Add record/delegate fast path for OP_RT_GET_PROP in VM by frostney · Pull Request #83 · frostney/GocciaScript

frostney · 2026-03-12T15:12:26Z

Summary

Adds record/delegate fast path for `OP_RT_GET_PROP` in `ExecuteRuntimeOp`.
When the receiver is a `TSouffleRecord`, the property is looked up directly on the record and its delegate chain before falling through to `FRuntimeOps.GetProperty`.
Most dynamic property accesses hit records, so this avoids the virtual call + full runtime property resolution for the common case.

Test plan

All JS tests pass in interpreter mode
All JS tests pass in bytecode mode

Made with Cursor

Summary by CodeRabbit

Bug Fixes
- Improved property access resolution: record-backed properties now try direct lookup, then delegation, before falling back to the generic resolver — reducing missed or incorrect property reads in complex object chains.

Dynamic property access via OP_RT_GET_PROP now checks if the receiver is a TSouffleRecord before calling FRuntimeOps.GetProperty. When it is, the record is queried directly, then the delegate chain is walked. This avoids the virtual method call + full runtime GetProperty (type switching, bridge caching) for the common case of accessing properties on objects. Made-with: Cursor

coderabbitai · 2026-03-12T15:12:37Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 743b0ff4-02d7-4966-93a9-d0e0f2f44c85

📥 Commits

Reviewing files that changed from the base of the PR and between 29d7823 and fea3d1e.

📒 Files selected for processing (1)

souffle/Souffle.VM.pas

🚧 Files skipped from review as they are similar to previous changes (1)

souffle/Souffle.VM.pas

📝 Walkthrough

Walkthrough

The VM's property-get opcode (OP_RT_GET_PROP) now tries a TSouffleRecord fast-path: direct Get, then DelegateGet, and only if both fail falls back to FRuntimeOps.GetProperty; non-record bases use the original runtime GetProperty path. (≤50 words)

Changes

Cohort / File(s)	Summary
VM Property Resolution `souffle/Souffle.VM.pas`	Updated `OP_RT_GET_PROP` handling: added local `PropKey`/`PropVal`; if base is a `TSouffleRecord` reference attempt `Get(PropKey)` → on miss try `DelegateGet(PropKey)` → on miss call `FRuntimeOps.GetProperty(PropKey)`. Non-record bases use `FRuntimeOps.GetProperty` with constant key.

Sequence Diagram(s)

sequenceDiagram
    participant VM as VM (ExecuteRuntimeOp)
    participant Record as TSouffleRecord
    participant Delegate as Delegate (owner/parent)
    participant Runtime as FRuntimeOps

    VM->>Record: If base is record -> Get(PropKey)
    alt Get returns value
        Record-->>VM: PropVal (assign)
    else Get fails
        Record->>Delegate: DelegateGet(PropKey)
        alt Delegate returns value
            Delegate-->>VM: PropVal (assign)
        else Delegate fails
            VM->>Runtime: GetProperty(base, PropKey)
            Runtime-->>VM: PropVal (assign or error)
        end
    end
    alt base not a record
        VM->>Runtime: GetProperty(base, PropKey)
        Runtime-->>VM: PropVal (assign or error)
    end

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

Promote 10 hot opcodes into main VM dispatch loop #81: Implements a similar delegation-aware fast-path for VM record/property access (Get then DelegateGet, then fallback).

Poem

🐇 I hop through fields with keen delight,
I check the record, then ask its sight,
If neighbors know where secrets hide,
I borrow answers, swift and wide—
A rabbit's leap finds values right.

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the main change: adding a fast path optimization for OP_RT_GET_PROP that handles records and delegates.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests
Post copyable unit tests in a comment
Commit unit tests in branch opt/rt-get-prop-fastpath

📝 Coding Plan

Generate coding plan for human review comments

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

github-actions · 2026-03-12T15:27:21Z

Suite Timing

Suite	Metric	Interpreted	Bytecode
Tests	Total	3463	3463
Tests	Passed	3422 ✅	3463 ✅
Tests	Skipped	41	0
Tests	Execution	152.8ms	155.6ms
Tests	Engine	311.2ms	642.7ms
Benchmarks	Total	254	254
Benchmarks	Duration	6.80min	11.00min

_{Measured on ubuntu-latest x64.}

github-actions · 2026-03-12T15:27:24Z

Benchmark Results

254 benchmarks

Interpreted: 🟢 10 improved · 🔴 71 regressed · 173 unchanged · avg -3.8%
Bytecode: 🟢 7 improved · 🔴 3 regressed · 244 unchanged · avg +1.1%

arraybuffer.js — Interp: 🔴 10, 4 unch. · avg -9.7% · Bytecode: 14 unch. · avg +2.3%

