Experiment: SOUFFLE_INLINE_STRING_MAX = 13 by frostney · Pull Request #96 · frostney/GocciaScript

frostney · 2026-03-15T23:39:12Z

Experiment

Reduces SOUFFLE_INLINE_STRING_MAX from 23 to 15, shrinking TSouffleValue from 26 bytes to 16 bytes. Strings longer than 15 chars spill to TSouffleHeapString on the GC heap.

Hypothesis: Smaller record size improves cache density in the VM register file (FRegisters), reducing cache misses in the dispatch loop. 15 chars is enough to cover most runtime strings (property names, typeof results, short literals) without heap spill, making this a less aggressive trade-off than the 7-char variant (PR #95).

Size comparison:

Variant	Inline max	Record size	Strings that spill
Current	23 chars	26 bytes	None (all common strings fit)
This PR	13 chars	16 bytes	14-23 char strings
PR #95	7 chars	10 bytes	8-23 char strings (6 test failures)

All tests pass in both modes — no strings in the test suite are in the 16-23 char range that would spill.

Test plan

Interpreter: 3422 passed, 0 failed, 41 skipped
Bytecode: 3463 passed, 0 failed, 0 skipped
CI benchmarks for performance comparison

Made with Cursor

Summary by CodeRabbit

Bug Fixes
- String equality now compares actual string content for heap-backed strings.
- Relational operators (>=, <=) consistently perform lexicographic comparison for string operands.
Optimizations
- Reduced inline string storage threshold so shorter strings are stored inline more often.
Tests
- Added comprehensive string-comparison tests covering lengths, boundaries, Unicode, and edge cases.

Shrinks TSouffleValue record from 26 bytes to 18 bytes. Strings longer than 15 chars spill to TSouffleHeapString. This tests whether the smaller register file footprint improves cache locality without the heap allocation penalty of the more aggressive 7-char variant. All tests pass in both modes. Made-with: Cursor

coderabbitai · 2026-03-15T23:39:22Z

📝 Walkthrough

Walkthrough

Lowers inline string threshold, compares heap-backed strings by value, treats string primitives as non-reference for method invocation, adds explicit string-aware >=/<= comparison, and introduces comprehensive string comparison tests including Unicode and boundary cases.

Changes

Cohort / File(s)	Summary
String Value Handling `souffle/Souffle.Value.pas`	Reduced `SOUFFLE_INLINE_STRING_MAX` from 23 to 13; updated `SouffleValuesEqual` to compare contents when both references are `TSouffleHeapString`, keeping other reference comparisons unchanged.
Runtime Operations `units/Goccia.Runtime.Operations.pas`	Modified `InvokeMethodForPrimitive` condition to treat string primitives as non-reference by adding `SouffleIsStringValue(APrimitive)` to the branch decision.
Evaluator Comparisons `units/Goccia.Evaluator.Comparison.pas`	Added explicit lexicographic handling for `>=` and `<=` when both operands are `TGocciaStringLiteralValue`, bypassing numeric coercion in that branch.
Tests `tests/language/expressions/comparison/string-comparison.js`	Added extensive string comparison tests (equality, inequality, ordering) covering empty, short, medium, long, boundary-length strings and broad Unicode/edge-case scenarios.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~22 minutes

Poem

🐰✨ I nibble bytes where strings grow and play,
From twenty-three down to thirteen I stray,
Heap tales judged by meaning, not by sight,
Primitives hop different in the moonlight,
Unicode blossoms — I thump and say hooray!

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: reducing SOUFFLE_INLINE_STRING_MAX from 23 to 13, which is the core experimental change affecting TSouffleValue size and cache density.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.

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

✨ Finishing Touches

📝 Generate docstrings (stacked PR)
📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

Create PR with unit tests
Post copyable unit tests in a comment
Commit unit tests in branch exp/inline-string-15

📝 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.}

Fix InvokeMethodForPrimitive to treat TSouffleHeapString as a primitive value (matching the existing pattern at line 1461). Without this, toString() returning a string >13 chars becomes a heap string reference, causing a false TypeError. Made-with: Cursor

Same fix as exp/inline-string-7: two TSouffleHeapString objects with identical content must compare by value, not pointer identity. This ensures string equality works correctly when strings spill from inline to heap representation. Made-with: Cursor

github-actions · 2026-03-15T23:54:05Z

Suite Timing

Suite	Metric	Interpreted	Bytecode
Tests	Total	3497	3497
Tests	Passed	3456 ✅	3497 ✅
Tests	Skipped	41	0
Tests	Execution	161.6ms	159.9ms
Tests	Engine	320.4ms	646.5ms
Benchmarks	Total	254	254
Benchmarks	Duration	6.69min	11.72min

_{Measured on ubuntu-latest x64.}

github-actions · 2026-03-15T23:54:08Z

Benchmark Results

254 benchmarks

Interpreted: 🟢 135 improved · 🔴 2 regressed · 117 unchanged · avg +7.8%
Bytecode: 🟢 100 improved · 🔴 11 regressed · 143 unchanged · avg +6.3%

arraybuffer.js — Interp: 🟢 12, 2 unch. · avg +16.9% · Bytecode: 14 unch. · avg +0.5%

