught exception: f() Expected SameValue(«"undefined"», «"undefined"») to be false)
• Windows: whoami crate's GetUserNameExW link fails on every doc-test (missing secur32 import lib) #220 language/function-code/10.4.3-1-13gs.js (runtime-fail: Uncaught exception: 'this' had incorrect value!)
• Codegen: arr[i]= writes silently dropped on module-level const arrays declared empty #221 language/function-code/10.4.3-1-15gs.js (runtime-fail: Uncaught exception: 'this' had incorrect value!)
• Codegen: FFI returns:"i64" treats string-pointer returns as Number; auto-detect to string is name-based #222 language/function-code/10.4.3-1-15-s.js (runtime-fail: Uncaught exception: f() Expected SameValue(«"undefined"», «"undefined"») to be false)
• fs.appendFileSync writes 0 bytes on native target — HIR variant has no LLVM codegen arm #226 language/function-code/10.4.3-1-2-s.js (runtime-fail: Uncaught exception: bar.call("1") Expected SameValue(«"string"», «"object"») to be true)
• perry/updater: bind version into signed payload to prevent old-binary replay #229 language/function-code/10.4.3-1-3-s.js (runtime-fail: Uncaught exception: bar.call() Expected SameValue(«"undefined"», «"object"») to be true)

Perry actual behavior from report

• 4 x runtime-fail: Uncaught exception: 'this' had incorrect value!
• 2 x runtime-fail: Uncaught exception: f() Expected SameValue(«"undefined"», «"undefined"») to be false
• 1 x runtime-fail: Uncaught exception: bar.call("1") Expected SameValue(«"string"», «"object"») to be true
• 1 x runtime-fail: Uncaught exception: bar.call() Expected SameValue(«"undefined"», «"object"») to be true
• 1 x runtime-fail: Uncaught exception: bar.call(1) Expected SameValue(«"number"», «"object"») to be true

Node / ECMAScript expected behavior

Sloppy calls box primitive/nullish this as required and strict calls preserve exact this values; direct and .call forms agree with ECMAScript.

Likely subsystem

function call dispatch / this-binding conversion

Acceptance criteria

• bar.call(1), bar.call("1"), and bar.call() rows return the expected typeof this.
• Generator/strict variants no longer report incorrect this.
• Direct function calls match strict/sloppy this rules.

Validation

Run the Test262 differential against the affected area and confirm these rows leave the gap report:

scripts/test262_subset.py --root vendor/test262 --dir language/function-code --sample-cap 99999 --report /tmp/FUNC_THIS_BINDING.json

This is Perry language/runtime semantics triage, not a Node core API compatibility issue.

[andrewtdiz]

Current main appears to have this strict/sloppy this binding slice covered by the existing focused regression.

Validation:

• base: 555e75715 (current origin/main in my checkout)
• target/release/perry --version -> perry 0.5.1112
• target/release/perry test-files/test_issue_3586_function_this_call.ts compiled successfully
• ./test_issue_3586_function_this_call -> issue-3586-ok

That regression covers the issue's main acceptance cases: strict .call(1) preserves primitive this, sloppy .call(1) boxes, sloppy direct call gets global this, first-class Function.prototype.call/apply, strict/sloppy replacement callback receivers, primitive getter boxing, and Function-constructor sloppy body behavior.

I could not run the exact Test262 command because vendor/test262 is not present in this checkout. Based on the local regression and the closed #3586 epic, this looks stale-open and ready for maintainer closure rather than a new PR.

[proggeramlug]

Verified on current main (macOS arm64): every this-binding behavior this issue lists now matches node --experimental-strip-types byte-for-byte:

form	Perry	Node
sloppy bar.call(1)	object	object
sloppy bar.call("1")	object	object
sloppy bar.call(true)	object	object
sloppy bar.call() / .call(null) / .call(undefined)	object	object
strict sbar.call(1)	number	number
strict sbar.call()	undefined	undefined
direct sloppy f()	object (globalThis)	object
direct strict fs()	undefined	undefined
generator sloppy/strict g.call(5)	object / number	same
bare extracted method bare()	object	object

So sloppy primitive/nullish this boxing and strict exact-this preservation are both implemented across direct, .call, generator, and method-extraction forms. I couldn't run the exact scripts/test262_subset.py differential here (it needs an external --root TC39 corpus checkout, not vendored in-repo), so I'm not closing this — but the documented rows appear resolved. Suggest a differential run to confirm the 10.4.3-1-* rows leave the gap report, then close.

[proggeramlug]

Verified resolved via the actual TC39 test262 differential (corpus pinned @ 4249661388, the repo's own pinned SHA):

python3 scripts/test262_subset.py --root <test262> --dir language/function-code --sample-cap 99999
→ language/function-code  pass=213  diff=0  rt-fail=4  compile-fail=0  parity=98.2%

All 9 rows this issue lists pass — 10.4.3-1-1-s, -2-s, -3-s, -13-s, -15-s, -100gs, -102gs, -13gs, -15gs are present in the corpus and none appear in the result's failure set. The only 4 remaining function-code runtime-fails are a different category (10.4.3-1-64gs/-65gs/-83gs/-84gs — negative tests where Node rejects but Perry runs clean, unrelated to this-binding).

Independently confirmed byte-for-byte against node --experimental-strip-types: sloppy primitive/nullish this boxing (bar.call(1)→object, .call("1"), .call(true), .call()/null/undefined→object) and strict exact-this preservation (.call(1)→number, .call()→undefined), across direct, .call, generator, and extracted-method forms.