Benchmark	Interpreted	Δ	Bytecode	Δ
create ArrayBuffer(0)	492,095 → 435,351	🔴 -11.5%	143,762 → 144,356	+0.4%
create ArrayBuffer(64)	484,754 → 426,173	🔴 -12.1%	140,282 → 143,206	+2.1%
create ArrayBuffer(1024)	383,187 → 324,243	🔴 -15.4%	128,217 → 131,444	+2.5%
create ArrayBuffer(8192)	172,803 → 139,487	🔴 -19.3%	80,804 → 80,495	-0.4%
slice full buffer (64 bytes)	576,117 → 526,468	🔴 -8.6%	387,252 → 398,243	+2.8%
slice half buffer (512 of 1024 bytes)	496,403 → 453,432	🔴 -8.7%	336,067 → 349,600	+4.0%
slice with negative indices	501,178 → 458,749	🔴 -8.5%	364,380 → 379,982	+4.3%
slice empty range	564,068 → 524,733	-7.0%	378,338 → 394,582	+4.3%
byteLength access	1,662,153 → 1,584,832	-4.7%	1,202,102 → 1,211,883	+0.8%
Symbol.toStringTag access	1,139,123 → 1,181,262	+3.7%	591,322 → 597,278	+1.0%
ArrayBuffer.isView	809,791 → 773,519	-4.5%	486,692 → 513,016	+5.4%
clone ArrayBuffer(64)	438,815 → 362,905	🔴 -17.3%	327,955 → 330,043	+0.6%
clone ArrayBuffer(1024)	353,812 → 308,248	🔴 -12.9%	252,601 → 255,990	+1.3%
clone ArrayBuffer inside object	299,371 → 271,702	🔴 -9.2%	153,200 → 158,083	+3.2%

arrays.js — Interp: 🔴 8, 11 unch. · avg -6.5% · Bytecode: 🟢 1, 18 unch. · avg +1.4%

Benchmark	Interpreted	Δ	Bytecode	Δ
Array.from length 100	14,896 → 13,965	-6.3%	13,723 → 13,767	+0.3%
Array.from 10 elements	251,673 → 231,444	🔴 -8.0%	160,582 → 165,568	+3.1%
Array.of 10 elements	343,275 → 311,988	🔴 -9.1%	225,478 → 230,390	+2.2%
spread into new array	362,475 → 338,674	-6.6%	683,353 → 692,321	+1.3%
map over 50 elements	30,661 → 28,965	-5.5%	24,025 → 23,852	-0.7%
filter over 50 elements	25,882 → 23,995	🔴 -7.3%	22,606 → 22,946	+1.5%
reduce sum 50 elements	29,268 → 27,743	-5.2%	20,575 → 19,923	-3.2%
forEach over 50 elements	24,438 → 23,281	-4.7%	25,494 → 25,617	+0.5%
find in 50 elements	37,986 → 35,816	-5.7%	30,164 → 30,209	+0.2%
sort 20 elements	13,065 → 12,396	-5.1%	11,949 → 12,028	+0.7%
flat nested array	125,572 → 120,704	-3.9%	361,463 → 391,202	🟢 +8.2%
flatMap	81,912 → 74,426	🔴 -9.1%	246,632 → 252,877	+2.5%
map inside map (5x5)	23,708 → 21,955	🔴 -7.4%	74,949 → 74,915	-0.0%
filter inside map (5x10)	17,491 → 16,143	🔴 -7.7%	14,246 → 14,485	+1.7%
reduce inside map (5x10)	21,125 → 19,662	-6.9%	14,520 → 14,718	+1.4%
forEach inside forEach (5x10)	17,535 → 16,967	-3.2%	16,369 → 16,306	-0.4%
find inside some (10x10)	15,081 → 13,996	🔴 -7.2%	11,208 → 11,511	+2.7%
map+filter chain nested (5x20)	5,895 → 5,511	-6.5%	4,743 → 4,822	+1.7%
reduce flatten (10x5)	42,981 → 39,242	🔴 -8.7%	6,030 → 6,198	+2.8%

async-await.js — Interp: 🔴 5, 1 unch. · avg -10.5% · Bytecode: 6 unch. · avg +2.3%

Benchmark	Interpreted	Δ	Bytecode	Δ
single await	422,158 → 376,590	🔴 -10.8%	281,334 → 288,843	+2.7%
multiple awaits	188,860 → 168,929	🔴 -10.6%	118,125 → 123,450	+4.5%
await non-Promise value	969,161 → 823,831	🔴 -15.0%	1,021,161 → 1,052,886	+3.1%
await with try/catch	409,820 → 364,457	🔴 -11.1%	277,650 → 280,286	+0.9%
await Promise.all	56,353 → 53,099	-5.8%	45,076 → 45,123	+0.1%
nested async function call	206,255 → 186,449	🔴 -9.6%	216,118 → 222,084	+2.8%