Benchmark	Interpreted	Δ	Bytecode	Δ
create ArrayBuffer(0)	422,378 → 495,695	🟢 +17.4%	144,098 → 147,711	+2.5%
create ArrayBuffer(64)	410,258 → 488,391	🟢 +19.0%	140,912 → 144,268	+2.4%
create ArrayBuffer(1024)	309,423 → 387,558	🟢 +25.3%	127,618 → 130,108	+2.0%
create ArrayBuffer(8192)	136,669 → 168,628	🟢 +23.4%	80,623 → 82,172	+1.9%
slice full buffer (64 bytes)	503,510 → 594,726	🟢 +18.1%	395,815 → 384,955	-2.7%
slice half buffer (512 of 1024 bytes)	431,493 → 516,458	🟢 +19.7%	343,411 → 339,914	-1.0%
slice with negative indices	438,625 → 524,440	🟢 +19.6%	374,336 → 368,065	-1.7%
slice empty range	498,505 → 585,757	🟢 +17.5%	396,238 → 383,994	-3.1%
byteLength access	1,583,894 → 1,685,910	+6.4%	1,205,985 → 1,187,366	-1.5%
Symbol.toStringTag access	1,174,935 → 1,143,843	-2.6%	587,177 → 601,616	+2.5%
ArrayBuffer.isView	739,575 → 805,174	🟢 +8.9%	497,837 → 499,686	+0.4%
clone ArrayBuffer(64)	365,854 → 450,085	🟢 +23.0%	328,206 → 334,415	+1.9%
clone ArrayBuffer(1024)	286,172 → 360,540	🟢 +26.0%	251,916 → 255,650	+1.5%
clone ArrayBuffer inside object	263,438 → 303,778	🟢 +15.3%	158,012 → 161,069	+1.9%

arrays.js — Interp: 🟢 17, 2 unch. · avg +9.6% · Bytecode: 🟢 10, 9 unch. · avg +11.8%

Benchmark	Interpreted	Δ	Bytecode	Δ
Array.from length 100	14,008 → 15,322	🟢 +9.4%	13,778 → 15,047	🟢 +9.2%
Array.from 10 elements	230,520 → 258,802	🟢 +12.3%	164,111 → 167,218	+1.9%
Array.of 10 elements	314,696 → 349,116	🟢 +10.9%	231,316 → 231,331	+0.0%
spread into new array	339,606 → 375,921	🟢 +10.7%	685,207 → 890,720	🟢 +30.0%
map over 50 elements	28,972 → 31,061	🟢 +7.2%	24,022 → 25,761	🟢 +7.2%
filter over 50 elements	24,196 → 26,038	🟢 +7.6%	22,819 → 25,467	🟢 +11.6%
reduce sum 50 elements	27,863 → 29,463	+5.7%	20,451 → 21,788	+6.5%
forEach over 50 elements	22,194 → 24,532	🟢 +10.5%	25,617 → 27,855	🟢 +8.7%
find in 50 elements	35,823 → 38,547	🟢 +7.6%	29,857 → 32,937	🟢 +10.3%
sort 20 elements	12,256 → 13,372	🟢 +9.1%	11,959 → 12,451	+4.1%
flat nested array	115,519 → 131,417	🟢 +13.8%	368,586 → 476,570	🟢 +29.3%
flatMap	73,059 → 83,309	🟢 +14.0%	249,237 → 320,452	🟢 +28.6%
map inside map (5x5)	21,786 → 24,091	🟢 +10.6%	74,758 → 107,036	🟢 +43.2%
filter inside map (5x10)	16,212 → 17,891	🟢 +10.4%	14,231 → 15,221	+7.0%
reduce inside map (5x10)	19,648 → 21,297	🟢 +8.4%	14,726 → 15,159	+2.9%
forEach inside forEach (5x10)	16,813 → 17,692	+5.2%	16,335 → 17,171	+5.1%
find inside some (10x10)	13,984 → 15,253	🟢 +9.1%	11,459 → 12,232	+6.7%
map+filter chain nested (5x20)	5,570 → 6,021	🟢 +8.1%	4,740 → 5,112	🟢 +7.8%
reduce flatten (10x5)	39,057 → 43,603	🟢 +11.6%	6,102 → 6,317	+3.5%

async-await.js — Interp: 🟢 6 · avg +14.2% · Bytecode: 6 unch. · avg -0.4%

Benchmark	Interpreted	Δ	Bytecode	Δ
single await	372,912 → 428,075	🟢 +14.8%	290,821 → 284,770	-2.1%
multiple awaits	165,457 → 192,579	🟢 +16.4%	123,008 → 123,960	+0.8%
await non-Promise value	839,954 → 980,423	🟢 +16.7%	1,052,244 → 1,040,683	-1.1%
await with try/catch	365,615 → 418,399	🟢 +14.4%	280,655 → 281,781	+0.4%
await Promise.all	51,648 → 57,571	🟢 +11.5%	44,257 → 43,959	-0.7%
nested async function call	187,672 → 209,270	🟢 +11.5%	218,830 → 218,922	+0.0%

classes.js — Interp: 🟢 31 · avg +13.7% · Bytecode: 🟢 13, 🔴 2, 16 unch. · avg +4.7%

