Repository navigation
Conversation
🎨 Storybook: ✅ Built — View Storybook🎭 Playwright: ✅ 1999 passed, 0 failed📊 Browser Reports
📦 Bundle: 9.35 MB gzipDetailsSummary
Category Glance App Entry Points — 3.71 kBMain entry bundles and manifests
Status: 1 added Graph Workspace — 1.4 MBGraph editor runtime, canvas, workflow orchestration
Status: 3 added Views & Navigation — 124 kBTop-level views, pages, and routed surfaces
Status: 17 added Panels & Settings — 591 kBConfiguration panels, inspectors, and settings screens
Status: 27 added User & Accounts — 27.5 kBAuthentication, profile, and account management bundles
Status: 11 added Editors & Dialogs — 125 kBModals, dialogs, drawers, and in-app editors
Status: 8 added UI Components — 112 kBReusable component library chunks
Status: 15 added Data & Services — 3.5 MBStores, services, APIs, and repositories
Status: 17 added Utilities & Hooks — 549 kBHelpers, composables, and utility bundles
Status: 37 added Vendor & Third-Party — 18.1 MBExternal libraries and shared vendor chunks
Status: 18 added Other — 15.1 MBBundles that do not match a named category
Status: 290 added ⚡ Performance Report
Absolute values
Raw data{
"timestamp": "2026-08-25T06:01:37.059Z",
"gitSha": "2e415978dddd65c2a6358d58bf32ed64253cd8ec",
"branch": "audit/ecs-test-fails",
"measurements": [
{
"name": "canvas-idle",
"durationMs": 2049.111000000011,
"styleRecalcs": 8,
"styleRecalcDurationMs": 7.352999999999998,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 495.409,
"heapDeltaBytes": -8782508,
"heapUsedBytes": 65936568,
"domNodes": -281,
"jsHeapTotalBytes": 3244032,
"scriptDurationMs": 9.404,
"eventListeners": -153,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "canvas-idle",
"durationMs": 2045.8770000000186,
"styleRecalcs": 8,
"styleRecalcDurationMs": 6.621000000000002,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 481.89400000000006,
"heapDeltaBytes": -12271124,
"heapUsedBytes": 61890128,
"domNodes": -281,
"jsHeapTotalBytes": 2719744,
"scriptDurationMs": 8.613000000000003,
"eventListeners": -155,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "canvas-mouse-sweep",
"durationMs": 2372.852000000023,
"styleRecalcs": 77,
"styleRecalcDurationMs": 42.418,
"layouts": 12,
"layoutDurationMs": 3.681,
"taskDurationMs": 1404.3000000000002,
"heapDeltaBytes": 7470008,
"heapUsedBytes": 81797976,
"domNodes": -282,
"jsHeapTotalBytes": 4554752,
"scriptDurationMs": 199.353,
"eventListeners": -155,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "canvas-mouse-sweep",
"durationMs": 2233.8009999999713,
"styleRecalcs": 79,
"styleRecalcDurationMs": 37.938,
"layouts": 12,
"layoutDurationMs": 3.334,
"taskDurationMs": 1218.129,
"heapDeltaBytes": 8678228,
"heapUsedBytes": 83542284,
"domNodes": -282,
"jsHeapTotalBytes": 4816896,
"scriptDurationMs": 169.74399999999997,
"eventListeners": -153,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333335,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "canvas-zoom-sweep",
"durationMs": 1707.0940000000405,
"styleRecalcs": 31,
"styleRecalcDurationMs": 14.815000000000001,
"layouts": 6,
"layoutDurationMs": 0.658,
"taskDurationMs": 343.37199999999996,
"heapDeltaBytes": 4450892,
"heapUsedBytes": 79542412,
"domNodes": 76,
"jsHeapTotalBytes": 5505024,
"scriptDurationMs": 10.684,
"eventListeners": 19,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.670000000000012,
"p95FrameDurationMs": 16.699999999999818
},
{
"name": "canvas-zoom-sweep",
"durationMs": 1715.283999999997,
"styleRecalcs": 30,
"styleRecalcDurationMs": 14.623,
"layouts": 6,
"layoutDurationMs": 0.5720000000000001,
"taskDurationMs": 343.37299999999993,
"heapDeltaBytes": 4441412,
"heapUsedBytes": 79623104,
"domNodes": 75,
"jsHeapTotalBytes": 5767168,
"scriptDurationMs": 10.546999999999997,
"eventListeners": 19,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "dom-widget-clipping",
"durationMs": 983.7390000000141,
"styleRecalcs": 8,
"styleRecalcDurationMs": 4.647000000000002,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 717.0150000000001,
"heapDeltaBytes": 11104724,
"heapUsedBytes": 85922200,
"domNodes": 12,
"jsHeapTotalBytes": 7077888,
"scriptDurationMs": 154.04399999999998,
"eventListeners": 2,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "dom-widget-clipping",
"durationMs": 943.2160000000067,
"styleRecalcs": 8,
"styleRecalcDurationMs": 4.655999999999999,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 700.066,
"heapDeltaBytes": 9444236,
"heapUsedBytes": 82583928,
"domNodes": 12,
"jsHeapTotalBytes": 7864320,
"scriptDurationMs": 148.073,
"eventListeners": 2,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-idle",
"durationMs": 2043.374999999969,
"styleRecalcs": 2,
"styleRecalcDurationMs": 1.471,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 526.053,