classes.js — Interp: 🔴 28, 3 unch. · avg -9.6% · Bytecode: 31 unch. · avg +0.3%

Benchmark	Interpreted	Δ	Bytecode	Δ
simple class new	131,235 → 115,794	🔴 -11.8%	362,637 → 369,962	+2.0%
class with defaults	105,936 → 93,494	🔴 -11.7%	252,253 → 255,632	+1.3%
50 instances via Array.from	6,241 → 5,627	🔴 -9.8%	6,476 → 6,536	+0.9%
instance method call	67,032 → 59,911	🔴 -10.6%	166,091 → 165,263	-0.5%
static method call	103,217 → 92,366	🔴 -10.5%	372,954 → 367,025	-1.6%
single-level inheritance	52,322 → 46,978	🔴 -10.2%	164,815 → 163,886	-0.6%
two-level inheritance	44,569 → 39,102	🔴 -12.3%	133,963 → 134,032	+0.1%
private field access	65,355 → 58,581	🔴 -10.4%	177,870 → 178,067	+0.1%
private methods	71,476 → 64,038	🔴 -10.4%	223,234 → 225,172	+0.9%
getter/setter access	73,748 → 66,257	🔴 -10.2%	175,774 → 173,011	-1.6%
class decorator (identity)	92,228 → 82,972	🔴 -10.0%	54,397 → 54,831	+0.8%
class decorator (wrapping)	53,042 → 47,507	🔴 -10.4%	38,119 → 38,764	+1.7%
identity method decorator	65,671 → 58,533	🔴 -10.9%	45,712 → 46,016	+0.7%
wrapping method decorator	53,385 → 48,580	🔴 -9.0%	39,044 → 39,243	+0.5%
stacked method decorators (x3)	37,639 → 34,466	🔴 -8.4%	28,749 → 28,538	-0.7%
identity field decorator	73,566 → 66,638	🔴 -9.4%	47,227 → 47,434	+0.4%
field initializer decorator	62,354 → 56,870	🔴 -8.8%	42,095 → 42,458	+0.9%
getter decorator (identity)	63,046 → 56,934	🔴 -9.7%	42,305 → 41,752	-1.3%
setter decorator (identity)	53,283 → 48,212	🔴 -9.5%	36,420 → 36,043	-1.0%
static method decorator	68,990 → 62,432	🔴 -9.5%	64,687 → 65,688	+1.5%
static field decorator	80,365 → 72,199	🔴 -10.2%	67,864 → 68,598	+1.1%
private method decorator	52,871 → 48,169	🔴 -8.9%	38,544 → 38,938	+1.0%
private field decorator	59,316 → 52,641	🔴 -11.3%	40,210 → 40,041	-0.4%
plain auto-accessor (no decorator)	100,578 → 90,590	🔴 -9.9%	55,332 → 55,594	+0.5%
auto-accessor with decorator	58,526 → 52,953	🔴 -9.5%	38,223 → 38,455	+0.6%
decorator writing metadata	46,578 → 43,529	-6.5%	43,789 → 44,048	+0.6%
static getter read	114,088 → 104,771	🔴 -8.2%	407,832 → 411,051	+0.8%
static getter/setter pair	84,296 → 79,711	-5.4%	213,805 → 215,527	+0.8%
inherited static getter	66,781 → 60,017	🔴 -10.1%	277,504 → 274,705	-1.0%
inherited static setter	70,670 → 64,724	🔴 -8.4%	214,256 → 213,601	-0.3%
inherited static getter with this binding	58,169 → 54,287	-6.7%	161,630 → 162,653	+0.6%

closures.js — Interp: 🔴 1, 10 unch. · avg -5.0% · Bytecode: 11 unch. · avg +0.1%

Benchmark	Interpreted	Δ	Bytecode	Δ
closure over single variable	136,921 → 128,985	-5.8%	657,175 → 651,873	-0.8%
closure over multiple variables	125,054 → 120,913	-3.3%	398,668 → 403,216	+1.1%
nested closures	132,187 → 127,291	-3.7%	599,708 → 613,650	+2.3%
function as argument	101,742 → 96,629	-5.0%	552,879 → 558,131	+0.9%
function returning function	126,142 → 119,342	-5.4%	635,498 → 633,074	-0.4%
compose two functions	75,577 → 72,791	-3.7%	380,889 → 385,976	+1.3%
fn.call	158,401 → 152,068	-4.0%	142,136 → 143,212	+0.8%
fn.apply	122,126 → 112,584	🔴 -7.8%	99,050 → 97,478	-1.6%
fn.bind	145,195 → 137,356	-5.4%	150,054 → 148,438	-1.1%
recursive sum to 50	12,461 → 11,984	-3.8%	40,195 → 40,237	+0.1%
recursive tree traversal	20,725 → 19,340	-6.7%	64,153 → 63,122	-1.6%

collections.js — Interp: 🔴 1, 11 unch. · avg -4.4% · Bytecode: 12 unch. · avg -0.0%

Benchmark	Interpreted	Δ	Bytecode	Δ
add 50 elements	7,546 → 7,210	-4.5%	5,833 → 5,873	+0.7%
has lookup (50 elements)	95,626 → 91,722	-4.1%	88,429 → 89,050	+0.7%
delete elements	50,385 → 48,596	-3.6%	37,136 → 37,312	+0.5%
forEach iteration	16,307 → 16,045	-1.6%	17,012 → 16,978	-0.2%
spread to array	35,313 → 31,596	🔴 -10.5%	151,615 → 152,576	+0.6%
deduplicate array	44,199 → 42,211	-4.5%	48,059 → 47,927	-0.3%
set 50 entries	5,532 → 5,212	-5.8%	5,912 → 5,908	-0.1%
get lookup (50 entries)	87,572 → 86,284	-1.5%	97,992 → 96,989	-1.0%
has check	133,930 → 130,551	-2.5%	154,798 → 155,170	+0.2%
delete entries	46,213 → 45,305	-2.0%	35,380 → 35,668	+0.8%
forEach iteration	17,213 → 16,018	-6.9%	16,780 → 16,972	+1.1%
keys/values/entries	9,144 → 8,619	-5.7%	23,183 → 22,404	-3.4%

destructuring.js — Interp: 🔴 13, 9 unch. · avg -7.3% · Bytecode: 🟢 2, 20 unch. · avg +1.9%