Benchmark	Interpreted	Δ	Bytecode	Δ
simple class new	114,878 → 132,052	🟢 +14.9%	361,703 → 385,852	+6.7%
class with defaults	92,659 → 105,485	🟢 +13.8%	257,876 → 269,994	+4.7%
50 instances via Array.from	5,501 → 6,394	🟢 +16.2%	6,087 → 6,847	🟢 +12.5%
instance method call	58,738 → 66,657	🟢 +13.5%	162,046 → 184,762	🟢 +14.0%
static method call	91,295 → 102,852	🟢 +12.7%	371,143 → 430,330	🟢 +15.9%
single-level inheritance	45,901 → 52,043	🟢 +13.4%	162,296 → 182,210	🟢 +12.3%
two-level inheritance	38,852 → 44,364	🟢 +14.2%	134,392 → 149,493	🟢 +11.2%
private field access	57,921 → 66,388	🟢 +14.6%	175,792 → 196,462	🟢 +11.8%
private methods	63,063 → 71,935	🟢 +14.1%	224,603 → 255,579	🟢 +13.8%
getter/setter access	65,534 → 73,704	🟢 +12.5%	173,780 → 189,951	🟢 +9.3%
class decorator (identity)	80,183 → 93,016	🟢 +16.0%	55,309 → 50,775	🔴 -8.2%
class decorator (wrapping)	46,876 → 53,441	🟢 +14.0%	39,296 → 38,040	-3.2%
identity method decorator	57,213 → 66,528	🟢 +16.3%	46,407 → 46,107	-0.6%
wrapping method decorator	47,390 → 53,796	🟢 +13.5%	39,755 → 42,350	+6.5%
stacked method decorators (x3)	33,607 → 38,077	🟢 +13.3%	28,400 → 26,599	-6.3%
identity field decorator	64,819 → 74,124	🟢 +14.4%	47,799 → 44,367	🔴 -7.2%
field initializer decorator	55,161 → 63,064	🟢 +14.3%	42,884 → 42,121	-1.8%
getter decorator (identity)	55,721 → 63,931	🟢 +14.7%	42,150 → 42,263	+0.3%
setter decorator (identity)	47,292 → 54,182	🟢 +14.6%	36,763 → 36,749	-0.0%
static method decorator	61,162 → 70,266	🟢 +14.9%	65,069 → 66,601	+2.4%
static field decorator	71,250 → 81,215	🟢 +14.0%	67,246 → 69,024	+2.6%
private method decorator	46,999 → 53,199	🟢 +13.2%	38,869 → 38,831	-0.1%
private field decorator	52,279 → 58,964	🟢 +12.8%	39,719 → 39,913	+0.5%
plain auto-accessor (no decorator)	87,053 → 102,515	🟢 +17.8%	56,042 → 55,925	-0.2%
auto-accessor with decorator	51,301 → 59,329	🟢 +15.6%	38,598 → 38,655	+0.1%
decorator writing metadata	42,552 → 47,970	🟢 +12.7%	44,166 → 44,081	-0.2%
static getter read	104,068 → 115,099	🟢 +10.6%	404,586 → 446,227	🟢 +10.3%
static getter/setter pair	78,046 → 84,158	🟢 +7.8%	217,079 → 238,808	🟢 +10.0%
inherited static getter	58,566 → 66,632	🟢 +13.8%	275,484 → 302,856	🟢 +9.9%
inherited static setter	63,266 → 70,937	🟢 +12.1%	215,147 → 235,310	🟢 +9.4%
inherited static getter with this binding	53,637 → 58,709	🟢 +9.5%	163,019 → 178,988	🟢 +9.8%

closures.js — Interp: 🟢 7, 4 unch. · avg +8.9% · Bytecode: 🟢 8, 3 unch. · avg +15.7%

Benchmark	Interpreted	Δ	Bytecode	Δ
closure over single variable	128,529 → 135,857	+5.7%	662,188 → 770,190	🟢 +16.3%
closure over multiple variables	118,737 → 126,486	+6.5%	400,210 → 472,883	🟢 +18.2%
nested closures	122,568 → 134,378	🟢 +9.6%	614,089 → 679,565	🟢 +10.7%
function as argument	95,779 → 102,264	+6.8%	554,512 → 693,080	🟢 +25.0%
function returning function	117,096 → 128,838	🟢 +10.0%	630,173 → 769,350	🟢 +22.1%
compose two functions	70,834 → 76,867	🟢 +8.5%	379,365 → 471,252	🟢 +24.2%
fn.call	149,432 → 163,059	🟢 +9.1%	142,190 → 142,433	+0.2%
fn.apply	109,240 → 124,337	🟢 +13.8%	96,570 → 99,694	+3.2%
fn.bind	135,737 → 148,278	🟢 +9.2%	150,038 → 151,831	+1.2%
recursive sum to 50	11,733 → 12,484	+6.4%	39,642 → 52,603	🟢 +32.7%
recursive tree traversal	19,094 → 21,434	🟢 +12.3%	63,302 → 75,339	🟢 +19.0%

collections.js — Interp: 🟢 6, 6 unch. · avg +7.5% · Bytecode: 12 unch. · avg -0.5%

Benchmark	Interpreted	Δ	Bytecode	Δ
add 50 elements	6,979 → 7,483	🟢 +7.2%	5,896 → 5,810	-1.5%
has lookup (50 elements)	88,888 → 93,217	+4.9%	88,833 → 88,163	-0.8%
delete elements	46,989 → 48,580	+3.4%	37,134 → 36,904	-0.6%
forEach iteration	15,121 → 16,668	🟢 +10.2%	16,671 → 17,397	+4.4%
spread to array	32,561 → 36,226	🟢 +11.3%	153,284 → 155,606	+1.5%
deduplicate array	41,157 → 44,708	🟢 +8.6%	48,442 → 48,998	+1.1%
set 50 entries	5,192 → 5,535	+6.6%	5,920 → 6,040	+2.0%
get lookup (50 entries)	84,648 → 88,743	+4.8%	96,402 → 90,913	-5.7%
has check	127,491 → 134,511	+5.5%	154,199 → 153,284	-0.6%
delete entries	44,837 → 46,635	+4.0%	35,362 → 34,921	-1.2%
forEach iteration	15,377 → 17,282	🟢 +12.4%	17,236 → 17,436	+1.2%
keys/values/entries	8,626 → 9,529	🟢 +10.5%	23,529 → 22,242	-5.5%

destructuring.js — Interp: 🟢 17, 5 unch. · avg +9.7% · Bytecode: 🟢 16, 6 unch. · avg +15.9%