"heapDeltaBytes": -20397680,
"heapUsedBytes": 78202040,
"domNodes": -273,
"jsHeapTotalBytes": -1961984,
"scriptDurationMs": 36.85099999999999,
"eventListeners": -183,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "large-graph-idle",
"durationMs": 2032.974999999965,
"styleRecalcs": 2,
"styleRecalcDurationMs": 1.902000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 535.771,
"heapDeltaBytes": -20581880,
"heapUsedBytes": 77799592,
"domNodes": -275,
"jsHeapTotalBytes": -3010560,
"scriptDurationMs": 37.94500000000001,
"eventListeners": -153,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333335,
"p95FrameDurationMs": 16.699999999999818
},
{
"name": "large-graph-pan",
"durationMs": 4157.63800000002,
"styleRecalcs": 63,
"styleRecalcDurationMs": 9.474,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 2981.695,
"heapDeltaBytes": 13993356,
"heapUsedBytes": 80037492,
"domNodes": 2,
"jsHeapTotalBytes": 5206016,
"scriptDurationMs": 1157.346,
"eventListeners": 10,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "large-graph-pan",
"durationMs": 4217.818999999963,
"styleRecalcs": 62,
"styleRecalcDurationMs": 8.750999999999998,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 3020.239,
"heapDeltaBytes": 3751524,
"heapUsedBytes": 70194516,
"domNodes": 0,
"jsHeapTotalBytes": 4718592,
"scriptDurationMs": 1191.891,
"eventListeners": 10,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-zoom",
"durationMs": 4826.885000000004,
"styleRecalcs": 66,
"styleRecalcDurationMs": 11.948999999999998,
"layouts": 60,
"layoutDurationMs": 7.784,
"taskDurationMs": 2947.84,
"heapDeltaBytes": 5492772,
"heapUsedBytes": 77618420,
"domNodes": 24,
"jsHeapTotalBytes": 3932160,
"scriptDurationMs": 1102.61,
"eventListeners": 12,
"totalBlockingTimeMs": 3,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-zoom",
"durationMs": 4678.093999999987,
"styleRecalcs": 66,
"styleRecalcDurationMs": 11.899,
"layouts": 60,
"layoutDurationMs": 7.620000000000001,
"taskDurationMs": 2852.5080000000003,
"heapDeltaBytes": 3159696,
"heapUsedBytes": 75194820,
"domNodes": 24,
"jsHeapTotalBytes": 4718592,
"scriptDurationMs": 1065.391,
"eventListeners": 12,
"totalBlockingTimeMs": 8,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "legacy-node-drag",
"durationMs": 6003.799000000015,
"styleRecalcs": 40,
"styleRecalcDurationMs": 5.1469999999999985,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 4378.874,
"heapDeltaBytes": 24573452,
"heapUsedBytes": 93794748,
"domNodes": 0,
"jsHeapTotalBytes": 6709248,
"scriptDurationMs": 1227.762,
"eventListeners": 196,
"totalBlockingTimeMs": 231,
"frameDurationMs": 16.670000000000012,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "legacy-node-drag",
"durationMs": 6001.71499999999,
"styleRecalcs": 40,
"styleRecalcDurationMs": 4.955000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 4369.59,
"heapDeltaBytes": 24405096,
"heapUsedBytes": 93551868,
"domNodes": 0,
"jsHeapTotalBytes": 5136384,
"scriptDurationMs": 1233.435,
"eventListeners": 196,
"totalBlockingTimeMs": 222,
"frameDurationMs": 16.66333333333338,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "minimap-idle",
"durationMs": 1990.8230000000344,
"styleRecalcs": 1,
"styleRecalcDurationMs": 0.41100000000000164,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 489.1669999999999,
"heapDeltaBytes": 12354024,
"heapUsedBytes": 78476644,
"domNodes": -275,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 38.059999999999995,
"eventListeners": -155,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "minimap-idle",
"durationMs": 2001.6910000000507,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 483.837,
"heapDeltaBytes": 12629256,
"heapUsedBytes": 79326316,
"domNodes": -279,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 38.370000000000005,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.670000000000012,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "subgraph-dom-widget-clipping",
"durationMs": 718.5210000000097,
"styleRecalcs": 46,
"styleRecalcDurationMs": 9.461,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 500.54200000000003,
"heapDeltaBytes": -6604376,
"heapUsedBytes": 69090580,
"domNodes": 18,
"jsHeapTotalBytes": 6815744,
"scriptDurationMs": 138.993,
"eventListeners": 8,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "subgraph-dom-widget-clipping",
"durationMs": 727.4649999999383,
"styleRecalcs": 46,
"styleRecalcDurationMs": 9.294,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 497.9939999999999,
"heapDeltaBytes": -6086592,
"heapUsedBytes": 67633716,
"domNodes": 18,
"jsHeapTotalBytes": 6291456,
"scriptDurationMs": 137.82599999999996,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-idle",
"durationMs": 2014.8019999999747,
