Performance report · 2026-10-04

dockview-vue 8.4 vs CustomLayout

A comparison of two window layout engines for a trading terminal: operation timings, scaling, memory, DOM, bundle size and features. Plus the results of the dockview 5.2 → 8.4 upgrade.

dockview 8.4.0CustomLayout Chromium 153.0.8010.12 headless · M3 Pro 3 passes × CPU ×1 and ×4 commit a143bbfa + uncommitted changes

01Summary

There is no clear winner. CustomLayout is several times faster at heavy lifecycle operations on large layouts (mount, restore, clear, splitter release). dockview is faster at small, frequent operations (splitter drag step, serialization, panel removal and moving light panels in large trees, first tab open) by 2–20× or more, but in absolute terms that is fractions of a millisecond to a few ms at ×1 (up to ~24 ms at ×4); moving heavy panels is only ×1.1–1.23. Under comparable load (flat-20 with heavy panels) the engines are equal; at size 4 the difference is a few ms.

Mounting 200 heavy panels, ×1

279 / 122 ms

CustomLayout is 2.3× faster, but keeps 20 live panels versus 200

Snapshot restore, 200 heavy, ×1

539 / 137 ms

CL / DV = 0.25×; both engines mount new ids

Splitter step, 200 heavy, ×1

0.42 / 1.99 ms

dockview is 4.7× faster; both are well under the 16.7 ms frame budget

Bundle, JS gzip

86.3 KB / 14.2 KB

CustomLayout is 6.1× smaller

CustomLayout is faster

Mount and restore with ≥ 20 heavy panels (except flat-20): from ×0.81–0.82 (mount, 20) to ×0.25–0.28 (restore, 200), the same at ×1 and ×4
Clear already wins at size 4 (×0.33–0.65)
Clear at ≥ 20: ×0.06–0.12 (240 → 13.3 ms at 200)
Splitter release, heavy panels: ×0.03–0.07 (46.9 → 2.5 ms at 200); with light panels, from parity to ×0.59–0.64 at 100–200 and ×4
Forced resize ×0.50–0.65 (heavy; the absolute gain is fractions of a ms, up to 1.9 ms at ×4); mount cycles ×0.61–0.67 (heavy; parity with light panels)
Tab click: visible latency is one frame lower (dockview defers activation to rAF)
At ×4 with 100–200 heavy panels, real drag is smoother: half as many dropped frames (84–89 vs 185); median frame 33 vs 50 ms, but it is quantized to frames and the difference does not pass the significance criterion

dockview is faster

Splitter drag step: 2–5.6×
Serialization: ~10× (11 vs 115 µs at 200), since getLayout makes a deep copy
Moving a panel in large trees: ×1.23 (heavy), ×4.3 (light, 200); in CustomLayout this is a full config reload
Removing a single panel with ≥ 50 panels: ~18–23× for heavy (0.27 vs 6 ms at 200), 3.4–7.8× for light; with 4–20 heavy panels the difference is not significant. The sharp drop in dockview's median from ~4.5 to ~0.25 ms at ≥ 50 has not been investigated (p95 is ~9–10 ms for both)
First tab open (tabs-20): in CustomLayout this is a lazy mount, 2.0–7.4 ms at ×1 (up to 23.6 ms at ×4) vs < 0.3 ms
Cold mount of 200 light panels: ×1.13
Real drag at ×4 with 20 heavy panels: half as many dropped frames (39 vs 80.5); at 50 the median frame is 16.7 vs 25 ms, but dropped frames are equal (88.5 vs 82), and the difference in median frame is not significant by the criterion

Parity

flat-20 mount (same number of mounted panels): 108.5 / 108.5 ms
Mount and restore at size 4, heavy panels: 4–7 % difference. With light panels CustomLayout is formally significantly faster by 14–28 %, but in absolute terms that is 0.7–4 ms
Tab switch via API, heavy panels: ×0.89–1.03
Moving heavy panels at 20–50: ×0.99–1.09 (with light panels dockview is faster from 20 already at ×4)
Real drag at ×1: both hold 60 fps at all sizes

Where CustomLayout's advantage comes from. Mostly architectural: CustomLayout mounts only tabs that have been visible at least once, while dockview creates a component for every panel. With 200 heavy panels, 20 stay live versus 200; on tabs-20, 2 versus 21. Where the number of mounted panels is the same (flat-20), heavy-panel mount time matches within 2 % (light: 4–10 % in CustomLayout's favor). For a real terminal with many hidden tabs this is a fair advantage, but it does not mean the layout code itself is faster.

Statistical significance. Out of 240 comparisons on ranked metrics (all sizes, both panel types, ×1 and ×4): 122 CustomLayout faster, 69 dockview faster, 28 parity (difference < 10 %), 21 not significant. "Significant" here means the faster engine won in each of the 3 passes and the interquartile ranges of the pooled samples do not overlap. This is a heuristic, not a formal test: with n = 6 samples per point (more for some metrics) and 3 passes, p-values were not computed. Multi-fold differences are robust; 10–25 % differences at small sizes are on the edge of noise.

What this means in practice. All of CustomLayout's losses in frequent operations (splitter step ≤ 2 ms at ×1, serialization ≤ 0.5 ms) fit within the frame budget even at ×4 with 200 panels (step 9.5 ms). The exception is opening a heavy tab for the first time at ×4 (lazy mount, 23.6 ms): roughly one dropped frame. The wins, meanwhile, fall on operations users notice: opening a workspace (hundreds of ms at ×4) and splitter release (185 → 12 ms at ×4/200). The decision should be based not on speed but on features (see section 09) and maintenance cost.

02Environment and methodology

Machine	Apple M3 Pro, 12 cores, 18 GB, AC power, battery 100%
OS	macOS 26.6.2 (25G83), Darwin 25.6.0
Browser	Playwright Chromium 153.0.8010.12 (headless, channel chromium); Playwright 1.63.0; page isolation per pass (`--isolate`)
Flags	`--js-flags=--expose-gc --enable-precise-memory-info --disable-renderer-backgrounding --disable-background-timer-throttling --disable-backgrounding-occluded-windows --disable-extensions`
Node / npm	v20.10.0 / 10.2.3
Libraries	dockview-vue 8.4.0, dockview-core 8.4.0, Vue 3.5.17, Vite 2.9.15
Build	production: `VITE_LAYOUT_BENCH=1 npm run build, served by vite preview :5050`
Git	`feat/custom-layout` @ `a143bbfa`, working tree with changes
Viewport	1920x1200 @1x, stage 1600x1000; crossOriginIsolated, timer resolution 5 µs, `gc()` available
Run time	2026-10-04T00:08:25Z — 2026-10-04T01:31:36Z (83 min), errors: 0
Safety	All `/api/*` are mocked; only 6 x GET /user went out (mocked 401); non-GET requests: 0; external: 0

How it is measured. The route /layout-benchmark mounts both engines in a 1600×1000 stage and runs identical scenarios through adapters. For each operation it records:

syncMs — the synchronous part of the operation until control returns (including the Vue flush) — the primary cost metric;
readyMs — when the DOM reached the target state (polled in rAF and after a frame) — quantized to frames;
paintMs — until the next paint after readiness;
special: perRemoveSyncMs, stepMs/commitMs, serializeUs, cycleMs, heapDeltaBytes.

Two panel types: light (a nearly empty component) and heavy (substantial DOM and work on mount). CPU throttling via CDP: ×1 and ×4 (simulating a slow machine). 3 passes per throttling level, each in a fresh page; engine order alternates between passes.

Aggregation. The chart bar is the median of per-pass medians; the whisker is the p95 of the pooled samples. With n = 6, p95 is effectively the maximum. A CL / DV ratio < 1 means CustomLayout is faster.

Layout sizes

Key	Panels	Groups	Tabs per group	Depth	Branching
4	4	4	1	2	2
20	20	10	2	3	3
flat-20	20	20	1	3	3
50	50	10	5	3	3
100	100	20	5	3	3
200	200	20	10	3	3
tabs-20	21	2	20	1	2

Iteration count reduced

With default settings one ×1 pass took ~30 min and ×4 about three times longer; the full run would have taken ~5 h. Iterations were reduced to fit in ~80 min. This widens confidence intervals, especially for small values at size 4.