Benchmark	Interpreted	Δ	Bytecode	Δ
simple array destructuring	427,029 → 450,252	+5.4%	822,371 → 1,042,293	🟢 +26.7%
with rest element	282,448 → 309,191	🟢 +9.5%	605,746 → 789,889	🟢 +30.4%
with defaults	432,459 → 455,072	+5.2%	831,995 → 947,858	🟢 +13.9%
skip elements	431,816 → 486,059	🟢 +12.6%	994,974 → 1,204,725	🟢 +21.1%
nested array destructuring	184,055 → 191,997	+4.3%	449,516 → 498,705	🟢 +10.9%
swap variables	495,103 → 556,699	🟢 +12.4%	1,195,361 → 1,426,600	🟢 +19.3%
simple object destructuring	296,889 → 324,749	🟢 +9.4%	566,310 → 625,403	🟢 +10.4%
with defaults	356,972 → 409,935	🟢 +14.8%	319,699 → 322,998	+1.0%
with renaming	299,212 → 357,667	🟢 +19.5%	659,255 → 724,606	🟢 +9.9%
nested object destructuring	149,274 → 159,886	🟢 +7.1%	265,364 → 287,252	🟢 +8.2%
rest properties	179,446 → 209,807	🟢 +16.9%	232,315 → 242,231	+4.3%
object parameter	92,137 → 102,995	🟢 +11.8%	192,241 → 206,621	🟢 +7.5%
array parameter	123,416 → 139,717	🟢 +13.2%	351,450 → 425,591	🟢 +21.1%
mixed destructuring in map	36,241 → 38,946	🟢 +7.5%	35,728 → 39,591	🟢 +10.8%
forEach with array destructuring	66,540 → 71,631	🟢 +7.7%	158,531 → 205,632	🟢 +29.7%
map with array destructuring	69,522 → 74,952	🟢 +7.8%	192,436 → 262,205	🟢 +36.3%
filter with array destructuring	71,697 → 78,372	🟢 +9.3%	230,082 → 300,751	🟢 +30.7%
reduce with array destructuring	78,185 → 84,143	🟢 +7.6%	208,103 → 286,837	🟢 +37.8%
map with object destructuring	79,908 → 86,422	🟢 +8.2%	78,205 → 82,268	+5.2%
map with nested destructuring	65,553 → 70,003	+6.8%	66,175 → 69,773	+5.4%
map with rest in destructuring	39,548 → 42,115	+6.5%	22,879 → 23,958	+4.7%
map with defaults in destructuring	60,733 → 66,196	🟢 +9.0%	38,228 → 40,222	+5.2%

fibonacci.js — Interp: 🟢 2, 6 unch. · avg +5.8% · Bytecode: 🟢 6, 2 unch. · avg +20.4%

Benchmark	Interpreted	Δ	Bytecode	Δ
recursive fib(15)	328 → 341	+3.9%	1,105 → 1,473	🟢 +33.3%
recursive fib(20)	29 → 31	+3.7%	100 → 133	🟢 +33.1%
recursive fib(15) typed	325 → 341	+4.8%	1,475 → 1,966	🟢 +33.3%
recursive fib(20) typed	29 → 30	+3.3%	129 → 178	🟢 +38.0%
iterative fib(20) via reduce	12,427 → 12,828	+3.2%	8,842 → 9,001	+1.8%
iterator fib(20)	9,405 → 10,047	+6.8%	14,284 → 15,957	🟢 +11.7%
iterator fib(20) via Iterator.from + take	14,624 → 16,180	🟢 +10.6%	16,718 → 17,293	+3.4%
iterator fib(20) last value via reduce	11,101 → 12,178	🟢 +9.7%	12,695 → 13,770	🟢 +8.5%

for-of.js — Interp: 7 unch. · avg +3.5% · Bytecode: 🟢 7 · avg +12.3%

Benchmark	Interpreted	Δ	Bytecode	Δ
for...of with 10-element array	48,132 → 50,615	+5.2%	150,308 → 166,799	🟢 +11.0%
for...of with 100-element array	5,435 → 5,616	+3.3%	21,364 → 24,834	🟢 +16.2%
for...of with string (10 chars)	34,038 → 36,191	+6.3%	121,372 → 137,775	🟢 +13.5%
for...of with Set (10 elements)	48,162 → 49,489	+2.8%	156,668 → 175,398	🟢 +12.0%
for...of with Map entries (10 entries)	30,669 → 31,408	+2.4%	52,573 → 57,814	🟢 +10.0%
for...of with destructuring	41,766 → 42,104	+0.8%	75,295 → 83,591	🟢 +11.0%
for-await-of with sync array	45,357 → 46,976	+3.6%	132,217 → 148,546	🟢 +12.3%

iterators.js — Interp: 🟢 3, 17 unch. · avg +4.7% · Bytecode: 🟢 1, 🔴 1, 18 unch. · avg +0.6%