"styleRecalcs": 10,
"styleRecalcDurationMs": 7.2829999999999995,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 480.65400000000005,
"heapDeltaBytes": 1533292,
"heapUsedBytes": 77627024,
"domNodes": -295,
"jsHeapTotalBytes": 3768320,
"scriptDurationMs": 8.049,
"eventListeners": -231,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "subgraph-idle",
"durationMs": 2023.2640000000401,
"styleRecalcs": 10,
"styleRecalcDurationMs": 7.443,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 475.8399999999999,
"heapDeltaBytes": -8797732,
"heapUsedBytes": 67300024,
"domNodes": -294,
"jsHeapTotalBytes": 4816896,
"scriptDurationMs": 8.123000000000001,
"eventListeners": -199,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "subgraph-mouse-sweep",
"durationMs": 1819.008999999994,
"styleRecalcs": 75,
"styleRecalcDurationMs": 32.586999999999996,
"layouts": 16,
"layoutDurationMs": 3.908,
"taskDurationMs": 1065.365,
"heapDeltaBytes": -2730508,
"heapUsedBytes": 72570408,
"domNodes": -298,
"jsHeapTotalBytes": 5341184,
"scriptDurationMs": 113.398,
"eventListeners": -201,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-mouse-sweep",
"durationMs": 1793.2260000000042,
"styleRecalcs": 76,
"styleRecalcDurationMs": 33.599,
"layouts": 16,
"layoutDurationMs": 4.0969999999999995,
"taskDurationMs": 1083.053,
"heapDeltaBytes": -3179576,
"heapUsedBytes": 72490936,
"domNodes": -296,
"jsHeapTotalBytes": 4816896,
"scriptDurationMs": 115.72799999999998,
"eventListeners": -201,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "subgraph-transition-enter",
"durationMs": 1416.8339999999944,
"styleRecalcs": 18,
"styleRecalcDurationMs": 29.356999999999996,
"layouts": 13,
"layoutDurationMs": 12.676,
"taskDurationMs": 913.181,
"heapDeltaBytes": 21810500,
"heapUsedBytes": 104803000,
"domNodes": 13671,
"jsHeapTotalBytes": 10747904,
"scriptDurationMs": 30.38500000000001,
"eventListeners": 2383,
"totalBlockingTimeMs": 125,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "viewport-pan-sweep",
"durationMs": 14298.000000000002,
"styleRecalcs": 245,
"styleRecalcDurationMs": 36.47599999999999,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 10567.526999999998,
"heapDeltaBytes": -26877592,
"heapUsedBytes": 71571760,
"domNodes": -272,
"jsHeapTotalBytes": 3805184,
"scriptDurationMs": 3989.8059999999996,
"eventListeners": -123,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "viewport-pan-sweep",
"durationMs": 14150.918000000047,
"styleRecalcs": 246,
"styleRecalcDurationMs": 37.48,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 10536.432999999999,
"heapDeltaBytes": -26707168,
"heapUsedBytes": 71920848,
"domNodes": -270,
"jsHeapTotalBytes": 3543040,
"scriptDurationMs": 3970.9089999999997,
"eventListeners": -153,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333326,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "vue-large-graph-idle",
"durationMs": 18100.53599999998,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 17125.267,
"heapDeltaBytes": -24650416,
"heapUsedBytes": 180536992,
"domNodes": -6548,
"jsHeapTotalBytes": 864256,
"scriptDurationMs": 338.372,
"eventListeners": -7475,
"totalBlockingTimeMs": 0,
"frameDurationMs": 17.776666666666642,
"p95FrameDurationMs": 16.80000000000291
},
{
"name": "vue-large-graph-idle",
"durationMs": 18006.193999999938,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 17161.273999999998,
"heapDeltaBytes": -30077896,
"heapUsedBytes": 180898008,
"domNodes": -6548,
"jsHeapTotalBytes": 3747840,
"scriptDurationMs": 328.805,
"eventListeners": -7471,
"totalBlockingTimeMs": 0,
"frameDurationMs": 18.333333333333332,
"p95FrameDurationMs": 16.80000000000291
},
{
"name": "vue-large-graph-pan",
"durationMs": 21781.929000000047,
"styleRecalcs": 175,
"styleRecalcDurationMs": 19.88000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 21388.703999999998,
"heapDeltaBytes": -26279292,
"heapUsedBytes": 181373624,
"domNodes": -6548,
"jsHeapTotalBytes": -176128,
"scriptDurationMs": 993.2170000000001,
"eventListeners": -7465,
"totalBlockingTimeMs": 159,
"frameDurationMs": 17.776666666666642,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "vue-large-graph-pan",
"durationMs": 22277.56099999999,
"styleRecalcs": 175,
"styleRecalcDurationMs": 20.121,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 21859.212,
"heapDeltaBytes": -24469952,
"heapUsedBytes": 187676656,
"domNodes": -6548,
"jsHeapTotalBytes": -4882432,
"scriptDurationMs": 1016.726,
"eventListeners": -7469,
"totalBlockingTimeMs": 167,
"frameDurationMs": 17.780000000000047,
"p95FrameDurationMs": 16.80000000000291
},
{
"name": "workflow-execution",
"durationMs": 153.53600000003098,
"styleRecalcs": 12,
"styleRecalcDurationMs": 21.194000000000003,
"layouts": 2,
"layoutDurationMs": 0.7410000000000001,