Parameter	Default	Used
`lifecycleIterations`	10	2
`operationIterations`	30	6
`warmup`	3	1
`blocks`	5	2
`cycles`	20	8
`dragIterations`	3	2
`serializeCount`	1000	300
`trustedRepeats`	5	2

Data quality checks

Completeness: 2344 aggregates, no NaN or empty values; both engines passed all scenarios; all real drags were verified (2/2 per engine/size/variant/pass).
Spread: for 19 of 480 points on ranked metrics (16 metrics × size × panel type × CPU × engine), per-pass medians differ by more than 1.6×; all but one are values < 3 ms, mostly at sizes 4 and 20. The only real outlier: restore-same-ids, CustomLayout, 100 heavy, ×4 — 57.9 ms in pass 1 vs 7.4 ms in the other two (the median of medians is unaffected).
Recheck: restore and remove-all at sizes 4 and 100 at ×4, 3 passes of 5 iterations: the outlier gave 7.55 ms (CV 0.06); 41 of 48 values are within 10 % of the main run, the other 7 differ by 13–32 % (mostly CustomLayout: restore-same-ids and restore at size 4, clear and remove-all at 100; plus dockview clear at size 4).
Uninvestigated oddities: tab-switch-api dockview light is faster at 4x (0.78 ms) than at 1x (1.51 ms), consistent across passes; remove-sequential perRemoveSyncMs dockview heavy drops from ~4.5 ms (size 20) to ~0.25 ms (size >=50) while its batched syncMs grows to 887 ms at 100 (4x): per-removal cost there is dominated by hidden tab panels.

03Caveats and fairness of the comparison

A methodology review found 9 issues; all were fixed before the final run. Below is what was fixed and what remains a limitation.

Fixed before the run

restore did different work. CustomLayout stores panels by id, so reloading the same snapshot is nearly a no-op. Now restore-replace loads a snapshot with new ids (both engines mount N panels); the old variant was moved to restore-same-ids.
readyMs lagged by a frame for work deferred to rAF (dockview's ResizeObserver). Added a recheck after the frame: dockview resize-observed 33.5 → 18.3 ms.
readyMs was presented as engine cost. It is now explicitly marked as frame-quantized and is not ranked.
remove-sequential compared batched work. Added a comparable metric perRemoveSyncMs.
Fixed engine order in mount-cycles. Each engine now goes first once per pass.
The CDP heap check mixed engines. Now done separately per engine, with alternating order.
Long tasks / TBT ignored the operation's own task. They are now matched by overlap with the window.
Real drag ignored the axis. The axis is now taken from prepare().

Remaining limitations

Mount counts differ by design: dockview creates a component for every tab, CustomLayout mounts only active tabs, lazily. Compare flat-20 with 20 and read alivePanels next to the timings.
CustomLayout has no fine-grained API for activate, move and add-to-group: those paths are getLayout → helper → loadLayout (a full reload), exactly as the app does it, and are not equivalent to dockview calls.
CustomLayout removeComponent refuses to remove the last remaining panel; the adapter falls back to clearLayout for it.
Headless frame times are indicative. Rank on syncMs, perRemoveSyncMs, stepMs, commitMs and serializeUs, not on readyMs or paintMs: both are quantized to display frames and mostly measure vsync wait.
readyMs is polled at each rAF and again in a task after that frame, so rAF-deferred work is seen in the frame it runs in. Read it together with framesToSettle (the syncMs extra), never as engine cost.
resize-observed: dockview lays out one frame later by design (ResizeObserver → rAF → layout), so its readyMs carries one extra frame of idle time; resize-forced isolates the layout cost. syncMs is only meaningful where readyInSync is 1.
tab-switch-warm: dockview activates on pointerdown and defers it one rAF (HTML5 DnD on); CustomLayout activates on click. Both get the same pointerdown → pointerup → click sequence. Its readyMs is user-visible latency in frames, not work; compare engine cost on tab-switch-api syncMs.
restore-replace loads a snapshot of the same shape with disjoint ids, so both engines unmount N and mount N panels. restore-same-ids reloads identical ids: CustomLayout keeps its id-keyed panels mounted and only diffs the tree, dockview fromJSON rebuilds every panel. Check the mounts extra before reading it as a speed difference.
remove-sequential removes all N panels in one synchronous block: dockview updates the DOM on every removePanel, CustomLayout replaces a reactive config N times and renders once. It measures batched API cost; perRemoveSyncMs (each removal flushed on its own) is the like-for-like per-operation metric.
mount-cycles runs each engine twice per pass in opposite order (each engine goes first once); heapDeltaBytes therefore has two samples per pass.
add-k appends 20 root groups to a 4-group tree; past ~16 root groups dockview clamps groups at its 100 px minimum while CustomLayout keeps shrinking them.
Reduced iterations (see above): only differences ≥ 1.5× are worth ranking; 10–25 % differences at size 4 are within noise.
Headless Chromium without GPU compositing: absolute numbers are lower than on users' real hardware; the ratios are what matter.
Synthetic panels are not real terminal widgets (TradingView charts, tables with data streams).

04Scenarios

Each metric has two charts (CPU ×1 and ×4) and a table. The bar is the median, the whisker is p95, and under each category is the CL / DV ratio (color marks the significant winner, gray means parity or noise). The toggle below switches the panel type in all charts.

Scenario list

1Cold mount `cold-mount · syncMs`

Empty engine → loading a tree of N panels, synchronous part until control returns. flat-20 is 20 groups with one tab each (same number of mounted panels in both engines). ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	23.9	25.9	23.2	1.08	22.7	24.2	21.6	0.95	0.95×	parity	6
4	×4	85.5	86.8	84.8	0.74	82.3	84.4	81.3	1.29	0.96×	parity	6
20	×1	62.7	64.4	61.1	1.28	51.6	53.1	50.1	1.12	0.82×	CL faster	6
20	×4	246	257	235	8.63	200	204	196	3.18	0.81×	CL faster	6
flat-20	×1	108	111	102	3.63	109	111	101	3.54	1.00×	parity	6
flat-20	×4	420	437	404	13.8	410	429	398	13.6	0.98×	parity	6
50	×1	88.5	89.4	87.1	0.89	54.5	55.3	53.6	0.74	0.62×	CL faster	6
50	×4	355	358	343	5.33	208	212	206	2.37	0.58×	CL faster	6
100	×1	196	197	195	0.60	114	118	109	3.95	0.58×	CL faster	6
100	×4	742	751	690	22.7	451	454	435	6.79	0.61×	CL faster	6
200	×1	279	291	276	5.04	122	126	118	3.28	0.44×	CL faster	6
200	×4	1,052	1,086	1,046	14.7	492	500	478	8.39	0.47×	CL faster	6

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	7.86	9.57	6.43	1.37	5.92	7.85	3.41	1.44	0.75×	CL faster	6
4	×4	14.9	15.7	14.1	0.53	10.8	11.9	10.0	0.82	0.72×	CL faster	6
20	×1	10.3	11.0	5.83	2.00	8.22	8.84	4.39	1.62	0.80×	not significant	6
20	×4	20.4	21.7	19.4	0.79	17.0	17.6	16.5	0.45	0.83×	CL faster	6
flat-20	×1	10.8	12.1	9.25	1.00	9.76	10.9	9.00	0.73	0.90×	parity	6
flat-20	×4	23.5	24.8	22.2	0.99	22.5	23.4	21.9	0.51	0.96×	parity	6
50	×1	11.0	11.9	10.4	0.57	10.6	11.6	10.1	0.53	0.96×	parity	6
50	×4	25.8	26.5	25.1	0.54	25.0	25.4	23.7	0.59	0.97×	parity	6
100	×1	15.2	15.8	13.7	0.85	15.9	17.3	15.3	0.76	1.05×	parity	6
100	×4	41.9	42.9	40.5	0.89	45.7	47.5	43.9	1.39	1.09×	parity	6
200	×1	18.8	19.7	17.9	0.70	21.2	22.2	20.1	0.82	1.13×	DV faster	6
200	×4	60.3	68.3	57.3	3.82	72.1	83.7	67.9	5.32	1.20×	DV faster	6

2Restore: new snapshot `restore-replace · syncMs`

Loading a snapshot of the same shape with different ids into an engine that already has a tree: both engines unmount N and mount N panels (a fair comparison after fix #1). ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	23.4	23.9	23.1	0.30	22.4	23.1	21.9	0.41	0.96×	parity	6
4	×4	92.7	94.4	91.4	1.26	86.5	87.7	84.8	1.26	0.93×	parity	6
20	×1	149	156	143	4.39	55.6	59.5	55.1	1.66	0.37×	CL faster	6
20	×4	553	570	542	10.6	241	246	224	7.76	0.44×	CL faster	6
50	×1	181	191	168	8.24	59.9	61.3	59.0	0.89	0.33×	CL faster	6
50	×4	700	721	689	12.5	245	251	228	9.14	0.35×	CL faster	6
100	×1	403	417	400	6.54	129	135	119	7.15	0.32×	CL faster	6
100	×4	1,553	1,580	1,535	15.4	505	527	471	22.7	0.32×	CL faster	6
200	×1	539	556	514	14.5	137	142	132	4.87	0.25×	CL faster	6
200	×4	2,008	2,152	1,982	77.7	556	577	545	12.6	0.28×	CL faster	6

