docs: update incremental benchmarks (3.9.6) #1032
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -6,6 +6,8 @@ Import resolution: native batch vs JS fallback throughput. | |
|
|
||
| | Version | Engine | Files | Full Build | No-op | 1-File | Resolve (native) | Resolve (JS) | | ||
| |---------|--------|------:|-----------:|------:|-------:|------------------:|-------------:| | ||
| | 3.9.6 | native | 744 | 3.0s | 10ms | 54ms | 7ms ~ | 11ms ↑7% | | ||
| | 3.9.6 | wasm | 744 | 14.0s | 131ms | 62ms | 7ms ~ | 11ms ↑7% | | ||
| | 3.9.4 | native | 668 | 2.1s ↓4% | 9ms ~ | 406ms ↑7% | 6ms ↑7% | 9ms ↓15% | | ||
| | 3.9.4 | wasm | 668 | 7.6s ~ | 19ms ↑6% | 61ms ↓90% | 6ms ↑7% | 9ms ↓15% | | ||
| | 3.9.3 | native | 667 | 2.2s ↓76% | 9ms ↑13% | 380ms ↓32% | 6ms ↓5% | 11ms ↑6% | | ||
|
|
@@ -55,25 +57,91 @@ Import resolution: native batch vs JS fallback throughput. | |
|
|
||
| ### Latest results | ||
|
|
||
| **Version:** 3.9.6 | **Files:** 744 | **Date:** 2026-04-30 | ||
|
Comment on lines
57
to
+60
Contributor
There was a problem hiding this comment.
The
Contributor
There was a problem hiding this comment. Tracked as a generator bug rather than a fix-in-place on this auto-generated file: #1042. The note that PR #1008 added was a manual edit, but |
||
|
|
||
| #### Native (Rust) | ||
|
|
||
| | Metric | Value | | ||
| |--------|------:| | ||
| | Full build | 3.0s | | ||
| | No-op rebuild | 10ms | | ||
| | 1-file rebuild | 54ms | | ||
|
|
||
| #### WASM | ||
|
|
||
| | Metric | Value | | ||
| |--------|------:| | ||
| | Full build | 14.0s | | ||
| | No-op rebuild | 131ms | | ||
| | 1-file rebuild | 62ms | | ||
|
|
||
| #### Import Resolution | ||
|
|
||
| | Metric | Value | | ||
| |--------|------:| | ||
| | Import pairs | 986 | | ||
| | Native batch | 7ms | | ||
| | JS fallback | 11ms | | ||
| | Per-import (native) | 0ms | | ||
| | Per-import (JS) | 0ms | | ||
| | Speedup ratio | 1.6x | | ||
|
|
||
| <!-- INCREMENTAL_BENCHMARK_DATA | ||
| [ | ||
| { | ||
| "version": "3.9.6", | ||
| "date": "2026-04-30", | ||
| "files": 744, | ||
| "wasm": { | ||
| "fullBuildMs": 14036, | ||
| "noopRebuildMs": 131, | ||
| "oneFileRebuildMs": 62, | ||
| "oneFilePhases": { | ||
| "setupMs": 5.1, | ||
| "collectMs": 6.6, | ||
| "detectMs": 10, | ||
| "parseMs": 4.3, | ||
| "insertMs": 0.4, | ||
| "resolveMs": 0.5, | ||
| "edgesMs": 1.7, | ||
| "structureMs": 3.1, | ||
| "rolesMs": 25.4, | ||
| "astMs": 0.6, | ||
| "complexityMs": 0.6, | ||
| "cfgMs": 0.4, | ||
| "dataflowMs": 0.4, | ||
| "finalizeMs": 0.4 | ||
| } | ||
| }, | ||
| "native": { | ||
| "fullBuildMs": 2986, | ||
| "noopRebuildMs": 10, | ||
| "oneFileRebuildMs": 54, | ||
| "oneFilePhases": { | ||
| "setupMs": 2.9, | ||
| "collectMs": 4.6, | ||
| "detectMs": 3, | ||
| "parseMs": 0.3, | ||
| "insertMs": 0.3, | ||
| "resolveMs": 0.4, | ||
| "edgesMs": 5.2, | ||
| "structureMs": 4.6, | ||
| "rolesMs": 24.9, | ||
| "astMs": 0.2, | ||
| "complexityMs": 0, | ||
| "cfgMs": 0, | ||
| "dataflowMs": 0, | ||
| "finalizeMs": 0.5 | ||
| } | ||
| }, | ||
| "resolve": { | ||
| "imports": 986, | ||
| "nativeBatchMs": 6.5, | ||
| "jsFallbackMs": 10.5, | ||
| "perImportNativeMs": 0, | ||
| "perImportJsMs": 0 | ||
| } | ||
| }, | ||
| { | ||
| "version": "3.9.5", | ||
| "date": "2026-04-23", | ||
|
|
||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The WASM engine shows significant regressions vs 3.9.4 (the last version with build data): full build went from 7.6s → 14.0s (~84% slower) and no-op rebuild from 19ms → 131ms (~589% slower). Because 3.9.5 had null build metrics, the generation script produced no
↑/↓annotations for these columns in the 3.9.6 WASM row, making the regressions invisible to casual readers of the table. It may be worth checking whether the generator should fall back to comparing against the nearest version that has build data when the immediately prior version has nulls.There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Tracked as a generator improvement: #1043. The blank trend cells are caused by
findPrevReleasereturning the immediately prior release (3.9.5) which has null build/rebuild metrics, sotrend()short-circuits to an empty string. The follow-up issue covers walking further back per-metric to find the nearest release with a non-null value. Worth flagging that the wasm numbers themselves may be partly environmental — the same investigation noted in #1042 about the 3.9.5 SIGKILLs is the right place to assess whether 3.9.6 wasm is a real regression or runner noise.