"taskDurationMs": 104.46099999999998,
"heapDeltaBytes": 3835824,
"heapUsedBytes": 75287620,
"domNodes": 115,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 5.0120000000000005,
"eventListeners": 49,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66999999999998,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "workflow-execution",
"durationMs": 463.4399999999914,
"styleRecalcs": 17,
"styleRecalcDurationMs": 23.953000000000003,
"layouts": 3,
"layoutDurationMs": 1.1430000000000002,
"taskDurationMs": 135.579,
"heapDeltaBytes": 6070500,
"heapUsedBytes": 76654500,
"domNodes": 123,
"jsHeapTotalBytes": 786432,
"scriptDurationMs": 7.679000000000002,
"eventListeners": 99,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
}
]
} |
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
Codecov Report❌ Patch coverage is
@@ Coverage Diff @@
## fix/ecs-review-findings #15654 +/- ##
========================================================
Coverage 82.54% 82.54%
========================================================
Files 1930 1930
Lines 110106 110134 +28
Branches 30165 30173 +8
========================================================
+ Hits 90885 90911 +26
+ Misses 18893 18886 -7
- Partials 328 337 +9
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 13 files with indirect coverage changes 🚀 New features to boost your workflow:
|
51903de to
97fc5d1
Compare
f975b08 to
6fc9acb
Compare
This comment has been minimized.
This comment has been minimized.
0756b7a to
075aa6a
Compare
075aa6a to
4069335
Compare
benjcooley
left a comment
There was a problem hiding this comment.
Review summary
| Severity | Count |
|---|---|
| BLOCKER | 1 |
| SHOULD FIX | 5 |
| TRIVIAL | 3 |
Three of the four production changes are sound; the fourth deletes links without tearing them down.
Blocking: realignInputLinkSlots now removes links by calling useLinkStore().updateEndpoints(scope, [], removals), which only displace()s the topology from the bucket. It skips everything LLink.disconnect() does — layoutStore.deleteLinkLayout, unregisterLinkTopology (so _graphScope and the useLinkStateStore entry survive), reroute pruning, and the onConnectionsChange(…, connected=false) callbacks on both endpoints. The reroute case is concrete: configure() prunes zero-link reroutes at LGraph.ts:2999-3003, before realignment at :3008, so a deleted link's reroutes persist at totalLinks === 0 — invisible but still box-selectable and still written back into the user's workflow. The same function already fires the connected=true callback for moves, so the removal path is inconsistent with its own neighbours.
Secondary: the !result.ok fallback deletes those links regardless of which validation error fired, so a duplicate-target between two moves (two serialized inputs sharing a name) also silently drops links that weren't blocking anything.
Verified as correct, for the record:
realignmentDataMap(LGraph.ts). The only realconfigure()override that mutatesinfo.inputsisComfyNode.configure(litegraphService.ts:473 / :580). It replacesdata.inputswith node-definition-ordered entries, whichLGraphNode.configurecopies ontothis.inputs— making the post-configure name→slot lookup self-consistent by construction, i.e. the oldnodeDataMapwas close to a no-op. The pre-configure snapshot is the only surviving record of the original name↔link pairing. Direction is right.defineEnumerableTopologyFacade(LLink.ts). I went looking for fallout from making 7 accessors own-enumerable and found none: noextends LLink, noObject.keys/for…in/JSON.stringify/structuredClone/cloneDeepover realLLinkinstances (every such site works onasSerialisable()output), noObject.assign(<LLink>, …), and Pinia only storeslink._state.LGraphNode.ts:1012already uses the same per-instance-accessor pattern. Only issue is the unhoistedgetOwnPropertyDescriptorscall.onAdded→rebuildInputWidgetBindings. Correct fix for #15565 —_convertToSubgraphImplnever callsconfigure()afterthis.add(subgraphNode), so this is the only place theUNASSIGNED_NODE_ID-keyed widget can be re-keyed. Flagged for cost and for the delete-on-failed-resolve edge, not correctness.
Note on CI: the six failing Playwright shards fail identically on the base branch (#15610), so they're inherited, not introduced here.
Test gaps: nothing covers what happens to reroutes or onConnectionsChange when realignment deletes a link, and the onAdded rebuild is only exercised through convert-to-subgraph, not through workflow load or paste — the two paths where it now runs on every subgraph node.
| const destinationSlots = new Set(moved.map(({ slot }) => slot)) | ||
| const removals = unmatched.filter( | ||
| (link) => | ||
| graph.links.has(link.id) && destinationSlots.has(link.target_slot) |
There was a problem hiding this comment.