Benchmark	Interpreted	Δ	Bytecode	Δ
Iterator.from({next}).toArray() — 20 elements	15,094 → 15,808	+4.7%	17,702 → 18,011	+1.7%
Iterator.from({next}).toArray() — 50 elements	6,541 → 6,808	+4.1%	7,975 → 8,226	+3.1%
spread pre-wrapped iterator — 20 elements	11,382 → 12,048	+5.9%	16,483 → 17,048	+3.4%
Iterator.from({next}).forEach — 50 elements	4,440 → 4,759	🟢 +7.2%	5,531 → 5,844	+5.6%
Iterator.from({next}).reduce — 50 elements	4,535 → 4,770	+5.2%	5,242 → 5,552	+5.9%
wrap array iterator	174,167 → 188,716	🟢 +8.4%	110,221 → 114,426	+3.8%
wrap plain {next()} object	10,426 → 10,714	+2.8%	11,546 → 12,265	+6.2%
map + toArray (50 elements)	4,587 → 4,870	+6.2%	5,734 → 5,698	-0.6%
filter + toArray (50 elements)	4,450 → 4,708	+5.8%	5,528 → 5,531	+0.1%
take(10) + toArray (50 element source)	26,683 → 27,766	+4.1%	27,503 → 28,116	+2.2%
drop(40) + toArray (50 element source)	6,546 → 6,666	+1.8%	8,349 → 8,059	-3.5%
chained map + filter + take (100 element source)	8,367 → 8,740	+4.5%	9,985 → 9,335	-6.5%
some + every (50 elements)	2,629 → 2,747	+4.5%	3,609 → 3,415	-5.4%
find (50 elements)	5,788 → 5,861	+1.3%	7,199 → 7,055	-2.0%
array.values().map().filter().toArray()	8,869 → 9,547	🟢 +7.6%	10,281 → 10,389	+1.1%
array.values().take(5).toArray()	210,354 → 222,262	+5.7%	163,056 → 151,617	🔴 -7.0%
array.values().drop(45).toArray()	198,854 → 208,419	+4.8%	138,445 → 148,573	🟢 +7.3%
map.entries() chained helpers	10,848 → 11,242	+3.6%	5,375 → 5,511	+2.5%
set.values() chained helpers	18,433 → 19,099	+3.6%	21,409 → 21,488	+0.4%
string iterator map + toArray	13,778 → 14,005	+1.7%	22,753 → 21,478	-5.6%

json.js — Interp: 🟢 4, 16 unch. · avg +5.2% · Bytecode: 🟢 2, 🔴 2, 16 unch. · avg +0.7%

Benchmark	Interpreted	Δ	Bytecode	Δ
parse simple object	165,668 → 176,065	+6.3%	144,409 → 138,039	-4.4%
parse nested object	105,420 → 110,648	+5.0%	93,840 → 84,767	🔴 -9.7%
parse array of objects	56,125 → 59,269	+5.6%	50,075 → 45,491	🔴 -9.2%
parse large flat object	49,880 → 51,013	+2.3%	47,696 → 45,324	-5.0%
parse mixed types	71,224 → 74,564	+4.7%	62,659 → 62,779	+0.2%
stringify simple object	196,868 → 209,814	+6.6%	156,088 → 169,535	🟢 +8.6%
stringify nested object	110,741 → 115,972	+4.7%	85,048 → 93,220	🟢 +9.6%
stringify array of objects	63,478 → 61,315	-3.4%	51,366 → 52,628	+2.5%
stringify mixed types	87,940 → 88,956	+1.2%	72,238 → 76,184	+5.5%
reviver doubles numbers	44,188 → 48,390	🟢 +9.5%	45,662 → 47,190	+3.3%
reviver filters properties	38,253 → 42,327	🟢 +10.6%	47,780 → 47,319	-1.0%
reviver on nested object	49,999 → 54,539	🟢 +9.1%	53,475 → 53,377	-0.2%
reviver on array	29,194 → 31,746	🟢 +8.7%	30,091 → 30,652	+1.9%
replacer function doubles numbers	47,928 → 51,104	+6.6%	54,371 → 55,226	+1.6%
replacer function excludes properties	60,199 → 63,874	+6.1%	62,995 → 65,535	+4.0%
array replacer (allowlist)	113,648 → 117,745	+3.6%	102,654 → 104,809	+2.1%
stringify with 2-space indent	97,115 → 101,236	+4.2%	80,398 → 82,008	+2.0%
stringify with tab indent	98,690 → 101,647	+3.0%	81,741 → 83,939	+2.7%
parse then stringify	54,486 → 57,488	+5.5%	52,315 → 52,022	-0.6%
stringify then parse	32,685 → 34,240	+4.8%	31,174 → 31,048	-0.4%

jsx.jsx — Interp: 🟢 12, 9 unch. · avg +7.7% · Bytecode: 🟢 20, 1 unch. · avg +16.9%