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	4.68	5.19	4.16	0.50	3.94	4.28	3.64	0.27	0.84×	CL faster	6
4	×4	7.95	11.2	7.22	1.39	6.85	7.21	6.22	0.37	0.86×	CL faster	6
20	×1	12.6	13.0	10.8	0.79	6.80	7.30	6.26	0.39	0.54×	CL faster	6
20	×4	28.1	30.2	27.5	0.94	15.4	16.0	14.8	0.39	0.55×	CL faster	6
50	×1	14.0	14.9	13.3	0.62	9.21	10.8	8.02	0.91	0.66×	CL faster	6
50	×4	37.5	39.0	36.5	1.05	24.6	25.3	23.8	0.51	0.65×	CL faster	6
100	×1	23.2	24.4	21.8	1.19	15.6	16.4	15.0	0.53	0.67×	CL faster	6
100	×4	82.6	86.1	78.3	2.96	47.5	49.6	46.4	1.22	0.57×	CL faster	6
200	×1	34.3	35.8	33.2	1.05	21.7	22.3	20.8	0.63	0.63×	CL faster	6
200	×4	136	156	132	8.78	79.4	88.8	78.5	4.15	0.58×	CL faster	6

3Restore: same ids `restore-same-ids · syncMs`

Reloading the same snapshot. CustomLayout keeps panels with the same ids and only diffs the tree (mounts = 0), while dockview fromJSON recreates everything. This is different work: the gap reflects architecture, not "speed". ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	24.3	27.1	23.2	1.41	0.43	0.48	0.39	0.03	0.02×	CL faster	6
4	×4	91.6	92.6	90.2	0.95	1.27	2.23	0.96	0.50	0.01×	CL faster	6
20	×1	145	151	141	3.85	0.81	0.96	0.68	0.13	<0.01×	CL faster	6
20	×4	557	581	548	12.4	3.24	3.55	2.49	0.41	<0.01×	CL faster	6
50	×1	185	189	181	2.52	0.93	1.00	0.87	0.05	<0.01×	CL faster	6
50	×4	699	713	675	15.6	3.51	4.11	2.91	0.41	<0.01×	CL faster	6
100	×1	397	403	389	5.68	1.67	1.88	1.62	0.10	<0.01×	CL faster	6
100	×4	1,553	1,562	1,543	7.28	7.40	109	6.81	41.5	<0.01×	CL faster	6
200	×1	521	550	501	19.5	2.51	2.64	2.35	0.11	<0.01×	CL faster	6
200	×4	2,056	2,287	1,996	114	10.9	11.6	10.1	0.65	<0.01×	CL faster	6

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	4.66	6.32	3.75	0.85	0.93	1.52	0.63	0.32	0.20×	CL faster	6
4	×4	7.74	8.31	7.14	0.49	1.17	1.67	0.64	0.43	0.15×	CL faster	6
20	×1	13.0	14.2	11.7	0.96	1.26	1.52	1.01	0.20	0.10×	CL faster	6
20	×4	27.1	28.0	26.5	0.49	2.88	3.22	2.38	0.32	0.11×	CL faster	6
50	×1	13.8	15.9	12.7	1.21	1.50	1.65	1.34	0.12	0.11×	CL faster	6
50	×4	35.3	36.1	34.7	0.52	3.44	4.18	3.14	0.35	0.10×	CL faster	6
100	×1	23.8	26.0	20.6	1.99	2.01	2.22	1.86	0.15	0.08×	CL faster	6
100	×4	79.1	79.7	77.9	0.73	6.33	6.68	5.33	0.50	0.08×	CL faster	6
200	×1	33.2	35.9	31.6	1.58	2.52	3.28	2.44	0.32	0.08×	CL faster	6
200	×4	126	143	125	7.98	9.35	11.2	8.92	0.82	0.07×	CL faster	6

4Clearing the layout `clear · syncMs`

clear() / clearLayout() on a tree of N panels. ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	3.75	3.84	3.65	0.08	2.41	2.48	2.33	0.06	0.64×	CL faster	6
4	×4	14.2	15.8	13.4	1.00	9.22	10.3	8.77	0.61	0.65×	CL faster	6
20	×1	77.4	78.2	76.2	0.82	5.59	6.19	5.36	0.28	0.07×	CL faster	6
20	×4	345	383	316	24.8	23.2	24.4	22.3	0.72	0.07×	CL faster	6
50	×1	93.5	97.2	81.8	5.60	6.12	6.63	5.72	0.32	0.07×	CL faster	6
50	×4	362	403	353	19.5	23.7	26.6	23.3	1.45	0.07×	CL faster	6
100	×1	215	218	211	2.67	13.9	21.1	11.9	3.48	0.06×	CL faster	6
100	×4	874	943	866	30.6	62.4	71.1	50.1	8.09	0.07×	CL faster	6
200	×1	240	242	239	1.50	13.3	14.9	12.6	0.87	0.06×	CL faster	6
200	×4	960	1,064	947	50.1	60.0	68.6	50.1	8.27	0.06×	CL faster	6

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	1.51	1.81	1.25	0.19	0.50	0.98	0.37	0.22	0.33×	CL faster	6
4	×4	2.98	4.29	2.47	0.67	1.11	1.33	0.94	0.17	0.37×	CL faster	6
20	×1	7.20	8.05	6.75	0.44	0.75	0.83	0.67	0.05	0.10×	CL faster	6
20	×4	17.0	19.6	14.4	1.83	2.03	2.33	1.73	0.27	0.12×	CL faster	6
50	×1	7.70	8.13	7.26	0.38	0.93	1.40	0.71	0.24	0.12×	CL faster	6
50	×4	21.0	26.3	18.4	3.25	2.61	3.69	2.27	0.53	0.12×	CL faster	6
100	×1	15.8	16.5	15.0	0.54	1.17	1.30	1.01	0.11	0.07×	CL faster	6
100	×4	50.6	64.4	46.7	6.33	3.28	4.88	3.05	0.70	0.06×	CL faster	6
200	×1	23.2	24.6	22.3	0.85	1.49	1.65	1.34	0.14	0.06×	CL faster	6
200	×4	86.2	94.5	83.4	4.46	4.85	6.80	4.21	0.99	0.06×	CL faster	6

5Removal one at a time (per operation) `remove-sequential · perRemoveSyncMs`

Removing a panel with a forced render after each operation: a comparable per-removal metric. ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	5.61	9.55	0.65	4.05	1.03	1.98	0.15	0.77	0.18×	not significant	24
4	×4	21.6	39.9	2.34	15.8	4.24	7.83	0.45	3.20	0.20×	not significant	24
20	×1	4.50	10.2	0.64	3.06	3.82	6.03	0.44	2.85	0.85×	not significant	120
20	×4	19.1	41.4	2.41	13.5	16.1	48.3	1.58	12.2	0.85×	not significant	120
50	×1	0.24	10.1	0.13	3.36	5.26	6.83	0.46	1.99	21.7×	DV faster	300
50	×4	1.19	40.5	0.13	13.2	21.1	27.7	1.56	8.28	17.7×	DV faster	300
100	×1	0.25	9.50	0.14	4.67	5.65	9.07	0.46	3.07	22.8×	DV faster	600
100	×4	1.16	40.2	0.14	19.1	22.3	41.3	1.78	11.5	19.1×	DV faster	600
200	×1	0.27	8.87	0.15	3.65	6.00	9.72	0.45	2.42	22.4×	DV faster	1200
200	×4	1.14	35.9	0.18	14.8	23.8	36.8	1.69	10.3	20.8×	DV faster	1200

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	0.70	1.39	0.38	0.36	0.76	1.60	0.11	0.55	1.09×	parity	24
4	×4	1.48	3.26	0.55	0.97	1.59	4.35	0.05	1.40	1.07×	parity	24
20	×1	0.33	0.87	0.14	0.27	0.62	1.73	0.08	0.46	1.89×	not significant	120
20	×4	0.88	2.38	0.10	0.74	1.55	4.37	0.06	1.17	1.75×	not significant	120
50	×1	0.13	0.77	0.05	0.22	0.53	1.60	0.05	0.49	4.08×	DV faster	300
50	×4	0.52	2.41	0.04	0.75	1.80	5.45	0.05	1.45	3.44×	DV faster	300
100	×1	0.14	0.78	0.04	0.24	0.81	2.46	0.05	0.67	5.59×	DV faster	600
100	×4	0.61	2.78	0.03	0.91	3.25	8.50	0.05	2.48	5.28×	DV faster	600
200	×1	0.15	0.64	0.04	0.21	1.16	2.65	0.05	0.75	7.77×	DV faster	1200
200	×4	0.61	2.60	0.03	0.94	4.74	11.0	0.06	3.19	7.80×	DV faster	1200