BLOCKER — link removal here bypasses the disconnect lifecycle.
updateEndpoints(scope, [], removals) only calls displace() on the topology bucket. Everything else that LLink.disconnect() (LLink.ts:623-632) and LGraph._removeLink() (LGraph.ts:1824-1830) do is skipped:
layoutStore.deleteLinkLayout(id)— orphaned link layout + segment entries.unregisterLinkTopology(link)— solink._graphScopestays set, itsuseLinkStateStoreentry is never released, andlinkByTopologystill resolves the dead topology.- reroute pruning. This matters concretely:
configure()runs its "drop reroutes that no live link passes through" loop at LGraph.ts:2999-3003, beforerealignInputLinkSlotsat :3008. A link deleted here that had reroutes leaves them attotalLinks === 0— invisible (they're only drawn viavisibleReroutesaccumulated during link rendering) but still box-selectable viagetItemsInRect(LGraphCanvas.ts:4441) and still re-serialized into the user's workflow. node.onConnectionsChange(INPUT, oldSlot, false, link, …)and the origin-side output notification. This same function already fires theconnected=truecallback for moves a few lines below, so the removal path is internally inconsistent — and per AGENTS.mdonConnectionsChangeis depended on by 40+ custom node repos for dynamic slot bookkeeping.
Keeping the removal inside the atomic batch is right (it's what makes the occupied-target validation pass), so don't swap to graph.removeLink. But after the batch succeeds, run the non-store half of teardown for each removed link: delete its layout, unregister its link state / _graphScope, prune now-empty reroutes, and fire the disconnect callback on both endpoints.
There was a problem hiding this comment.
Correction — downgrading this from BLOCKER to SHOULD FIX. I went back and verified each consequence I asserted; two of them don't survive scrutiny and the reachability is much narrower than I implied. Details, so you can check my work:
Retracting the layout-leak claim. Link layouts are written only at render time (litegraphLinkAdapter.ts:341, inside renderLink), so nothing exists to leak during a fresh configure(). And on a re-configure of an already-rendered graph the omission is swamped by a pre-existing gap: layoutStore.handleClearGraph (layoutStore.ts:864-889) clears ynodes/contentSizes/ygroups/yreroutes but not linkLayouts/linkSegmentLayouts — only clearViewGeometry() does that, and it fires solely on a currentGraph reference change (GraphCanvas.vue:272,281). Stale link layouts already outlive every in-place workflow switch for all links. Not attributable to this PR.
Reachability: the deletion path can't fire for backend-registered nodes. ComfyNode.configure (litegraphService.ts:446 and :552) rebuilds data.inputs = [...definedInputs, ...extraInputs], and extraInputs (:470-472) keeps every serialized input whose name isn't a node-definition name. LGraphNode.configure then copies that onto this.inputs. So every serialized input name is present post-configure, unmatched is provably empty, and no link is deleted. Those two overrides are the only configure overrides in src/extensions, src/core, src/services.
That leaves two reachable cases: SubgraphNode (its configure discards the serialized inputs entirely, rebuilds from subgraph.inputNode.slots, then _internalConfigureAfterSlots filters on _subgraphSlot), and third-party classes registered directly via LiteGraph.registerNodeType that prune inputs. Note Subgraph extends LGraph (LGraph.ts:3160) and Subgraph.configure calls super.configure, so this also runs for nodes inside subgraph definitions.
Reroute orphan: confirmed, but milder than I said. The ordering holds — prune at LGraph.ts:3001-3005, realign at :3008. Reroute.linkIds is derived live (rerouteStore.buildMembershipIndex runs inside a computed over useLinkStore().graphTopologies), so the reroute genuinely drops to totalLinks === 0 rather than dangling on a dead id. It's never drawn (only reroutes walked during link rendering enter _visibleReroutes), but getItemsInRect (LGraphCanvas.ts:4441) iterates graph.reroutes.values() unconditionally, so it's an invisible box-selectable item; it's serialized, and pruned on the next load. Junk data, not breakage. Cheapest fix is moving the prune loop to after line 3008.
Link-state leak: confirmed as a leak, not as a bug. unregisterLinkState is skipped and useLinkStateStore has no bulk clear (only get/register/unregister), so the entry survives permanently. But _addLink → observeLinkId bumps lastLinkId past every loaded id before realign runs, so no newly minted link can reuse the freed id in-session, and the only reuse path (reloading the same JSON into the same graph id) re-triggers the same deletion. I could not construct a case where a surviving link inherits the stale color/_pos/data.
Also benign: leaving link._graphScope set can't cause a cross-delete — deleteLink and validateEndpointUpdates are both guarded by ownsPlacement (linkStore.ts:264-272), which fails once the topology is no longer the byId occupant.
What genuinely stands: the missing onConnectionsChange(…, false, …). disconnectInput fires it on both endpoints (LGraphNode.ts:3486-3500). dynamicWidgets.ts:611-617 consumes the false branch and — unlike the matchType handler directly above it at :306 — does not bail on app.configuringGraph, so it is live during load; it calls autogrowInputDisconnected (:500), which compacts the dynamic slot group. Deleting a link without it leaves an autogrow node holding a stale slot. That's a real behavioural gap, and it's inconsistent with this same function firing the connected=true callback for moves a few lines below.