Benchmark	Interpreted	Δ	Bytecode	Δ
simple array destructuring	446,487 → 409,958	🔴 -8.2%	857,899 → 859,938	+0.2%
with rest element	292,260 → 295,525	+1.1%	584,356 → 594,117	+1.7%
with defaults	449,456 → 418,897	-6.8%	852,064 → 836,241	-1.9%
skip elements	463,134 → 441,446	-4.7%	976,784 → 978,512	+0.2%
nested array destructuring	184,890 → 182,655	-1.2%	440,910 → 447,455	+1.5%
swap variables	540,145 → 514,015	-4.8%	1,196,824 → 1,180,549	-1.4%
simple object destructuring	330,198 → 295,317	🔴 -10.6%	540,408 → 564,480	+4.5%
with defaults	414,566 → 357,751	🔴 -13.7%	327,171 → 316,446	-3.3%
with renaming	351,818 → 318,847	🔴 -9.4%	625,540 → 657,628	+5.1%
nested object destructuring	162,482 → 149,024	🔴 -8.3%	259,676 → 264,521	+1.9%
rest properties	205,511 → 187,522	🔴 -8.8%	231,129 → 238,365	+3.1%
object parameter	103,540 → 92,008	🔴 -11.1%	182,758 → 196,525	🟢 +7.5%
array parameter	136,681 → 125,963	🔴 -7.8%	352,831 → 362,728	+2.8%
mixed destructuring in map	39,583 → 36,558	🔴 -7.6%	34,952 → 37,678	🟢 +7.8%
forEach with array destructuring	71,159 → 67,636	-5.0%	156,944 → 161,160	+2.7%
map with array destructuring	74,170 → 69,844	-5.8%	186,731 → 191,324	+2.5%
filter with array destructuring	76,806 → 72,165	-6.0%	230,868 → 233,247	+1.0%
reduce with array destructuring	82,822 → 75,728	🔴 -8.6%	207,438 → 210,862	+1.7%
map with object destructuring	86,402 → 79,291	🔴 -8.2%	76,951 → 79,640	+3.5%
map with nested destructuring	70,694 → 65,632	🔴 -7.2%	63,858 → 65,267	+2.2%
map with rest in destructuring	42,256 → 39,752	-5.9%	23,406 → 23,417	+0.0%
map with defaults in destructuring	66,872 → 59,416	🔴 -11.1%	38,830 → 38,617	-0.5%

fibonacci.js — Interp: 8 unch. · avg -2.9% · Bytecode: 8 unch. · avg -1.2%

Benchmark	Interpreted	Δ	Bytecode	Δ
recursive fib(15)	346 → 330	-4.5%	1,128 → 1,131	+0.2%
recursive fib(20)	31 → 30	-2.8%	103 → 102	-1.4%
recursive fib(15) typed	334 → 335	+0.3%	1,498 → 1,468	-2.0%
recursive fib(20) typed	31 → 30	-4.0%	136 → 133	-2.1%
iterative fib(20) via reduce	12,878 → 12,318	-4.4%	8,892 → 8,651	-2.7%
iterator fib(20)	9,839 → 9,513	-3.3%	14,981 → 14,741	-1.6%
iterator fib(20) via Iterator.from + take	15,299 → 14,717	-3.8%	16,592 → 16,760	+1.0%
iterator fib(20) last value via reduce	11,534 → 11,495	-0.3%	12,791 → 12,708	-0.6%

for-of.js — Interp: 7 unch. · avg -0.9% · Bytecode: 7 unch. · avg +0.6%

Benchmark	Interpreted	Δ	Bytecode	Δ
for...of with 10-element array	48,998 → 48,160	-1.7%	151,903 → 152,978	+0.7%
for...of with 100-element array	5,599 → 5,522	-1.4%	21,625 → 21,427	-0.9%
for...of with string (10 chars)	35,842 → 35,289	-1.5%	118,969 → 119,964	+0.8%
for...of with Set (10 elements)	49,442 → 49,002	-0.9%	159,569 → 159,285	-0.2%
for...of with Map entries (10 entries)	30,927 → 31,309	+1.2%	47,611 → 48,976	+2.9%
for...of with destructuring	42,121 → 41,725	-0.9%	71,700 → 73,071	+1.9%
for-await-of with sync array	46,490 → 46,094	-0.9%	130,360 → 128,640	-1.3%

iterators.js — Interp: 20 unch. · avg -0.4% · Bytecode: 🔴 1, 19 unch. · avg +0.6%

Benchmark	Interpreted	Δ	Bytecode	Δ
Iterator.from({next}).toArray() — 20 elements	15,399 → 14,561	-5.4%	17,300 → 16,860	-2.5%
Iterator.from({next}).toArray() — 50 elements	6,688 → 6,387	-4.5%	7,662 → 7,630	-0.4%
spread pre-wrapped iterator — 20 elements	11,838 → 11,402	-3.7%	16,355 → 16,179	-1.1%
Iterator.from({next}).forEach — 50 elements	4,647 → 4,649	+0.1%	5,386 → 5,527	+2.6%
Iterator.from({next}).reduce — 50 elements	4,709 → 4,647	-1.3%	5,134 → 5,202	+1.3%
wrap array iterator	187,070 → 177,812	-4.9%	112,725 → 111,022	-1.5%
wrap plain {next()} object	10,562 → 10,464	-0.9%	13,100 → 12,955	-1.1%
map + toArray (50 elements)	4,756 → 4,810	+1.1%	5,666 → 5,933	+4.7%
filter + toArray (50 elements)	4,590 → 4,632	+0.9%	5,423 → 5,760	+6.2%
take(10) + toArray (50 element source)	27,318 → 26,936	-1.4%	29,051 → 29,875	+2.8%
drop(40) + toArray (50 element source)	6,583 → 6,479	-1.6%	8,110 → 8,621	+6.3%
chained map + filter + take (100 element source)	8,533 → 8,549	+0.2%	9,873 → 10,240	+3.7%
some + every (50 elements)	2,651 → 2,681	+1.2%	3,323 → 3,388	+2.0%
find (50 elements)	5,751 → 5,883	+2.3%	7,331 → 7,403	+1.0%
array.values().map().filter().toArray()	9,045 → 9,427	+4.2%	10,247 → 10,107	-1.4%
array.values().take(5).toArray()	220,441 → 216,828	-1.6%	158,739 → 146,106	🔴 -8.0%
array.values().drop(45).toArray()	207,549 → 202,282	-2.5%	145,806 → 147,235	+1.0%
map.entries() chained helpers	11,142 → 11,398	+2.3%	5,489 → 5,473	-0.3%
set.values() chained helpers	18,918 → 19,266	+1.8%	21,622 → 21,147	-2.2%
string iterator map + toArray	13,625 → 14,496	+6.4%	22,437 → 21,974	-2.1%