6Removing all N in one block `remove-sequential · syncMs`

All N removals in a single synchronous task. dockview updates the DOM on every removePanel, CustomLayout renders once at the end: this measures the batch cost of the API, not identical work. ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	3.76	3.88	3.67	0.08	2.39	2.53	2.24	0.10	0.64×	CL faster	6
4	×4	15.6	16.1	14.4	0.68	9.73	10.4	9.46	0.32	0.63×	CL faster	6
20	×1	77.7	79.4	76.1	1.25	5.83	6.27	5.62	0.23	0.08×	CL faster	6
20	×4	325	364	320	18.1	24.7	27.1	24.2	1.12	0.08×	CL faster	6
50	×1	87.8	93.1	82.3	4.74	6.85	7.43	6.59	0.35	0.08×	CL faster	6
50	×4	361	400	343	23.5	28.2	29.4	27.1	0.85	0.08×	CL faster	6
100	×1	216	223	213	3.91	16.8	19.8	15.0	2.18	0.08×	CL faster	6
100	×4	887	982	872	50.9	78.4	89.4	62.3	10.6	0.09×	CL faster	6
200	×1	242	245	240	1.62	26.0	30.4	23.7	2.93	0.11×	CL faster	6
200	×4	971	1,048	952	42.0	115	124	95.9	10.7	0.12×	CL faster	6

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	1.60	2.02	1.32	0.26	0.60	0.66	0.47	0.08	0.38×	CL faster	6
4	×4	3.65	4.90	3.13	0.64	1.22	1.92	0.95	0.34	0.33×	CL faster	6
20	×1	6.26	6.54	5.41	0.45	1.13	1.28	0.98	0.12	0.18×	CL faster	6
20	×4	16.5	20.1	15.1	2.03	3.14	4.02	2.87	0.41	0.19×	CL faster	6
50	×1	7.32	7.52	7.05	0.23	2.10	2.27	1.92	0.12	0.29×	CL faster	6
50	×4	22.8	28.5	20.5	3.41	6.56	7.10	6.22	0.33	0.29×	CL faster	6
100	×1	13.8	15.4	13.6	0.66	5.22	5.49	4.83	0.29	0.38×	CL faster	6
100	×4	54.4	61.9	48.5	5.68	18.3	20.1	17.8	0.84	0.34×	CL faster	6
200	×1	23.2	24.2	22.9	0.49	13.4	14.2	12.9	0.49	0.58×	CL faster	6
200	×4	91.5	98.3	89.4	3.49	51.3	56.5	50.6	2.50	0.56×	CL faster	6

7Adding a panel to the root `add-k · syncMs`

Adding 20 root groups to a tree of 4 (one at a time, median per addition). CustomLayout has no fine-grained API — the path is getLayout → helper → loadLayout, as in the app. ranked metric

CPU ×1

CPU ×4

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
light, 4	×1	3.00	4.73	1.52	0.83	4.09	5.30	1.76	1.00	1.37×	not significant	120
light, 4	×4	4.47	7.69	2.63	1.34	7.22	10.6	3.16	2.20	1.62×	not significant	120
heavy, 4	×1	6.25	8.27	5.67	0.76	6.50	8.16	5.48	0.87	1.04×	parity	120
heavy, 4	×4	24.5	27.7	21.9	1.59	27.8	33.8	21.7	3.96	1.13×	not significant	120

8Adding a tab to a group `add-k-tab · syncMs`

Adding 20 tabs to an existing group (median per addition). ranked metric

CPU ×1

CPU ×4

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
light, 4	×1	2.04	3.02	0.79	0.61	2.51	4.54	1.43	0.89	1.23×	not significant	120
light, 4	×4	1.66	3.29	0.60	0.72	4.80	6.80	3.13	1.01	2.88×	DV faster	120
heavy, 4	×1	7.79	8.94	5.59	0.87	8.49	9.87	5.98	1.15	1.09×	parity	120
heavy, 4	×4	23.5	27.0	21.0	1.72	27.5	31.6	22.7	2.65	1.17×	DV faster	120

9Switching a tab via the API `tab-switch-api · syncMs`

Programmatic tab activation in a group of 20 tabs — the main "engine cost" metric for tabs. ranked metric

CPU ×1

CPU ×4

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
light, tabs-20	×1	1.51	2.20	0.42	0.46	1.60	1.88	1.12	0.22	1.05×	parity	36
light, tabs-20	×4	0.78	1.55	0.15	0.38	2.72	3.65	2.38	0.42	3.46×	DV faster	36
heavy, tabs-20	×1	6.76	7.98	5.13	0.96	5.99	6.72	4.94	0.59	0.89×	not significant	36
heavy, tabs-20	×4	18.1	19.9	16.3	0.97	18.6	25.0	17.5	1.92	1.03×	parity	36

10First tab open `tab-switch-first · syncMs`

First activation of each tab: for CustomLayout this is a lazy panel mount, for dockview the panel is already mounted. ranked metric

CPU ×1

CPU ×4

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
light, tabs-20	×1	0.28	0.41	0.12	0.08	1.98	4.51	1.20	0.93	7.07×	DV faster	114
light, tabs-20	×4	0.05	0.71	0.02	0.23	3.73	5.03	2.27	0.72	83.0×	DV faster	114
heavy, tabs-20	×1	0.09	0.15	0.06	0.03	7.38	9.13	6.70	0.69	79.8×	DV faster	114
heavy, tabs-20	×4	0.07	0.79	0.04	0.27	23.6	27.9	21.8	1.91	337.4×	DV faster	114

11Tab click (visible latency) `tab-switch-warm · readyMs`

A real pointerdown → pointerup → click sequence. dockview defers activation by one rAF, so readyMs ≈ one frame. This is frame latency, not engine work. reference metric

CPU ×1

CPU ×4

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
light, tabs-20	×1	17.5	19.1	15.9	0.87	1.60	1.91	1.23	0.19	0.09×	CL faster	36
light, tabs-20	×4	17.6	18.9	14.8	1.00	3.00	3.77	2.14	0.43	0.17×	CL faster	36
heavy, tabs-20	×1	24.6	26.9	15.6	2.32	6.31	6.67	5.11	0.41	0.26×	CL faster	36
heavy, tabs-20	×4	34.2	36.4	27.6	1.82	19.2	23.4	17.2	1.83	0.56×	CL faster	36

12Splitter drag: step `splitter-drag · stepMs`

Synthetic splitter movement: 40 steps of 6 px, synchronous processing time per step. ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	0.25	0.37	0.13	0.06	0.51	0.76	0.29	0.14	2.05×	DV faster	960
4	×4	0.23	1.07	0.09	0.34	1.27	2.28	0.25	0.50	5.58×	DV faster	960
20	×1	0.29	0.42	0.15	0.07	0.62	0.73	0.52	0.05	2.16×	DV faster	960
20	×4	0.88	1.49	0.15	0.37	2.66	3.74	1.47	0.78	3.01×	DV faster	960
50	×1	0.23	0.32	0.15	0.04	0.82	0.94	0.67	0.09	3.57×	DV faster	960
50	×4	0.94	1.63	0.14	0.38	3.37	4.29	2.24	0.50	3.57×	DV faster	960
100	×1	0.37	0.49	0.32	0.27	1.42	1.57	1.29	0.29	3.85×	DV faster	960
100	×4	2.09	2.80	0.98	0.41	6.51	7.55	5.24	0.74	3.11×	DV faster	960
200	×1	0.42	0.58	0.33	0.67	1.99	2.20	1.81	0.12	4.73×	DV faster	960
200	×4	2.39	3.08	0.67	0.43	9.51	10.8	7.77	0.93	3.97×	DV faster	960

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	0.60	0.96	0.18	0.23	0.68	1.73	0.30	0.46	1.12×	not significant	960
4	×4	0.12	0.90	0.06	0.30	0.58	1.51	0.14	0.44	4.81×	not significant	960
20	×1	0.53	1.30	0.22	0.34	0.87	2.07	0.47	0.41	1.64×	not significant	960
20	×4	0.30	1.13	0.11	0.35	1.48	2.78	0.65	0.58	4.88×	DV faster	960
50	×1	0.37	0.53	0.26	0.08	1.09	1.68	0.87	0.30	2.99×	DV faster	960
50	×4	0.72	1.28	0.13	0.35	2.20	3.49	1.43	0.55	3.08×	DV faster	960
100	×1	0.57	0.70	0.38	0.09	1.92	2.62	1.44	0.30	3.38×	DV faster	960
100	×4	1.89	2.67	0.66	0.72	4.21	5.40	3.18	0.54	2.23×	DV faster	960
200	×1	0.58	0.72	0.38	0.07	2.59	3.24	2.02	0.29	4.42×	DV faster	960
200	×4	2.03	2.76	0.67	0.61	6.29	7.92	5.14	0.73	3.10×	DV faster	960

13Splitter drag: release `splitter-drag · commitMs`

Splitter release: committing sizes and the final repaint. ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	13.8	15.1	12.6	0.73	0.68	0.99	0.47	0.20	0.05×	CL faster	12
4	×4	32.9	33.7	32.2	0.48	2.04	3.07	1.38	0.47	0.06×	CL faster	12
20	×1	24.4	26.1	21.8	1.06	0.79	0.88	0.72	0.05	0.03×	CL faster	12
20	×4	83.9	90.4	82.1	2.69	3.76	16.4	2.77	3.69	0.04×	CL faster	12
50	×1	23.4	24.0	20.9	1.03	1.09	1.14	0.94	0.07	0.05×	CL faster	12
50	×4	85.0	93.7	83.7	2.70	4.45	5.61	3.97	0.48	0.05×	CL faster	12
100	×1	45.2	46.1	44.0	0.63	1.72	1.96	1.66	0.10	0.04×	CL faster	12
100	×4	176	182	172	3.44	8.42	10.0	7.54	0.70	0.05×	CL faster	12
200	×1	46.9	48.5	45.8	0.74	2.52	2.80	2.36	0.13	0.05×	CL faster	12
200	×4	185	189	178	3.87	12.2	13.0	10.7	0.78	0.07×	CL faster	12