So: worth fixing, and the reroute prune reorder is nearly free, but this isn't merge-blocking and I shouldn't have filed it as such.
There was a problem hiding this comment.
One sharpening on reachability, since it's the crux of whether this matters.
Conditions 1 and 3 aren't independent constraints — they come free. In a saved file a link's target_slot equals the index of its input in the serialized inputs array, so if the input at index k is dropped, every later linked input shifts down and some link necessarily moves into slot k — exactly where the doomed link sits. The whole question is condition 2 (a serialized input name absent from node.inputs after configure()).
For first-party code, condition 2 is closed off in both directions:
widgetInputs.ts:577-595is the one place that genuinely deletes named inputs during load, and it runs fromonGraphConfigured→app.ts:917triggerCallbackOnAllNodes→graph.onConfigureat LGraph.ts:3043, i.e. after realign at :3008.widgetInputs.ts:599onConfigureis gated on!app.configuringGraph(the depth counter set by theLGraph.prototype.configurewrapper at app.ts:903-911).dynamicWidgetsautogrow only adds during configure; its removal path is deferred torequestAnimationFrame(:606), andmatchTypereturns early onapp.configuringGraph(:312).
The gap worth knowing about: node.onConfigure fires at LGraphNode.ts:1192, inside the node-configure loop — before realign. Nothing gates third-party callbacks the way the first-party ones gate themselves, so an extension that calls removeInput from onConfigure or a nodeCreated hook puts an ordinary Comfy node into condition 2, and the link is deleted without the onConnectionsChange(…, false, …) that its own dynamic-input bookkeeping is waiting on. Per the entity-callback warning in AGENTS.md that's a known ecosystem pattern, and it's the case I'd actually design the teardown for.
Separately, worth a look while you're in here: SubgraphNode._internalConfigureAfterSlots:543 prunes with a raw this.inputs = this.inputs.filter(input => input._subgraphSlot) rather than removeInput, which is why an orphaned link is sitting at a live slot in the first place. That prune landed for a real report (#9977), so instance/definition divergence does happen — though the divergence attested there is duplicate names rather than absent ones, which can't trigger the deletion path.
There was a problem hiding this comment.
Addressed in 72613a372c: successful atomic removals now complete link teardown, increment graph version, notify both endpoints, and run reroute pruning after realignment.
| removals.map((link) => link._state) | ||
| ) | ||
| if (!result.ok) { | ||
| if (removals.length) { |
There was a problem hiding this comment.
SHOULD FIX — the fallback deletes removals regardless of why the batch failed.
validateEndpointUpdates (linkStore.ts:275-327) can also return duplicate-target (two serialized inputs sharing a name both resolve to slots[0]), duplicate-topology, or unowned-topology. In those cases the unmatched links weren't blocking anything, yet they get deleted anyway — silent link loss during workflow load on an unrelated validation error.
Gate on the error code, and don't discard the retry's result:
if (result.error.code === 'occupied-target' && removals.length) {
const removed = useLinkStore().updateEndpoints(
graphScopeOf(graph),
[],
removals.map((link) => link._state)
)
if (!removed.ok) console.error('Failed to vacate blocked input slots', removed.error)
}There was a problem hiding this comment.
Addressed in 464cc93e57: the unconditional removal-only fallback was deleted. A rejected batch now logs and leaves all moves and removals unchanged.
| } | ||
|
|
||
| for (let pass = 0; pass < referencedNames.size; pass++) { | ||
| for (let pass = 0; pass < Math.max(1, referencedNames.size); pass++) { |
There was a problem hiding this comment.
SHOULD FIX — Math.max(1, …) is a no-op.
It only changes behaviour when referencedNames.size === 0, and in that case moved is built by iterating referencedNames, so it's necessarily empty and the loop breaks on the very next line. The unmatched/removals work added below sits after that break, so it isn't reached either.
It reads like it fixes something but changes nothing. Revert to referencedNames.size unless there's a case I'm missing.
| for (let pass = 0; pass < Math.max(1, referencedNames.size); pass++) { | |
| for (let pass = 0; pass < referencedNames.size; pass++) { |
| targetSlot: fields.target_slot | ||
| }) | ||
| console.warn( | ||
| `Dropping competing link to occupied input ${fields.target_id}:${fields.target_slot}`, |
There was a problem hiding this comment.
TRIVIAL — the production log string was reshaped to satisfy a loose test assertion.
linkDeduplication.conflictingOrigins.test.ts:87 asserts expect(warn.mock.calls.flat().join(' ')).toContain('3:0'). That failed only because the second argument is an object and join(' ') stringifies it as [object Object] — the ids were already there in targetNodeId / targetSlot. Better to fix the assertion than the message:
expect(warn).toHaveBeenCalledWith(
expect.any(String),
expect.objectContaining({ targetNodeId: toNodeId(3), targetSlot: 0 })
)Related: LGraph.inputSlotRealign.test.ts:327 now hard-codes the full message text, which is a change-detector assertion. Also note fields.target_id is interpolated raw here while the dedup key above uses toNodeId(fields.target_id), so the two can disagree for string node ids.