Benchmark	Interpreted	Δ	Bytecode	Δ
simple element	189,974 → 218,641	🟢 +15.1%	657,306 → 788,451	🟢 +20.0%
self-closing element	207,342 → 219,159	+5.7%	695,438 → 807,862	🟢 +16.2%
element with string attribute	165,917 → 181,340	🟢 +9.3%	474,865 → 566,725	🟢 +19.3%
element with multiple attributes	146,012 → 152,663	+4.6%	416,307 → 463,210	🟢 +11.3%
element with expression attribute	161,275 → 173,544	🟢 +7.6%	469,102 → 546,012	🟢 +16.4%
text child	195,737 → 211,761	🟢 +8.2%	673,857 → 776,164	🟢 +15.2%
expression child	192,451 → 208,689	🟢 +8.4%	649,670 → 779,545	🟢 +20.0%
mixed text and expression	180,965 → 196,336	🟢 +8.5%	593,133 → 705,811	🟢 +19.0%
nested elements (3 levels)	72,433 → 81,927	🟢 +13.1%	258,598 → 316,089	🟢 +22.2%
sibling children	55,405 → 60,856	🟢 +9.8%	193,189 → 230,030	🟢 +19.1%
component element	144,549 → 152,586	+5.6%	463,023 → 534,842	🟢 +15.5%
component with children	86,769 → 95,108	🟢 +9.6%	292,589 → 339,679	🟢 +16.1%
dotted component	116,918 → 129,744	🟢 +11.0%	343,189 → 404,398	🟢 +17.8%
empty fragment	213,014 → 219,726	+3.2%	699,081 → 817,489	🟢 +16.9%
fragment with children	54,818 → 59,831	🟢 +9.1%	193,575 → 236,131	🟢 +22.0%
spread attributes	104,170 → 112,404	🟢 +7.9%	110,958 → 119,192	🟢 +7.4%
spread with overrides	92,526 → 98,588	+6.6%	82,203 → 88,525	🟢 +7.7%
shorthand props	154,368 → 165,007	+6.9%	447,119 → 527,627	🟢 +18.0%
nav bar structure	26,473 → 27,495	+3.9%	81,359 → 102,826	🟢 +26.4%
card component tree	31,231 → 32,134	+2.9%	90,796 → 110,187	🟢 +21.4%
10 list items via Array.from	14,204 → 14,997	+5.6%	24,512 → 26,032	+6.2%

numbers.js — Interp: 🟢 8, 3 unch. · avg +8.1% · Bytecode: 🟢 5, 6 unch. · avg +5.3%

Benchmark	Interpreted	Δ	Bytecode	Δ
integer arithmetic	558,441 → 558,232	-0.0%	1,723,543 → 2,082,693	🟢 +20.8%
floating point arithmetic	575,953 → 625,183	🟢 +8.5%	1,882,398 → 2,247,537	🟢 +19.4%
number coercion	173,550 → 195,756	🟢 +12.8%	139,737 → 140,636	+0.6%
toFixed	99,633 → 109,686	🟢 +10.1%	221,505 → 224,952	+1.6%
toString	152,232 → 165,721	🟢 +8.9%	780,604 → 848,595	🟢 +8.7%
valueOf	215,257 → 241,594	🟢 +12.2%	1,127,743 → 1,304,670	🟢 +15.7%
toPrecision	140,677 → 144,353	+2.6%	435,451 → 466,638	🟢 +7.2%
Number.isNaN	294,345 → 327,234	🟢 +11.2%	183,756 → 178,840	-2.7%
Number.isFinite	295,419 → 319,326	🟢 +8.1%	177,626 → 170,078	-4.2%
Number.isInteger	286,457 → 315,888	🟢 +10.3%	193,070 → 187,552	-2.9%
Number.parseInt and parseFloat	238,706 → 248,809	+4.2%	163,785 → 154,939	-5.4%

objects.js — Interp: 🟢 1, 6 unch. · avg +2.1% · Bytecode: 🟢 4, 3 unch. · avg +7.6%

Benchmark	Interpreted	Δ	Bytecode	Δ
create simple object	452,604 → 451,426	-0.3%	907,086 → 1,022,635	🟢 +12.7%
create nested object	207,510 → 224,930	🟢 +8.4%	387,952 → 442,390	🟢 +14.0%
create 50 objects via Array.from	8,894 → 9,353	+5.2%	8,710 → 8,845	+1.5%
property read	617,561 → 590,240	-4.4%	762,539 → 920,498	🟢 +20.7%
Object.keys	285,698 → 298,349	+4.4%	213,029 → 202,011	-5.2%
Object.entries	107,024 → 105,474	-1.4%	64,275 → 64,393	+0.2%
spread operator	180,388 → 185,425	+2.8%	197,494 → 215,202	🟢 +9.0%

promises.js — Interp: 🔴 1, 11 unch. · avg -1.6% · Bytecode: 🟢 1, 🔴 3, 8 unch. · avg -5.2%

Benchmark	Interpreted	Δ	Bytecode	Δ
Promise.resolve(value)	542,391 → 568,046	+4.7%	374,932 → 369,399	-1.5%
new Promise(resolve => resolve(value))	193,092 → 198,537	+2.8%	169,543 → 151,236	🔴 -10.8%
Promise.reject(reason)	536,046 → 571,624	+6.6%	352,162 → 338,547	-3.9%
resolve + then (1 handler)	170,277 → 168,643	-1.0%	161,075 → 150,979	-6.3%
resolve + then chain (3 deep)	63,979 → 61,740	-3.5%	72,068 → 65,810	🔴 -8.7%
resolve + then chain (10 deep)	20,912 → 19,021	🔴 -9.0%	21,457 → 23,297	🟢 +8.6%
reject + catch + then	95,237 → 92,570	-2.8%	98,005 → 84,831	🔴 -13.4%
resolve + finally + then	82,996 → 82,245	-0.9%	77,781 → 72,839	-6.4%
Promise.all (5 resolved)	31,619 → 30,497	-3.5%	25,775 → 24,155	-6.3%
Promise.race (5 resolved)	33,740 → 32,519	-3.6%	28,982 → 27,067	-6.6%
Promise.allSettled (5 mixed)	26,941 → 26,406	-2.0%	22,323 → 21,519	-3.6%
Promise.any (5 mixed)	32,260 → 30,150	-6.5%	26,550 → 25,614	-3.5%

strings.js — Interp: 🟢 3, 🔴 1, 7 unch. · avg +2.7% · Bytecode: 🟢 4, 🔴 3, 4 unch. · avg -1.6%