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	1.14	1.62	0.84	0.25	1.15	2.10	0.59	0.50	1.00×	parity	12
4	×4	1.03	1.58	0.65	0.31	1.34	1.83	0.97	0.26	1.30×	not significant	12
20	×1	2.00	2.95	0.90	0.61	1.46	3.22	1.05	0.56	0.73×	not significant	12
20	×4	2.74	3.25	2.34	0.27	2.55	3.87	2.03	0.63	0.93×	parity	12
50	×1	2.23	2.55	1.74	0.27	1.64	2.29	1.53	0.23	0.74×	CL faster	12
50	×4	4.11	4.83	3.67	0.29	3.89	4.96	3.32	0.54	0.95×	parity	12
100	×1	2.82	2.93	2.57	0.09	2.82	3.46	2.43	0.34	1.00×	parity	12
100	×4	9.85	10.3	8.93	0.48	5.85	6.88	5.26	0.48	0.59×	CL faster	12
200	×1	4.03	4.24	3.72	0.16	3.60	4.86	3.40	0.38	0.89×	CL faster	12
200	×4	14.4	15.7	13.6	0.55	9.24	11.1	8.07	0.78	0.64×	CL faster	12

14Real drag (CDP mouse) `splitter-drag-trusted · frameMs`

Trusted mouse events via CDP Input; median frame interval during the drag. 16.7 ms = 60 fps, 33 ms = 30 fps. reference metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	16.7	16.7	16.7	0.00	16.7	16.7	16.7	0.00	1.00×	parity	486
4	×4	16.7	16.7	16.7	2.36	16.7	16.7	16.7	0.00	1.00×	parity	985
20	×1	16.7	16.7	16.7	0.00	16.7	16.7	16.7	0.00	1.00×	parity	504
20	×4	16.7	33.3	16.7	10.3	25.0	33.3	16.7	8.34	1.50×	not significant	1116
50	×1	16.7	16.7	16.7	0.00	16.7	16.7	16.7	0.00	1.00×	parity	504
50	×4	16.7	33.3	16.7	11.7	25.0	33.3	16.7	9.42	1.50×	not significant	1354
100	×1	16.7	16.7	16.7	2.58	16.7	16.7	16.7	1.08	1.00×	parity	998
100	×4	50.0	66.7	16.7	23.8	33.3	66.7	16.7	24.7	0.67×	not significant	1390
200	×1	16.7	16.7	16.7	3.33	16.7	16.7	16.7	0.54	1.00×	parity	995
200	×4	50.0	66.7	16.7	25.1	33.3	66.7	16.7	25.0	0.67×	not significant	1401

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	16.7	16.7	16.7	1.07	16.7	16.7	16.7	0.76	1.00×	parity	482
4	×4	16.7	16.7	16.7	1.69	16.7	16.7	16.7	0.76	1.00×	parity	485
20	×1	16.7	16.7	16.7	0.00	16.7	16.7	16.7	0.00	1.00×	parity	480
20	×4	16.7	16.7	16.7	0.00	16.7	16.7	16.7	0.00	1.00×	parity	480
50	×1	16.7	16.7	16.7	0.00	16.7	16.7	16.7	0.00	1.00×	parity	480
50	×4	16.7	16.7	16.7	0.00	16.7	16.7	16.7	0.00	1.00×	parity	674
100	×1	16.7	16.7	16.7	0.00	16.7	16.7	16.7	0.00	1.00×	parity	481
100	×4	16.7	33.3	16.7	7.08	16.7	16.7	16.7	0.63	1.00×	parity	1012
200	×1	16.7	16.7	16.7	0.00	16.7	16.7	16.7	0.00	1.00×	parity	481
200	×4	16.7	33.3	16.7	6.96	16.7	16.7	16.7	0.93	1.00×	parity	1077

15Real drag: dropped frames `splitter-drag-trusted · droppedFrames`

Number of dropped frames per drag (median). reference metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	0	0	0	0	0	0	0	0	1.00×	parity	6
4	×4	2.5	3	2	0.5	0	0	0	0	<0.01×	CL faster	6
20	×1	0	0	0	0	0	0	0	0	1.00×	parity	6
20	×4	39	48	20	10.3	80.5	81	80	0.5	2.06×	DV faster	6
50	×1	0	0	0	0	0	0	0	0	1.00×	parity	6
50	×4	88.5	98	82	5.3	82	90	81	3.4	0.93×	parity	6
100	×1	1	1	1	0	0.5	1	0	0.5	0.50×	not significant	6
100	×4	184.5	191	177	4.7	83.5	91	82	3.4	0.45×	CL faster	6
200	×1	1.5	3	1	0.8	0	1	0	0.4	<0.01×	CL faster	6
200	×4	185	194	183	4.3	88.5	90	85	2.0	0.48×	CL faster	6

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	0.5	1	0	0.5	0	1	0	0.4	<0.01×	not significant	6
4	×4	0.5	1	0	0.5	0	1	0	0.4	<0.01×	not significant	6
20	×1	0	0	0	0	0	0	0	0	1.00×	parity	6
20	×4	0	0	0	0	0	0	0	0	1.00×	parity	6
50	×1	0	0	0	0	0	0	0	0	1.00×	parity	6
50	×4	0	0	0	0	0	0	0	0	1.00×	parity	6
100	×1	0	0	0	0	0	0	0	0	1.00×	parity	6
100	×4	39	41	37	1.4	0	1	0	0.4	<0.01×	CL faster	6
200	×1	0	0	0	0	0	0	0	0	1.00×	parity	6
200	×4	40	42	37	2.0	0.5	1	0	0.5	0.01×	CL faster	6

16Container resize (forced layout) `resize-forced · syncMs`

Resizing the stage with an explicit layout call — isolates the layout cost. ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	0.29	0.37	0.17	0.05	0.18	0.24	0.13	0.03	0.64×	CL faster	36
4	×4	1.14	1.49	0.23	0.32	0.68	1.08	0.10	0.33	0.60×	CL faster	36
20	×1	0.39	0.53	0.31	0.06	0.25	0.34	0.21	0.04	0.65×	CL faster	36
20	×4	1.63	2.44	0.65	0.38	1.06	1.61	0.23	0.33	0.65×	not significant	36
50	×1	0.49	0.61	0.41	0.06	0.25	0.29	0.22	0.02	0.51×	CL faster	36
50	×4	1.84	2.46	1.02	0.36	1.06	1.68	0.60	0.32	0.57×	CL faster	36
100	×1	0.85	1.10	0.69	0.15	0.45	0.54	0.32	0.05	0.53×	CL faster	36
100	×4	3.56	3.88	2.78	0.32	2.05	2.79	1.41	0.38	0.57×	CL faster	36
200	×1	0.92	1.05	0.76	0.09	0.47	0.56	0.39	0.04	0.51×	CL faster	36
200	×4	3.88	4.63	2.95	0.47	1.95	2.72	1.08	0.38	0.50×	CL faster	36

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	0.50	1.38	0.23	0.34	0.43	0.75	0.14	0.20	0.86×	not significant	36
4	×4	0.85	1.20	0.15	0.27	0.43	0.95	0.08	0.31	0.51×	CL faster	36
20	×1	0.60	1.10	0.39	0.22	0.43	1.29	0.33	0.27	0.72×	not significant	36
20	×4	1.43	2.17	0.91	0.38	0.95	1.54	0.18	0.35	0.66×	CL faster	36
50	×1	0.62	0.91	0.46	0.11	0.39	0.51	0.29	0.06	0.63×	CL faster	36
50	×4	1.88	2.47	0.90	0.42	1.07	1.67	0.18	0.31	0.57×	CL faster	36
100	×1	1.07	1.41	0.78	0.19	0.75	0.96	0.59	0.11	0.70×	CL faster	36
100	×4	3.39	4.16	2.25	0.46	1.90	2.31	1.05	0.32	0.56×	CL faster	36
200	×1	0.97	1.28	0.75	0.13	0.70	0.92	0.52	0.10	0.72×	CL faster	36
200	×4	3.71	4.36	2.99	0.44	1.84	2.27	1.02	0.36	0.49×	CL faster	36