json.js — Interp: 20 unch. · avg -3.2% · Bytecode: 🟢 1, 19 unch. · avg +1.8%

Benchmark	Interpreted	Δ	Bytecode	Δ
parse simple object	177,679 → 170,462	-4.1%	145,370 → 137,384	-5.5%
parse nested object	110,241 → 108,341	-1.7%	90,988 → 90,135	-0.9%
parse array of objects	58,870 → 56,544	-4.0%	51,028 → 49,920	-2.2%
parse large flat object	50,592 → 49,542	-2.1%	45,953 → 46,015	+0.1%
parse mixed types	74,679 → 71,822	-3.8%	62,390 → 64,611	+3.6%
stringify simple object	206,353 → 195,245	-5.4%	152,952 → 158,988	+3.9%
stringify nested object	112,786 → 112,358	-0.4%	83,777 → 86,463	+3.2%
stringify array of objects	61,107 → 63,500	+3.9%	49,476 → 52,951	🟢 +7.0%
stringify mixed types	87,687 → 87,566	-0.1%	69,177 → 71,915	+4.0%
reviver doubles numbers	47,958 → 45,221	-5.7%	40,536 → 42,749	+5.5%
reviver filters properties	41,905 → 39,138	-6.6%	46,241 → 47,075	+1.8%
reviver on nested object	53,867 → 51,267	-4.8%	52,887 → 53,731	+1.6%
reviver on array	30,594 → 29,664	-3.0%	29,165 → 30,163	+3.4%
replacer function doubles numbers	50,987 → 48,617	-4.6%	52,143 → 52,983	+1.6%
replacer function excludes properties	63,150 → 61,202	-3.1%	62,683 → 63,026	+0.5%
array replacer (allowlist)	116,907 → 115,032	-1.6%	98,501 → 99,913	+1.4%
stringify with 2-space indent	102,055 → 96,979	-5.0%	78,964 → 80,700	+2.2%
stringify with tab indent	103,396 → 97,431	-5.8%	79,607 → 80,721	+1.4%
parse then stringify	56,598 → 54,608	-3.5%	51,102 → 52,261	+2.3%
stringify then parse	33,788 → 32,868	-2.7%	30,921 → 31,315	+1.3%

jsx.jsx — Interp: 🔴 1, 20 unch. · avg -4.2% · Bytecode: 21 unch. · avg +0.9%

Benchmark	Interpreted	Δ	Bytecode	Δ
simple element	220,775 → 206,071	-6.7%	641,003 → 652,304	+1.8%
self-closing element	220,208 → 211,596	-3.9%	676,498 → 685,671	+1.4%
element with string attribute	182,283 → 173,058	-5.1%	466,593 → 473,841	+1.6%
element with multiple attributes	153,235 → 147,813	-3.5%	405,224 → 407,621	+0.6%
element with expression attribute	175,137 → 161,817	🔴 -7.6%	465,319 → 458,293	-1.5%
text child	213,156 → 204,327	-4.1%	652,676 → 657,820	+0.8%
expression child	209,310 → 198,707	-5.1%	637,288 → 647,055	+1.5%
mixed text and expression	193,808 → 195,131	+0.7%	578,274 → 586,608	+1.4%
nested elements (3 levels)	80,476 → 77,859	-3.3%	256,877 → 257,052	+0.1%
sibling children	60,963 → 57,504	-5.7%	191,435 → 191,309	-0.1%
component element	153,252 → 144,861	-5.5%	449,838 → 453,924	+0.9%
component with children	95,518 → 90,588	-5.2%	288,466 → 290,997	+0.9%
dotted component	129,421 → 122,300	-5.5%	331,387 → 338,243	+2.1%
empty fragment	222,277 → 212,719	-4.3%	694,151 → 689,560	-0.7%
fragment with children	59,474 → 57,412	-3.5%	189,235 → 190,711	+0.8%
spread attributes	113,554 → 108,437	-4.5%	108,146 → 111,895	+3.5%
spread with overrides	99,144 → 94,536	-4.6%	80,912 → 83,217	+2.8%
shorthand props	163,041 → 155,509	-4.6%	433,200 → 440,955	+1.8%
nav bar structure	27,754 → 27,326	-1.5%	81,390 → 81,998	+0.7%
card component tree	32,598 → 31,590	-3.1%	89,791 → 88,654	-1.3%
10 list items via Array.from	15,036 → 14,859	-1.2%	24,307 → 24,449	+0.6%