Benchmark	Interpreted	Δ	Bytecode	Δ
string concatenation	412,491 → 388,539	-5.8%	432,927 → 369,797	🔴 -14.6%
template literal	701,207 → 594,674	🔴 -15.2%	710,564 → 692,997	-2.5%
string repeat	386,916 → 421,495	🟢 +8.9%	1,073,171 → 1,067,603	-0.5%
split and join	130,588 → 132,431	+1.4%	332,060 → 324,855	-2.2%
indexOf and includes	158,883 → 173,588	🟢 +9.3%	679,573 → 738,123	🟢 +8.6%
toUpperCase and toLowerCase	238,073 → 252,860	+6.2%	697,971 → 734,665	+5.3%
slice and substring	152,806 → 160,445	+5.0%	750,533 → 841,177	🟢 +12.1%
trim operations	179,109 → 182,931	+2.1%	858,043 → 658,172	🔴 -23.3%
replace and replaceAll	198,884 → 207,480	+4.3%	735,738 → 620,994	🔴 -15.6%
startsWith and endsWith	128,988 → 140,162	🟢 +8.7%	571,541 → 614,308	🟢 +7.5%
padStart and padEnd	189,025 → 198,306	+4.9%	633,648 → 684,908	🟢 +8.1%

typed-arrays.js — Interp: 🟢 6, 16 unch. · avg +4.6% · Bytecode: 🟢 3, 19 unch. · avg +2.4%

Benchmark	Interpreted	Δ	Bytecode	Δ
new Int32Array(0)	322,635 → 312,067	-3.3%	129,190 → 130,819	+1.3%
new Int32Array(100)	297,419 → 294,693	-0.9%	123,303 → 126,200	+2.3%
new Int32Array(1000)	170,110 → 196,174	🟢 +15.3%	62,356 → 68,743	🟢 +10.2%
new Float64Array(100)	268,456 → 274,221	+2.1%	109,364 → 102,487	-6.3%
Int32Array.from([...])	181,854 → 190,448	+4.7%	159,264 → 154,572	-2.9%
Int32Array.of(1, 2, 3, 4, 5)	317,333 → 318,422	+0.3%	247,535 → 255,899	+3.4%
sequential write 100 elements	3,592 → 3,572	-0.6%	13,984 → 14,337	+2.5%
sequential read 100 elements	3,700 → 3,623	-2.1%	10,688 → 11,373	+6.4%
Float64Array write 100 elements	3,336 → 3,410	+2.2%	13,237 → 13,837	+4.5%
fill(42)	47,545 → 54,427	🟢 +14.5%	46,357 → 43,353	-6.5%
slice()	207,348 → 219,164	+5.7%	190,715 → 186,047	-2.4%
map(x => x * 2)	7,817 → 8,405	🟢 +7.5%	8,101 → 8,458	+4.4%
filter(x => x > 50)	8,108 → 8,465	+4.4%	8,341 → 8,797	+5.5%
reduce (sum)	7,741 → 8,071	+4.3%	7,239 → 7,629	+5.4%
sort()	173,927 → 201,859	🟢 +16.1%	152,200 → 150,610	-1.0%
indexOf()	437,403 → 498,341	🟢 +13.9%	376,510 → 365,880	-2.8%
reverse()	330,780 → 371,013	🟢 +12.2%	279,421 → 276,111	-1.2%
create view over existing buffer	391,926 → 394,395	+0.6%	136,291 → 138,080	+1.3%
subarray()	448,853 → 463,553	+3.3%	369,622 → 359,172	-2.8%
set() from array	581,486 → 617,030	+6.1%	246,892 → 266,486	🟢 +7.9%
for-of loop	5,146 → 5,075	-1.4%	17,546 → 21,115	🟢 +20.3%
spread into array	18,576 → 18,018	-3.0%	55,459 → 56,800	+2.4%

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

Covers ===, !==, <, > across empty, short (1-5 char), medium (6-13 char), long (14+ char) and cross-boundary strings. Includes unicode tests for BMP characters, emoji, CJK, mixed ASCII+unicode, and escape sequences. Tests at the 13-char inline threshold boundary validate correctness when strings transition from inline to heap-allocated representation. Made-with: Cursor

coderabbitai

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@souffle/Souffle.Value.pas`:
- Line 11: SOUFFLE_INLINE_STRING_MAX is set to 13 but the experiment is scoped
to 15; update the constant SOUFFLE_INLINE_STRING_MAX to 15 so string
inlining/spill behavior matches the experimental configuration, ensuring
benchmarks use the intended 15-char threshold.

In `@tests/language/expressions/comparison/string-comparison.js`:
- Line 8: Replace self-comparisons like expect("" === "").toBe(true) with
comparisons between two distinct bindings holding the same value: declare two
separate constants (e.g., const left = ""; const right = ""; ) and update the
assertion to use those bindings (e.g., expect(left === right).toBe(true) or
expect(left).toBe(right)); apply this pattern to all flagged expect(... === ...)
occurrences so each side is a distinct binding rather than the identical
literal/variable.

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 76b81bee-9f5b-4827-8562-c5ec4a86e5ce

📥 Commits

Reviewing files that changed from the base of the PR and between cd0ce40 and e5013a4.

📒 Files selected for processing (3)

souffle/Souffle.Value.pas
tests/language/expressions/comparison/string-comparison.js
units/Goccia.Runtime.Operations.pas

GreaterThanOrEqual and LessThanOrEqual coerced both operands to numbers before checking for string-string comparison. Non-numeric strings like "abc" became NaN, triggering an early false return. Add string-string early exit paths matching the pattern already used in GreaterThan and LessThan. Restore <= and >= assertions in string comparison tests. Made-with: Cursor

…ions Rename describe/test labels that referenced inline strings, heap allocation, and character limit boundaries. Tests are unchanged. Made-with: Cursor

coderabbitai

♻️ Duplicate comments (1)

tests/language/expressions/comparison/string-comparison.js (1)

8-9: ⚠️ Potential issue | 🟠 Major

Resolve Biome noSelfCompare errors by avoiding literal/identical self-comparisons.