17Resize (via ResizeObserver) `resize-observed · readyMs`

Resize observer only. dockview does ResizeObserver → rAF → layout, so its readyMs includes a frame of waiting. Do not rank — read together with framesToSettle. reference metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	22.1	24.9	20.2	1.34	0.16	0.23	0.12	0.03	<0.01×	CL faster	36
4	×4	31.8	34.6	29.8	1.34	0.52	1.04	0.09	0.40	0.02×	CL faster	36
20	×1	24.6	31.9	22.5	2.38	0.24	0.29	0.19	0.02	<0.01×	CL faster	36
20	×4	45.4	79.7	41.9	14.0	1.03	1.56	0.25	0.32	0.02×	CL faster	36
50	×1	23.6	29.1	21.7	2.21	0.25	0.37	0.21	0.04	0.01×	CL faster	36
50	×4	58.4	88.9	37.4	17.1	1.25	1.77	0.60	0.33	0.02×	CL faster	36
100	×1	31.4	33.7	22.2	2.84	0.45	0.55	0.36	0.05	0.01×	CL faster	36
100	×4	99.0	151	83.0	21.5	2.13	2.65	1.23	0.37	0.02×	CL faster	36
200	×1	30.4	35.6	22.6	3.78	0.47	0.59	0.36	0.06	0.02×	CL faster	36
200	×4	111	153	91.0	21.6	1.90	2.88	1.07	0.46	0.02×	CL faster	36

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	18.7	20.8	15.5	1.37	0.32	0.90	0.15	0.23	0.02×	CL faster	36
4	×4	18.1	19.6	16.6	0.81	0.28	0.99	0.08	0.32	0.02×	CL faster	36
20	×1	18.3	20.5	16.1	1.15	0.40	1.00	0.32	0.22	0.02×	CL faster	36
20	×4	19.6	21.4	17.9	0.88	0.94	1.35	0.16	0.35	0.05×	CL faster	36
50	×1	18.7	19.8	16.8	0.95	0.44	0.58	0.30	0.07	0.02×	CL faster	36
50	×4	20.0	22.4	12.0	1.65	0.89	1.45	0.21	0.28	0.04×	CL faster	36
100	×1	19.4	20.8	17.7	0.84	0.75	0.94	0.58	0.10	0.04×	CL faster	36
100	×4	20.6	23.9	17.9	1.77	1.77	2.36	1.11	0.33	0.09×	CL faster	36
200	×1	19.8	21.3	17.1	1.02	0.71	0.88	0.55	0.10	0.04×	CL faster	36
200	×4	25.4	41.4	17.5	4.55	1.83	2.41	1.06	0.38	0.07×	CL faster	36

18Moving a panel between groups `move · syncMs`

Moving a panel to another group. In CustomLayout: getLayout → moveItemToStack → loadLayout. ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
20	×1	9.78	10.6	8.81	0.54	9.64	10.5	8.88	0.48	0.99×	parity	36
20	×4	37.9	40.8	34.1	1.71	37.9	42.2	35.6	3.52	1.00×	parity	36
50	×1	10.1	10.7	9.02	0.49	10.9	12.2	9.88	0.58	1.09×	parity	36
50	×4	38.6	41.9	35.5	4.66	42.3	62.6	39.3	6.58	1.09×	parity	36
100	×1	10.1	10.6	9.74	0.24	11.8	12.4	11.1	0.34	1.17×	DV faster	36
100	×4	40.9	43.9	38.7	1.47	46.0	73.6	42.6	7.45	1.12×	DV faster	36
200	×1	10.2	10.7	9.78	0.22	12.6	13.4	11.6	0.50	1.23×	DV faster	36
200	×4	40.5	43.3	38.4	1.29	49.9	52.4	45.1	1.56	1.23×	DV faster	36

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
20	×1	2.03	3.24	0.99	0.69	2.53	3.03	2.03	0.30	1.25×	not significant	36
20	×4	2.07	3.08	1.01	0.45	5.37	6.93	3.90	1.08	2.59×	DV faster	36
50	×1	1.60	3.30	1.10	0.71	3.47	3.98	2.67	0.36	2.18×	DV faster	36
50	×4	2.78	3.61	1.86	0.52	7.65	9.35	6.13	0.76	2.75×	DV faster	36
100	×1	1.23	1.50	0.89	0.25	4.79	5.92	3.64	0.57	3.88×	DV faster	36
100	×4	3.10	3.99	2.18	0.44	10.6	11.8	9.10	0.89	3.42×	DV faster	36
200	×1	1.34	1.94	1.09	0.28	5.74	6.77	4.50	0.68	4.27×	DV faster	36
200	×4	3.75	4.86	2.36	0.53	14.6	15.7	12.6	0.75	3.90×	DV faster	36

19Serialization `serialize · serializeUs`

toJSON() / getLayout(), microseconds per call (300 calls per measurement). getLayout makes a deep copy. ranked metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	0.86	1.22	0.72	0.13	8.25	9.25	7.20	0.57	9.61×	DV faster	36
4	×4	2.17	3.52	0.55	1.03	24.3	26.7	22.9	1.25	11.2×	DV faster	36
20	×1	1.82	2.05	1.58	0.15	17.6	18.2	17.1	0.39	9.67×	DV faster	36
20	×4	6.16	9.83	3.43	1.57	71.0	79.8	69.0	2.37	11.5×	DV faster	36
50	×1	3.01	3.60	2.75	0.81	33.4	36.5	31.5	1.38	11.1×	DV faster	36
50	×4	11.8	54.3	9.07	14.2	130	141	125	4.57	11.0×	DV faster	36
100	×1	6.07	7.32	5.42	0.49	66.7	70.1	64.3	1.43	11.0×	DV faster	36
100	×4	23.0	28.6	20.6	2.34	259	276	249	7.53	11.3×	DV faster	36
200	×1	11.0	17.6	10.6	1.91	115	120	113	1.83	10.5×	DV faster	36
200	×4	41.5	54.7	39.3	4.21	452	474	432	11.6	10.9×	DV faster	36

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	1.08	3.22	0.68	0.67	9.82	14.1	6.45	2.21	9.07×	DV faster	36
4	×4	1.87	3.18	0.53	1.03	24.7	25.8	23.0	0.74	13.2×	DV faster	36
20	×1	2.43	3.27	2.02	0.35	19.3	23.2	17.7	1.70	7.92×	DV faster	36
20	×4	6.23	8.20	3.67	1.11	70.5	72.3	67.8	1.20	11.3×	DV faster	36
50	×1	3.46	4.33	3.13	0.35	33.5	37.4	31.6	1.38	9.69×	DV faster	36
50	×4	10.8	13.6	8.03	1.67	130	137	127	2.56	12.1×	DV faster	36
100	×1	6.15	7.35	5.87	0.48	67.2	69.9	65.5	1.01	10.9×	DV faster	36
100	×4	23.1	29.9	21.4	2.88	262	273	252	5.72	11.4×	DV faster	36
200	×1	10.8	12.3	10.4	0.57	116	118	113	1.23	10.7×	DV faster	36
200	×4	42.3	80.8	39.1	18.5	459	476	443	9.73	10.9×	DV faster	36

20Serialization + JSON.stringify `serialize-stringify · serializeUs`

Same, plus JSON.stringify of the result. reference metric

CPU ×1

CPU ×4

Table: heavy panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	1.98	2.72	1.53	0.28	8.95	9.63	7.90	0.51	4.51×	DV faster	36
4	×4	5.90	7.52	3.57	1.04	27.5	29.3	24.3	1.19	4.66×	DV faster	36
20	×1	9.48	10.8	8.70	0.62	19.9	20.4	18.9	0.44	2.10×	DV faster	36
20	×4	33.8	39.2	31.0	2.19	78.8	83.3	75.2	2.09	2.33×	DV faster	36
50	×1	16.1	23.8	15.2	3.15	36.5	37.9	35.1	0.62	2.27×	DV faster	36
50	×4	65.5	96.6	60.3	10.5	141	152	137	4.00	2.15×	DV faster	36
100	×1	31.7	33.7	30.8	0.94	72.9	74.7	71.0	1.10	2.30×	DV faster	36
100	×4	128	136	123	4.03	283	306	274	9.73	2.21×	DV faster	36
200	×1	56.9	59.5	55.6	0.82	125	127	123	3.57	2.20×	DV faster	36
200	×4	226	233	216	3.91	499	524	474	16.3	2.21×	DV faster	36

CPU ×1

CPU ×4

Table: light panels

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
4	×1	2.47	6.95	1.52	1.40	10.4	12.2	6.88	2.00	4.20×	DV faster	36
4	×4	6.25	8.03	3.43	1.27	27.4	28.5	25.5	1.04	4.38×	DV faster	36
20	×1	13.2	15.6	11.3	1.11	20.7	24.2	19.2	1.46	1.57×	DV faster	36
20	×4	33.9	37.7	32.0	1.53	78.2	84.1	74.9	2.53	2.31×	DV faster	36
50	×1	19.6	21.6	18.7	0.85	36.8	38.8	35.7	0.88	1.88×	DV faster	36
50	×4	63.7	69.4	59.9	3.05	145	151	139	3.16	2.28×	DV faster	36
100	×1	31.4	35.6	30.5	1.48	73.4	75.1	71.9	0.83	2.33×	DV faster	36
100	×4	126	136	120	4.00	288	300	276	6.48	2.28×	DV faster	36
200	×1	57.0	60.1	55.6	1.39	126	127	123	1.10	2.20×	DV faster	36
200	×4	227	236	219	7.19	496	521	484	11.1	2.18×	DV faster	36