numbers.js — Interp: 🔴 1, 10 unch. · avg -3.3% · Bytecode: 11 unch. · avg +2.1%

Benchmark	Interpreted	Δ	Bytecode	Δ
integer arithmetic	556,156 → 548,742	-1.3%	1,682,106 → 1,734,404	+3.1%
floating point arithmetic	577,798 → 603,746	+4.5%	1,878,489 → 1,907,269	+1.5%
number coercion	192,696 → 189,093	-1.9%	135,889 → 139,217	+2.4%
toFixed	108,923 → 103,390	-5.1%	219,686 → 223,647	+1.8%
toString	162,747 → 159,469	-2.0%	768,063 → 777,885	+1.3%
valueOf	240,331 → 227,507	-5.3%	1,113,307 → 1,132,001	+1.7%
toPrecision	145,637 → 146,976	+0.9%	432,897 → 439,481	+1.5%
Number.isNaN	324,631 → 300,639	🔴 -7.4%	179,758 → 182,602	+1.6%
Number.isFinite	310,003 → 294,126	-5.1%	174,893 → 176,230	+0.8%
Number.isInteger	311,094 → 289,726	-6.9%	189,612 → 193,931	+2.3%
Number.parseInt and parseFloat	255,878 → 238,230	-6.9%	154,053 → 161,214	+4.6%

objects.js — Interp: 🟢 1, 6 unch. · avg +1.9% · Bytecode: 7 unch. · avg +1.0%

Benchmark	Interpreted	Δ	Bytecode	Δ
create simple object	458,224 → 455,556	-0.6%	877,829 → 875,175	-0.3%
create nested object	215,881 → 216,039	+0.1%	386,733 → 397,642	+2.8%
create 50 objects via Array.from	9,405 → 9,168	-2.5%	8,742 → 8,785	+0.5%
property read	595,301 → 618,734	+3.9%	760,264 → 766,326	+0.8%
Object.keys	285,836 → 297,084	+3.9%	204,244 → 207,501	+1.6%
Object.entries	100,962 → 109,389	🟢 +8.3%	63,412 → 63,355	-0.1%
spread operator	185,708 → 186,097	+0.2%	187,965 → 190,933	+1.6%

promises.js — Interp: 🟢 6, 6 unch. · avg +7.7% · Bytecode: 🔴 1, 11 unch. · avg +0.4%

Benchmark	Interpreted	Δ	Bytecode	Δ
Promise.resolve(value)	542,926 → 550,759	+1.4%	342,201 → 351,586	+2.7%
new Promise(resolve => resolve(value))	193,581 → 198,826	+2.7%	162,269 → 170,613	+5.1%
Promise.reject(reason)	531,509 → 546,118	+2.7%	341,315 → 356,980	+4.6%
resolve + then (1 handler)	163,659 → 174,139	+6.4%	158,572 → 163,930	+3.4%
resolve + then chain (3 deep)	60,453 → 68,186	🟢 +12.8%	70,736 → 71,657	+1.3%
resolve + then chain (10 deep)	20,271 → 22,210	🟢 +9.6%	21,791 → 22,124	+1.5%
reject + catch + then	93,755 → 99,934	+6.6%	90,899 → 93,304	+2.6%
resolve + finally + then	81,707 → 85,814	+5.0%	77,007 → 78,311	+1.7%
Promise.all (5 resolved)	29,963 → 33,370	🟢 +11.4%	26,547 → 26,513	-0.1%
Promise.race (5 resolved)	32,068 → 35,583	🟢 +11.0%	29,024 → 28,040	-3.4%
Promise.allSettled (5 mixed)	25,508 → 27,993	🟢 +9.7%	24,161 → 22,831	-5.5%
Promise.any (5 mixed)	29,436 → 33,330	🟢 +13.2%	28,622 → 25,822	🔴 -9.8%

strings.js — Interp: 🟢 1, 10 unch. · avg +4.5% · Bytecode: 🔴 1, 10 unch. · avg -0.7%

Benchmark	Interpreted	Δ	Bytecode	Δ
string concatenation	349,879 → 419,143	🟢 +19.8%	436,360 → 434,372	-0.5%
template literal	635,493 → 643,967	+1.3%	701,546 → 663,104	-5.5%
string repeat	415,382 → 414,143	-0.3%	1,111,534 → 1,068,933	-3.8%
split and join	131,998 → 137,035	+3.8%	328,456 → 330,173	+0.5%
indexOf and includes	170,417 → 171,847	+0.8%	676,560 → 685,238	+1.3%
toUpperCase and toLowerCase	248,071 → 257,166	+3.7%	745,809 → 688,993	🔴 -7.6%
slice and substring	155,056 → 164,410	+6.0%	750,107 → 761,262	+1.5%
trim operations	183,963 → 191,822	+4.3%	857,533 → 863,895	+0.7%
replace and replaceAll	200,776 → 213,577	+6.4%	731,170 → 734,882	+0.5%
startsWith and endsWith	139,376 → 140,143	+0.6%	569,851 → 587,555	+3.1%
padStart and padEnd	197,849 → 203,673	+2.9%	617,887 → 627,593	+1.6%

typed-arrays.js — Interp: 🟢 2, 🔴 3, 17 unch. · avg -0.6% · Bytecode: 🟢 3, 19 unch. · avg +2.9%