There was a problem hiding this comment.
Superseded by #15689 — don't spend an edit on this. That PR rewrites this warn into reportDroppedLink (a single formatted string plus a trackLinkDedupDrop telemetry call) and replaces the assertion in linkDeduplication.conflictingOrigins.test.ts with an exact toHaveBeenCalledWith(...) rather than the loose toContain('3:0'). That's the fix I was asking for here, so this comment resolves itself in the rebase.
Worth flagging that the same test is a hard collision: both PRs convert it.fails('warns when a link is dropped in favour of a different origin') to it(...) and both rewrite the same assertion line, incompatibly. #15689's version should win.
For what it's worth, I checked and #15689 doesn't touch realignInputLinkSlots at all — its linkDeduplication.ts changes are confined to normalizeConfiguredTopology / remapLinkReferences / the new prune + telemetry helpers. So the realignment work in this PR, and my other two comments on it (the Math.max no-op at :165 and the unconditional removals deletion at :198), should survive the rebase unchanged and still need addressing.
| } | ||
|
|
||
| function defineEnumerableTopologyFacade(link: LLink): void { | ||
| const descriptors = Object.getOwnPropertyDescriptors(LLink.prototype) |
There was a problem hiding this comment.
SHOULD FIX — hoist this out of the constructor.
Object.getOwnPropertyDescriptors(LLink.prototype) materialises a descriptor object for every one of the ~48 prototype members and then throws away 41 of them — on every LLink construction. That's one per link on every workflow load (LGraph.ts:2847, :2888, :2990, plus once per subgraph definition) and one per user connect (LGraphNode.ts:3201, :3294, SubgraphInput.ts:92, …). Prototype descriptors are frozen after class definition, so this is pure waste:
const TOPOLOGY_DESCRIPTORS = (() => {
const d = Object.getOwnPropertyDescriptors(LLink.prototype)
return Object.fromEntries(
(['id','type','origin_id','origin_slot','target_id','target_slot','parentId'] as const)
.map((key) => [key, { ...d[key], enumerable: true }])
)
})()(evaluated after the class declaration, then Object.defineProperties(this, TOPOLOGY_DESCRIPTORS)).
The approach itself checks out — I looked for breakage from making these 7 own-enumerable and found none: no extends LLink anywhere, no Object.keys/for…in/JSON.stringify/structuredClone/cloneDeep over real LLink instances (every such site operates on asSerialisable()/serialize() output), no Object.assign(<LLink>, …), and Pinia only ever stores link._state, never the link. LGraphNode.ts:1012-1023 already uses the same per-instance-accessor trick, so the shape cost is an established pattern here.
|
|
||
| override onAdded(_graph: LGraph): void { | ||
| this.invalidatePromotedViews() | ||
| this.rebuildInputWidgetBindings() |
There was a problem hiding this comment.
SHOULD FIX — this wires previously-dead code into a hot lifecycle path, and it's redundant in two of the three callers.
rebuildInputWidgetBindings had zero callers on fix/ecs-review-findings. It has effectively never run in production, and it now runs for every SubgraphNode added to any graph.
It's genuinely needed for _convertToSubgraphImpl (LGraph.ts:2243 this.add(subgraphNode) is never followed by a configure(), so this is the only chance to re-key off UNASSIGNED_NODE_ID). But the two high-volume paths call configure() immediately after add():
LGraph.configure— LGraph.ts:2964this.add(node, true)then :2971node.configure(nodeData)LGraphCanvas._deserializeItems— :4248graph.add(node)then :4249node.configure(info)
In both, configure() blows away this.inputs wholesale and re-runs _resolveInputWidget for every slot. So the full teardown + interior-link walk + registerWidget + _setConcreteSlots() + widget-promoted dispatch this does is thrown away microseconds later — per promoted input, per subgraph node, on every workflow load and every paste. The duplicate widget-promoted also double-fires createPromotionErrorReconciler.ts:74 (queueing a redundant dropOutOfScope/rescanHost microtask) and SubgraphEditor.vue:90.
The only thing onAdded actually needs to do is migrate the store entries whose id encodes the stale node id — no re-resolution required:
for (const input of this.inputs) {
const stale = input.widgetId
if (!stale) continue
const fresh = widgetId(this.rootGraph.id, this.id, input.name)
if (fresh === stale) continue
const state = store.getWidget(stale)
if (state) store.registerWidget(fresh, state, /* render state */)
store.deleteWidget(stale)
input.widgetId = fresh
}| } | ||
|
|
||
| const activeIds = new Set(this.inputs.map((input) => input.widgetId)) | ||
| for (const { id } of previousBindings.values()) { |
There was a problem hiding this comment.
SHOULD FIX — a failed re-resolve destroys the user's promoted value instead of leaving it alone.