21Mount/unmount cycles `mount-cycles · cycleMs`

8 cycles of "mount 50 panels → unmount the whole engine", two rounds per pass with alternating engine order. ranked metric

CPU ×1

CPU ×4

Size	CPU	dockview				CustomLayout				CL / DV	Verdict	n
Size	CPU	median	p95	min	σ	median	p95	min	σ	CL / DV	Verdict	n
light, 50	×1	133	137	118	4.64	133	137	118	4.08	1.00×	parity	48
light, 50	×4	150	154	140	3.04	135	150	128	6.76	0.90×	parity	48
heavy, 50	×1	252	270	219	13.7	169	188	161	7.66	0.67×	CL faster	48
heavy, 50	×4	686	712	640	17.8	419	440	399	14.6	0.61×	CL faster	48

05Scaling

Median by panel count (4, 20, 50, 100, 200; sizes with tabs in groups, see the size table). Two charts per operation: CPU ×1 and ×4. Panel type follows the toggle above (currently: heavylight).

Cold mount

Restore (new ids)

Cleanup

Removal, per operation

Splitter step

Serialization

Cold mount

Restore (new ids)

Cleanup

Removal, per operation

Splitter step

Serialization

How many panels are actually mounted

alivePanels after cold mount: dockview / CustomLayout. This explains why the CustomLayout curves grow more slowly: it mounts only active tabs.

Panels	4	20	flat-20	50	100	200	tabs-20
heavy	4 / 4	20 / 10	20 / 20	50 / 10	100 / 20	200 / 20	21 / 2
light	4 / 4	20 / 10	20 / 20	50 / 10	100 / 20	200 / 20	21 / 2

06Memory and leaks

Neither engine shows DOM leaks: after 8 mount/unmount cycles of 50 panels there are 0 live panels, and growth in document nodes and body children is 0. Heap: for dockview after a forced GC the median growth is +2.4 MB (light) and +15.4–30.8 MB (heavy, ×4 and ×1), but the spread is large (heavy at ×4 ranges from 0.2 to 15.9 MB). A slope of ~15–25 KB per cycle adds only ~0.1–0.2 MB over 8 cycles, so the growth is more likely one-off retention (caches/closures) than a linear leak. For CustomLayout the median growth is negative (−0.6…−6.2 MB), but for heavy panels the range reaches +3.7 MB, and the slope at ×4 is unstable (median +943 KB/cycle, spread from −7 KB to +1.9 MB/cycle). With 8 cycles and 6 measurements nothing can be definitively called a leak — a long run (20+ cycles) is needed. CDP usedSize before/after depends heavily on garbage left by the previous scenario (passes 1 and 3 start with a large heap), so do not use it to compare engines.

Panels	CPU	Metric	dockview, median [min … max]	CustomLayout, median [min … max]
light	×1	Heap growth after 8 cycles	2.4 MB [783.6 KB … 2.4 MB]	-937.2 KB [-2.0 MB … 21.9 KB]
light	×1	Heap slope per cycle	24.0 KB [4.8 KB … 26.2 KB]	857 B [584 B … 1.1 KB]
light	×1	Live panels after unmount	0 [0 … 0]	0 [0 … 0]
light	×1	Δ document nodes	0 [0 … 0]	0 [0 … 0]
light	×1	Δ body children	0 [0 … 0]	0 [0 … 0]
light	×1	Cycle time	133 [118 … 139]	133 [118 … 137]
light	×4	Heap growth after 8 cycles	2.4 MB [1.3 MB … 2.4 MB]	-649.0 KB [-1.5 MB … 22.0 KB]
light	×4	Heap slope per cycle	24.5 KB [4.9 KB … 25.1 KB]	601 B [-4.8 KB … 1.2 KB]
light	×4	Live panels after unmount	0 [0 … 0]	0 [0 … 0]
light	×4	Δ document nodes	0 [0 … 0]	0 [0 … 0]
light	×4	Δ body children	0 [0 … 0]	0 [0 … 0]
light	×4	Cycle time	150 [140 … 156]	135 [128 … 151]
heavy	×1	Heap growth after 8 cycles	30.8 MB [16.0 MB … 31.7 MB]	-6.2 MB [-16.1 MB … 3.7 MB]
heavy	×1	Heap slope per cycle	14.9 KB [-7.7 KB … 50.7 KB]	-3.3 KB [-1.9 MB … 1.9 MB]
heavy	×1	Live panels after unmount	0 [0 … 0]	0 [0 … 0]
heavy	×1	Δ document nodes	0 [0 … 0]	0 [0 … 0]
heavy	×1	Δ body children	0 [0 … 0]	0 [0 … 0]
heavy	×1	Cycle time	252 [219 … 272]	169 [161 … 191]
heavy	×4	Heap growth after 8 cycles	15.4 MB [208.5 KB … 15.9 MB]	-6.2 MB [-16.1 MB … 3.7 MB]
heavy	×4	Heap slope per cycle	15.1 KB [3.7 KB … 37.1 KB]	943.2 KB [-7.3 KB … 1.9 MB]
heavy	×4	Live panels after unmount	0 [0 … 0]	0 [0 … 0]
heavy	×4	Δ document nodes	0 [0 … 0]	0 [0 … 0]
heavy	×4	Δ body children	0 [0 … 0]	0 [0 … 0]
heavy	×4	Cycle time	686 [640 … 732]	419 [399 … 474]

CDP Runtime.getHeapUsage around mount-cycles (per engine)

Pass	CPU	dockview Δ usedSize	CustomLayout Δ usedSize	Order
1	×1	-48.2 MB	-15.9 MB	dockview → custom
2	×1	16.0 MB	-2.2 MB	custom → dockview
3	×1	-48.2 MB	-16.0 MB	dockview → custom
1	×4	-2.2 MB	27.5 KB	dockview → custom
2	×4	42.3 KB	-2.2 MB	custom → dockview
3	×4	-2.2 MB	42.4 KB	dockview → custom

07DOM nodes

Counted after mount, CPU ×1, dockview / CustomLayout. "Chrome" means engine nodes excluding panel content (splitters, headers, tabs). Panel type follows the toggle.

"chrome" nodes (excluding panel content)

Size	live panels	visible	detached	nodes in scene	panel nodes	"chrome" nodes	off-scene nodes
4	4 / 4	4 / 4	0 / 0	5,719 / 5,724	5,612 / 5,612	107 / 112	0 / 0
20	20 / 10	10 / 10	10 / 0	14,321 / 14,406	14,030 / 14,030	291 / 376	0 / 0
flat-20	20 / 20	20 / 20	0 / 0	28,521 / 28,614	28,060 / 28,060	461 / 554	0 / 0
50	50 / 10	10 / 10	40 / 0	14,511 / 14,716	14,030 / 14,030	481 / 686	0 / 0
100	100 / 20	20 / 20	80 / 0	29,101 / 29,440	28,060 / 28,060	1,041 / 1,380	0 / 0
200	200 / 20	20 / 20	180 / 0	29,701 / 30,440	28,060 / 28,060	1,641 / 2,380	0 / 0
tabs-20	21 / 2	2 / 2	19 / 0	2,984 / 3,053	2,806 / 2,806	178 / 247	0 / 0
4	4 / 4	4 / 4	0 / 0	119 / 124	12 / 12	107 / 112	0 / 0
20	20 / 10	10 / 10	10 / 0	321 / 406	30 / 30	291 / 376	0 / 0
flat-20	20 / 20	20 / 20	0 / 0	521 / 614	60 / 60	461 / 554	0 / 0
50	50 / 10	10 / 10	40 / 0	511 / 716	30 / 30	481 / 686	0 / 0
100	100 / 20	20 / 20	80 / 0	1,101 / 1,440	60 / 60	1,041 / 1,380	0 / 0
200	200 / 20	20 / 20	180 / 0	1,701 / 2,440	60 / 60	1,641 / 2,380	0 / 0
tabs-20	21 / 2	2 / 2	19 / 0	184 / 253	6 / 6	178 / 247	0 / 0

08Bundle size

In the app build both engines are bundled into the main index chunk, so sizes are taken from separate library builds of each engine (vite, same minification config).

JS and CSS, gzip / brotli compression

	dockview	CustomLayout	DV / CL
JS raw	361.5 KB	42.5 KB	8.5×
JS gzip	86.3 KB	14.2 KB	6.1×
JS brotli	72.0 KB	12.7 KB	5.7×
CSS gzip	11.0 KB	4.4 KB	2.5×
CSS brotli	9.2 KB	3.9 KB	2.4×
Source modules	dockview-core 700.6 KB, dockview-vue 18.4 KB	custom-layout 90.2 KB
App index chunk	2.440.8 KB raw / 730.6 KB gzip (both engines inside)

Against the ~731 KB gzip main chunk, the difference of ~79 KB is about 11% of the bundle.