These assertions compare the same operand to itself, which trips lint/suspicious/noSelfCompare and can block CI. Rewrite them to compare distinct bindings containing equal values.

Suggested patch pattern

 test("empty string comparisons", () => {
-  expect("" === "").toBe(true);
-  expect("" !== "").toBe(false);
+  const emptyA = "";
+  const emptyB = `${""}`;
+  expect(emptyA === emptyB).toBe(true);
+  expect(emptyA !== emptyB).toBe(false);
   expect("" === "a").toBe(false);
   expect("" !== "a").toBe(true);
   expect("a" === "").toBe(false);
 });

 test("same content same length", () => {
-  expect("abc" === "abc").toBe(true);
+  const abcA = "abc";
+  const abcB = `ab${"c"}`;
+  expect(abcA === abcB).toBe(true);
-  expect("abc" !== "abc").toBe(false);
+  expect(abcA !== abcB).toBe(false);
 });

Also applies to: 16-18, 43-44, 60-61, 77-77, 85-85, 87-87, 158-158, 187-187, 193-193, 204-204, 210-210, 216-216

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@tests/language/expressions/comparison/string-comparison.js` around lines 8 -
9, The failing tests use literal self-comparisons like expect("" ===
"").toBe(true) which trigger the noSelfCompare linter; instead create distinct
bindings and compare them (e.g. const a = ""; const b = ""; then use expect(a
=== b).toBe(true) or expect(a !== b).toBe(false)) so the same logical assertion
is preserved but the operands are different identifiers; update all similar
occurrences (uses of expect(... === ...) and expect(... !== ...) in this file)
accordingly.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Duplicate comments:
In `@tests/language/expressions/comparison/string-comparison.js`:
- Around line 8-9: The failing tests use literal self-comparisons like expect(""
=== "").toBe(true) which trigger the noSelfCompare linter; instead create
distinct bindings and compare them (e.g. const a = ""; const b = ""; then use
expect(a === b).toBe(true) or expect(a !== b).toBe(false)) so the same logical
assertion is preserved but the operands are different identifiers; update all
similar occurrences (uses of expect(... === ...) and expect(... !== ...) in this
file) accordingly.

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: a30f3bc4-6818-42aa-9181-e725a047c634

📥 Commits

Reviewing files that changed from the base of the PR and between e5013a4 and 8e71ebc.

📒 Files selected for processing (2)

tests/language/expressions/comparison/string-comparison.js
units/Goccia.Evaluator.Comparison.pas

Replace identical literal operands on both sides of === / !== / <= / >= with distinct const bindings to satisfy Biome's noSelfCompare rule. Same logical assertions are preserved. Made-with: Cursor

frostney added 2 commits March 15, 2026 23:45

This was referenced Mar 16, 2026

Experiment: SOUFFLE_INLINE_STRING_MAX = 7 #95

Closed

Experiment: 8-byte TSouffleValue with reference pools #97

Closed

coderabbitai Bot reviewed Mar 16, 2026

View reviewed changes

Comment thread souffle/Souffle.Value.pas

Comment thread tests/language/expressions/comparison/string-comparison.js Outdated

frostney added 2 commits March 16, 2026 15:25

Remove VM implementation details from string comparison test descript…

8e71ebc

…ions Rename describe/test labels that referenced inline strings, heap allocation, and character limit boundaries. Tests are unchanged. Made-with: Cursor

coderabbitai Bot reviewed Mar 16, 2026

View reviewed changes

frostney changed the title ~~Experiment: SOUFFLE_INLINE_STRING_MAX = 15~~ Experiment: SOUFFLE_INLINE_STRING_MAX = 13 Mar 16, 2026

Avoid self-comparison in string comparison tests

1a68d14

Replace identical literal operands on both sides of === / !== / <= / >= with distinct const bindings to satisfy Biome's noSelfCompare rule. Same logical assertions are preserved. Made-with: Cursor

frostney merged commit f0257b1 into main Mar 16, 2026
9 checks passed

frostney deleted the exp/inline-string-15 branch March 16, 2026 17:13

frostney mentioned this pull request Mar 18, 2026

Convert spike PDFs to Markdown, update VM docs, add optimization log #105

Merged

4 tasks

frostney added the performance Performance improvement label Apr 9, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Experiment: SOUFFLE_INLINE_STRING_MAX = 13#96

Experiment: SOUFFLE_INLINE_STRING_MAX = 13#96
frostney merged 7 commits into
mainfrom
exp/inline-string-15

frostney commented Mar 15, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot commented Mar 15, 2026 •

edited

Loading

Walkthrough

Changes

Estimated code review effort

Poem

Uh oh!

github-actions Bot commented Mar 15, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Mar 15, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot left a comment

Uh oh!

Uh oh!

Uh oh!

coderabbitai Bot left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

frostney commented Mar 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Experiment

Test plan

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Mar 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Estimated code review effort

Poem

Uh oh!

github-actions Bot commented Mar 15, 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 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark Results

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

frostney commented Mar 15, 2026 •

edited

Loading

coderabbitai Bot commented Mar 15, 2026 •

edited

Loading

github-actions Bot commented Mar 15, 2026 •

edited

Loading

github-actions Bot commented Mar 15, 2026 •

edited

Loading