Benchmark	Interpreted	Δ	Bytecode	Δ
new Int32Array(0)	311,367 → 322,440	+3.6%	125,858 → 127,724	+1.5%
new Int32Array(100)	290,435 → 296,492	+2.1%	121,025 → 122,399	+1.1%
new Int32Array(1000)	193,011 → 170,754	🔴 -11.5%	66,080 → 65,727	-0.5%
new Float64Array(100)	266,353 → 267,589	+0.5%	107,796 → 109,748	+1.8%
Int32Array.from([...])	189,518 → 183,579	-3.1%	158,303 → 160,557	+1.4%
Int32Array.of(1, 2, 3, 4, 5)	300,735 → 318,979	+6.1%	237,366 → 241,027	+1.5%
sequential write 100 elements	3,470 → 3,660	+5.5%	12,647 → 13,752	🟢 +8.7%
sequential read 100 elements	3,591 → 3,785	+5.4%	10,133 → 10,594	+4.5%
Float64Array write 100 elements	3,384 → 3,363	-0.6%	12,403 → 13,288	🟢 +7.1%
fill(42)	54,442 → 47,850	🔴 -12.1%	46,082 → 46,272	+0.4%
slice()	216,285 → 205,151	-5.1%	186,587 → 188,175	+0.9%
map(x => x * 2)	8,231 → 8,337	+1.3%	7,723 → 7,699	-0.3%
filter(x => x > 50)	8,166 → 8,517	+4.3%	8,140 → 8,162	+0.3%
reduce (sum)	7,766 → 8,057	+3.7%	6,881 → 7,052	+2.5%
sort()	197,987 → 170,107	🔴 -14.1%	156,526 → 157,147	+0.4%
indexOf()	488,404 → 454,773	-6.9%	359,305 → 373,194	+3.9%
reverse()	360,919 → 339,236	-6.0%	276,774 → 278,970	+0.8%
create view over existing buffer	406,268 → 401,764	-1.1%	134,735 → 136,453	+1.3%
subarray()	443,810 → 457,114	+3.0%	348,849 → 363,822	+4.3%
set() from array	628,149 → 603,968	-3.8%	239,539 → 248,334	+3.7%
for-of loop	4,738 → 5,178	🟢 +9.3%	17,626 → 17,800	+1.0%
spread into array	17,235 → 18,549	🟢 +7.6%	46,524 → 54,601	🟢 +17.4%

_{Measured on ubuntu-latest x64. Changes within ±7% are considered insignificant.}

frostney · 2026-03-12T15:29:56Z

TODO: Let's re-review this after #81

- Rewrite optimization-log.md with correct PR numbers for all 22 PRs since #73 (previous version had multiple wrong PR-to-description mappings) - Add missing PRs: #73, #74, #75, #76, #77, #79, #80, #81, #82, #83, #85, #88, #89, plus infrastructure PRs #90, #92, #98 - Fix GC section: Collect-after-each-file is deliberate, not a workaround; any GC revisit needs fresh analysis - Fix minor items: link to actual issues #103 and #104, correct naming convention context (shared units should NOT use Goccia.* prefix) - Create docs/decision-log.md with 1-3 sentence summaries linking to detailed documentation - Link fpc-generics-performance.md and fpc-dispatch-performance.md from AGENTS.md, code-style.md, design-decisions.md, and optimization-log.md Made-with: Cursor

…105) * Convert spike PDFs to Markdown, update VM docs, add optimization log - Convert all 4 PDF spike documents (string performance, hashmap performance, generics performance, dispatch performance) to Markdown for better AI agent parsing - Fix stale TSouffleValue documentation: 26 bytes → 16 bytes, string[23] → string[13] across souffle-vm.md, design-decisions.md - Update test count to 3,501 across all docs (AGENTS.md, README.md, souffle-vm.md, design-decisions.md) - Add svkString to TSouffleValue kinds list in AGENTS.md - Add THashMap optimization details to AGENTS.md component table - Add docs/optimization-log.md with full VM optimization history, experiment results, remaining work, and key insights for fresh context pickup - Add hashmap spike update note documenting TScopeMap removal, TOrderedStringMap refactoring, and THashMap hash optimization Made-with: Cursor * Fix optimization log PR numbers, add decision log, link all spikes - Rewrite optimization-log.md with correct PR numbers for all 22 PRs since #73 (previous version had multiple wrong PR-to-description mappings) - Add missing PRs: #73, #74, #75, #76, #77, #79, #80, #81, #82, #83, #85, #88, #89, plus infrastructure PRs #90, #92, #98 - Fix GC section: Collect-after-each-file is deliberate, not a workaround; any GC revisit needs fresh analysis - Fix minor items: link to actual issues #103 and #104, correct naming convention context (shared units should NOT use Goccia.* prefix) - Create docs/decision-log.md with 1-3 sentence summaries linking to detailed documentation - Link fpc-generics-performance.md and fpc-dispatch-performance.md from AGENTS.md, code-style.md, design-decisions.md, and optimization-log.md Made-with: Cursor * Address PR review comments: fix ratios, bytes/chars, stale refs - Fix generics spike ratio values (arithmetic errors from original PDF) - Change "13 characters" to "13 bytes" in souffle-vm.md (FPC string[N] is byte-counted), add UTF-8 note - Mark TScopeMap as removed/historical in hashmap spike document - Fix Key Insight #4 PR reference (was #88 async/await, not boolean) - Remove stale GOCCIA_EXT_EVAL_CLASS / FPendingClasses / decorator bridge references from AGENTS.md, souffle-vm.md, design-decisions.md (decorators are now compiled natively via BEGIN/APPLY/FINISH opcodes) - Fix "44 second" → "44-second" hyphenation in string spike Made-with: Cursor * Expand TScopeMap history in hashmap spike with full timeline The spike originally recommended TScopeMap for scope bindings based on micro-benchmarks at N≤10. After implementation (PR #66), real-world profiling showed the assumption about scope sizes was incorrect — global and bridged scopes far exceed 20 bindings, causing a 2.7× regression. The update section now tells the complete timeline: original recommendation → implementation → profiling discovery → revert. Made-with: Cursor * Add grill-me interview skill Introduce a new 'grill-me' agent skill that relentlessly interviews users about a plan or design, exploring each branch of the decision tree until a shared understanding is reached and dependencies are resolved. The skill prefers to inspect the codebase when that can answer a question. Updated skills lock to include the new skill.