09Features: speed isn't everything

dockview is a mature library with a broad API; CustomLayout (~3.7k lines of TS/Vue and ~0.8k lines of SCSS themes in src/components/custom-layout) covers what the terminal needs today, but everything else would have to be written and maintained in-house.

Capability	dockview 8.4	CustomLayout
Nested row/column splits	yes	yes (MAX_TREE_DEPTH, config validation)
Tabbed groups	yes	yes (stacks)
Tab DnD between groups and to edges	yes, customizable overlays and zones	yes: group zones and root edges
Dragging a whole group	yes	yes
Touch dragging	limited — relies on HTML5 DnD	yes, custom `use-touch-drag`
Floating groups	yes (now persisted in the workspace)	no
Popout to a separate window	yes	no
Maximize group	yes (maximize)	yes (stack fullscreen)
Tab overflow	yes (dropdown list)	yes ("+N" menu, menu item slot)
Lazy mounting of hidden tabs	no — a component is created for every panel (alivePanels = N)	yes — only tabs that have ever been visible
Fine-grained API (activate, move, addPanel to a group, sizes)	yes, plus dozens of events (onDidActivePanelChange, onDidLayoutChange, onWillDrop…)	minimal: add/remove/load/get/clear; activation and moving go through a full config reload
Serialization	toJSON — ~11 µs at 200 panels	getLayout — deep copy, ~115 µs
Minimum sizes	in pixels (group no smaller than 100 px)	in percent (no less than 5%); groups keep shrinking
Accessibility	partial: aria-live announcements (disabled in the app), limited keyboard navigation	minimal
Themes	built-in themes + CSS variables `--dv-*`	7 themes (default, light, terminal, sunset, gl, midnight, ocean)
Tab customization	tab/header components, header actions, watermark	`tabComponent`, overflow menu slot
Other layout types	gridview, splitview, paneview	no
Support and updates	external dependency; the 5.2 → 8.4 upgrade required fixes and surfaced a blocker	own code — full control, all maintenance and testing on the team
Size (JS gzip)	86.3 KB	14.2 KB

10dockview 5.2 → 8.4 upgrade

Changes

dockview-vue 5.2.0 → 8.4.0, plus an explicit dependency dockview 8.4.0; dockview-core is no longer imported directly.
Style import moved to dockview-vue/dist/styles/dockview.css.
Payload onDidActivePanelChange is now { panel, origin }.
The component root became a fragment — the layout's scoped styles were moved to :deep().
floating-group-drag-handle="tabbar" — to keep the 5.x look.
Light theme: color-scheme: light and --dv-floating-border.
Floating groups are now persisted (dockviewToGl / glToDockview — floatingGroups).
.dv-shell got isolation: isolate.
New spec dockview-integration.spec.ts (8 tests).

Blocker found in the browser smoke test

Vue casts an omitted Boolean prop to false. As a result, proportionalLayout became false, and core 8.4 produced NaN sizes for nested splits; the same cast disabled the root edge drop zones. Fixed with explicit props:

<DockviewVue
  :proportional-layout="true"
  :dnd-edges="undefined"
  :announcements="false" />

Also, panels are mounted via Teleport one tick after addPanel/fromJSON — the benchmark waits for nextTick + requestAnimationFrame before stopping the timers.

Checks

Unit tests 229/229 after the upgrade (271/271 after adding the benchmark); the build passes.
Typecheck: 99 errors, all pre-existing, no new ones.
Smoke: /dockview-demo 18/18; /trading-dockview 8/10 — 2 long-standing issues (a closed tab is not hidden in the spec after reload; report table console errors caused by the mock).
Known and not fixed: glToDockview ignores missing and percentage sizes.

11How to rerun

In the browser

VITE_LAYOUT_BENCH=1 npm run build
npx vite preview --port 5050
# open http://localhost:5050/layout-benchmark

The /layout-benchmark route is available only in a build with VITE_LAYOUT_BENCH=1. The page offers selection of scenarios, sizes and engines, progress, and a results table with JSON export. From the console:

await window.__layoutBench.run({
  scenarios: ['cold-mount', 'restore'],
  sizes: [4, 50, 200],
})

Important: after the run, rebuild without the flag — dist/ currently contains a build with the benchmark route.

Full run via Playwright

Script: perf-reports/layout-benchmark/run.mjs (the runner for this run). It builds the project, measures bundle size, starts preview on :5050, mocks all /api/* (non-GET requests are blocked), runs passes with CDP throttling and real drag, and writes results.json next to itself.

cd perf-reports/layout-benchmark
PLAYWRIGHT_DIR=/path/to/playwright node run.mjs --isolate \
  --passes=3 --throttles=1,4 --trusted-repeats=5
# quick run:        --quick
# selected only:    --scenarios=restore,remove-all
# no rebuild:       --no-build
# custom config:    --config='{"sizes":[4,100]}'

Next to the script are mocks.mjs (mocks for /api/*) and size/ (a separate engine build for bundle measurement). Playwright is not a project dependency: install it in any folder (npm i playwright) and pass the path via PLAYWRIGHT_DIR. The defaults (10/30 iterations, 20 cycles) take ~5 h; this report was produced with reduced values via --config (the command is in the header of run.mjs).

12Raw data

The page embeds JSON (<script type="application/json" id="bench-results">): meta, env, bundle, safety, all 2344 aggregates, comparisons, the re-check, and pass summaries (CDP heap, trusted drag without per-sample arrays). Per-sample data (the full results.json, 13 MB) is not embedded due to its size.

Generated from results.json and env.json · run 2026-10-04T00:08:25.617Z — 2026-10-04T01:31:36.053Z · local data only, no external scripts.

01Summary

CustomLayout is faster

dockview is faster

Parity

02Environment and methodology

Layout sizes

Iteration count reduced

Data quality checks

03Caveats and fairness of the comparison

Fixed before the run

Remaining limitations

04Scenarios

1Cold mount cold-mount · syncMs

2Restore: new snapshot restore-replace · syncMs

3Restore: same ids restore-same-ids · syncMs

4Clearing the layout clear · syncMs

5Removal one at a time (per operation) remove-sequential · perRemoveSyncMs

6Removing all N in one block remove-sequential · syncMs

7Adding a panel to the root add-k · syncMs

8Adding a tab to a group add-k-tab · syncMs

9Switching a tab via the API tab-switch-api · syncMs

10First tab open tab-switch-first · syncMs

11Tab click (visible latency) tab-switch-warm · readyMs

12Splitter drag: step splitter-drag · stepMs

13Splitter drag: release splitter-drag · commitMs

14Real drag (CDP mouse) splitter-drag-trusted · frameMs

15Real drag: dropped frames splitter-drag-trusted · droppedFrames

16Container resize (forced layout) resize-forced · syncMs

17Resize (via ResizeObserver) resize-observed · readyMs

18Moving a panel between groups move · syncMs

19Serialization serialize · serializeUs

20Serialization + JSON.stringify serialize-stringify · serializeUs

21Mount/unmount cycles mount-cycles · cycleMs

05Scaling

How many panels are actually mounted

06Memory and leaks

CDP Runtime.getHeapUsage around mount-cycles (per engine)

07DOM nodes

08Bundle size

09Features: speed isn't everything

10dockview 5.2 → 8.4 upgrade

Changes

Blocker found in the browser smoke test

Checks

11How to rerun

In the browser

Full run via Playwright

12Raw data

1Cold mount `cold-mount · syncMs`

2Restore: new snapshot `restore-replace · syncMs`

3Restore: same ids `restore-same-ids · syncMs`

4Clearing the layout `clear · syncMs`

5Removal one at a time (per operation) `remove-sequential · perRemoveSyncMs`

6Removing all N in one block `remove-sequential · syncMs`

7Adding a panel to the root `add-k · syncMs`

8Adding a tab to a group `add-k-tab · syncMs`

9Switching a tab via the API `tab-switch-api · syncMs`

10First tab open `tab-switch-first · syncMs`

11Tab click (visible latency) `tab-switch-warm · readyMs`

12Splitter drag: step `splitter-drag · stepMs`

13Splitter drag: release `splitter-drag · commitMs`

14Real drag (CDP mouse) `splitter-drag-trusted · frameMs`

15Real drag: dropped frames `splitter-drag-trusted · droppedFrames`

16Container resize (forced layout) `resize-forced · syncMs`

17Resize (via ResizeObserver) `resize-observed · readyMs`

18Moving a panel between groups `move · syncMs`

19Serialization `serialize · serializeUs`

20Serialization + JSON.stringify `serialize-stringify · serializeUs`

21Mount/unmount cycles `mount-cycles · cycleMs`