_resolveInputWidget is best-effort: it returns without calling _setWidget when subgraphInput.linkIds is empty or every link.resolve() misses (the three console.warn paths at :627/:636/:641). The loop above also continues before calling it at all when input._subgraphSlot is missing — and input.widgetId has already been deleted by then.
In any of those cases input.widgetId is undefined, so the old id isn't in activeIds and this line deletes it. The promoted widget's value is gone permanently, rather than just staying where it was.
Only delete an old id when the input actually acquired a different one:
for (const [input, { id }] of previousBindings) {
if (input.widgetId && input.widgetId !== id) store.deleteWidget(id)
}| // add before configure, otherwise configure cannot create links | ||
| this.add(node, true) | ||
| nodeDataMap.set(node.id, n_info) | ||
| realignmentDataMap.set(node.id, { |
There was a problem hiding this comment.
TRIVIAL — the snapshot retains far more than it needs.
realignInputLinkSlots reads only nodeData.id and inputs[].name / inputs[].link, but this keeps a shallow copy of every node's entire serialized payload (including the widgets_values / properties references) plus a fresh object per input, alive until :3008. On a few-thousand-node workflow that's a second copy of the whole node array for what amounts to a name→link table.
A narrower structural param type on realignInputLinkSlots would let this be { id: node.id, inputs: n_info.inputs?.map(({ name, link }) => ({ name, link })) }.
The direction of this change is right, though — I checked the only real configure() override that touches info.inputs (ComfyNode.configure, litegraphService.ts:473 and :580). It replaces data.inputs with node-definition-ordered entries, which LGraphNode.configure then copies straight onto this.inputs, making the post-configure name→slot lookup self-consistent by construction — i.e. the old nodeDataMap was a near no-op. The pre-configure snapshot is the only surviving record of the original workflow's name↔link pairing.
| this.addWidget('number', PROMOTED_INPUT, 0, () => {}) | ||
| } | ||
| } | ||
| LiteGraph.registered_node_types[CONVERTIBLE_NODE_TYPE] = ConvertibleNode |
There was a problem hiding this comment.
TRIVIAL — writing directly into LiteGraph.registered_node_types skips registerNodeType, which also sets .type, .title and the category registration. It works for this test, but the file already has a fixture for this (registerTestSubgraphNodeTypes) and LiteGraph.registerNodeType is used in the sibling realign specs.
Also: the if (!registered) guard means onTestFinished is only scheduled on the first call within a test. Fine as long as this helper is only ever called from a test body — it would silently leak the registration if someone moved it into beforeEach.
4069335 to
8f1c609
Compare
|
Verdict correction: no blockers. I re-verified the one BLOCKER I filed and downgraded it to SHOULD FIX — full reasoning in the thread on Revised counts:
The changes-requested state now rests on the SHOULD FIX items, not on anything merge-blocking. The two I'd most want addressed before merge are the unconditional Apologies for the overstated first pass. |
8f1c609 to
2831056
Compare
2831056 to
cdedaca
Compare
christian-byrne
left a comment
There was a problem hiding this comment.
Sequence gate only.
- Current head reviewed.
- Land #15689 first.
- Rebase, preserve both contracts, re-request.
Full context for agent readers
Reviewed exact head cdedacaba87ea3d510cc40e8cb2981fc7366fdfb against its stacked base fix/ecs-review-findings, including implementation correctness, lifecycle/ecosystem regressions, tests, open review threads, workspace invariants, exceptions, roadmap, related open PRs, and the recorded ECS land order.
I found no independent blocking correctness defect in this delta. The seven own-enumerable LLink forwarding descriptors match the settled D-c6-01 compatibility contract. The remaining callback/cardinality concerns are non-blocking follow-ups already represented in the existing discussion.
Approval is not valid yet because DQ-27/xc-102 records a concrete overlap order: #15689 must land first after human approval. Then this stack must rebase while preserving #15689's authoritative-reference survivor, pruning, and telemetry semantics and replaying this PR's unmatched-link realignment behavior. The overlap is real in LGraph.inputSlotRealign.test.ts, linkDeduplication.conflictingOrigins.test.ts, and linkDeduplication.ts.
After that rebase, rerun the conflicting-origin, rejected-batch, and input-slot realignment detectors and request review on the resulting exact head.
|
Landed in main via #15924 (squash e7d0d3e). Closing. Full context for agent readersThis PR was part of the 7-PR ECS stack (#15544 #15610 #15654 #15666 #15670 #15783 #15786) merged as a single squash through vehicle PR #15924 (branch refactor/ecs-thermonuclear-cleanup-main = stack tip + main merge + CI fixes), per the stack-merge plan agreed between Christian and DrJKL on 2026-08-25.
|
Summary
Re-inventories the remaining expected test failures from the ECS migration and fixes the cases owned by ECS compatibility and lifecycle behavior.
Changes
LLinktopology facades, handles dropped inputs during link-slot realignment, and re-keys promoted subgraph widgets after node insertion.Review Focus
Please focus on the atomic removal/move behavior during input-slot realignment and promoted-widget value preservation while replacing provisional node IDs.
Fixes #15565