Mirror PR #83's GET fast path: when the receiver is a TSouffleRecord, call PutChecked() directly instead of bridging to FRuntimeOps.SetProperty(). Non-record types (blueprints, setters, private fields) still fall through to the bridge for full property resolution. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Add record fast path for OP_RT_SET_PROP in VM dispatch Mirror PR #83's GET fast path: when the receiver is a TSouffleRecord, call PutChecked() directly instead of bridging to FRuntimeOps.SetProperty(). Non-record types (blueprints, setters, private fields) still fall through to the bridge for full property resolution. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Fix SET_PROP fast path: skip records with blueprints (setters/frozen) Records with a Blueprint may have setters, non-writable properties, or frozen/sealed state that PutChecked() does not handle. Restrict the fast path to plain records (no blueprint) and fall through to the bridge for class instances. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Revert SET_PROP fast path — PutChecked misses accessor/freeze/seal semantics PutChecked does not handle accessor properties (getters/setters defined via Object.defineProperty), does not throw on non-writable/frozen/sealed violations, and does not cover Object.preventExtensions. The blueprint check alone was insufficient — plain objects modified with defineProperty, freeze, or seal also break. OP_RT_SET_PROP is only emitted for destructuring patterns, not a hot enough path to warrant a partial fast path that breaks correctness. Revert to always delegating to FRuntimeOps.SetProperty which handles the full property descriptor protocol. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Correct SET_PROP fast path: guard all property descriptor cases The fast path now only fires when ALL conditions are met: - Receiver is TSouffleRecord (not blueprint, wrapped value, etc.) - Record has no instance setters (HasSetters = false) - Record has no blueprint (no class setter hierarchy) - Key is not a private property (does not start with #) - PutChecked succeeds (property is writable AND record is extensible) If any condition fails, falls through to FRuntimeOps.SetProperty which handles the full property descriptor protocol: setter invocation, blueprint setter hierarchy, private property flags, TypeError on non-writable/frozen/sealed violations. The previous attempts failed because: - v1: No guards at all — broke setters, frozen, sealed, defineProperty - v2: Blueprint-only guard — missed Object.defineProperty accessor properties, Object.freeze, Object.seal on plain records - v3 (reverted): Gave up entirely This version correctly handles all cases by checking HasSetters, Blueprint, private keys, and PutChecked return value before committing to the fast path. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Handle full record SetProperty protocol natively in VM Instead of a partial fast path that falls through to the bridge, handle the complete TSouffleRecord SetProperty protocol in the VM dispatch: 1. Instance setters: check Rec.HasSetters, look up Rec.Setters.Get(key), invoke the setter closure via ExecuteFunction — all Souffle-native 2. Blueprint setter hierarchy: walk Bp.Setters chain via new SetPropertyViaBlueprintSetter helper — all Souffle-native 3. Normal put: Rec.PutChecked handles writable/extensible checks 4. Write violation: on PutChecked failure, call FRuntimeOps.PropertyWriteViolation for the language-level TypeError Add PropertyWriteViolation as a virtual method on the base TSouffleRuntimeOperations contract (default no-op) so the VM can request the error without knowing about Goccia types. The bridge (FRuntimeOps.SetProperty) is now only reached for non-record types (TSouffleBlueprint static fields, TGocciaWrappedValue). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

frostney added the implement-feedback label Mar 12, 2026

frostney added 2 commits March 14, 2026 09:29

Merge branch 'main' into opt/rt-get-prop-fastpath

29d7823

Merge branch 'main' into opt/rt-get-prop-fastpath

fea3d1e

frostney merged commit cd0ce40 into main Mar 15, 2026
9 checks passed

frostney deleted the opt/rt-get-prop-fastpath branch March 15, 2026 22:47

coderabbitai Bot mentioned this pull request Mar 15, 2026

Add primitive fast path for OP_RT_TYPEOF in VM dispatch #77

Closed

2 tasks

frostney mentioned this pull request Mar 23, 2026

Add record fast path for OP_RT_SET_PROP in VM dispatch #108

Merged

2 tasks

frostney added the performance Performance improvement label Apr 9, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add record/delegate fast path for OP_RT_GET_PROP in VM#83

Add record/delegate fast path for OP_RT_GET_PROP in VM#83
frostney merged 3 commits intomainfrom
opt/rt-get-prop-fastpath

frostney commented Mar 12, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented Mar 12, 2026 •

edited

Loading

Walkthrough

Changes

Sequence Diagram(s)

Estimated code review effort

Possibly related PRs

Poem

Uh oh!

github-actions Bot commented Mar 12, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Mar 12, 2026 •

edited

Loading

Uh oh!

frostney commented Mar 12, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

frostney commented Mar 12, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Mar 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Sequence Diagram(s)

Estimated code review effort

Possibly related PRs

Poem

Uh oh!

github-actions Bot commented Mar 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Suite Timing

Uh oh!

github-actions Bot commented Mar 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark Results

Uh oh!

frostney commented Mar 12, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

frostney commented Mar 12, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented Mar 12, 2026 •

edited

Loading

github-actions Bot commented Mar 12, 2026 •

edited

Loading

github-actions Bot commented Mar 12, 2026 •

edited

Loading