vuejs

devtools

⚙️ Devtools for debugging Vue.js applications.

by vuejs

TypeScript

TypeScript78.2%
Vue20.3%
CSS0.9%
HTML0.3%
JavaScript0.3%

Commit activity

Last 12 weeks · 53 commits

Full graph →

Community health

2 of 6 standards met

Community profile →
37
✓README✓License○Contributing○Code of Conduct○Issue Template○PR Template

Recent PRs & issues

Active · Last activity 1h ago
See all on GitHub →
chjans

fix: prevent duplicate DevTools server listener registration ##created with help of AI Problem Repeated navigation in an Inertia/Vue application causes the Vue Devtools Chrome extension (7.7.7) to become progressively slower. Investigation revealed that can be invoked repeatedly through the birpc interface. Each invocation registers additional handlers on , without checking whether those handlers have already been registered. After several navigations, the internal hook registry contained 45 distinct handlers for each of the following events: This caused thousands of redundant inspector and router messages, despite relatively few Vue component lifecycle events. Fix Make idempotent using a module-scoped initialization guard. Subsequent RPC calls no longer register duplicate listeners. Verification Reproduced with Vue 3.5.42 and Inertia 2.1.6. Reproduced using a minimal Vue application without Pinia or Vue Router. Confirmed repeated RPC invocations using Chrome's debugger. Confirmed accumulation of 45 distinct callbacks per internal messaging hook. Verified that deduplicating listeners and preventing further registration resolves the performance degradation. Considerations The guard assumes that the underlying RPC broadcast instance remains valid across subsequent initialization requests. Connection teardown and reconnection behavior should be verified before merging.

chjans · 2d ago
ValentinYoushkevich

Summary Closes #975. used to serve the v6 docs, and the early v7 guide pages moved in #382, so those URLs now return 404. The v6 extension still links to them: its popup's Troubleshooting link goes to (), and its open-in-editor console hint points to . This adds , which VitePress copies into the published output ( and do the same). v6 pages (, ) go to the same path on , and the early v7 pages without go to their current pages. Every source URL returns 404 today and every target returns 200.

ValentinYoushkevich · 2d ago
webfansplz

Summary Classify Vapor nodes as render effects and attribute them to the owning component through , preserving VDOM support. Fix Options API computed recomputation on Vue 3.6 by accepting pointing directly to a computed ref, alongside Vue 3.5's shape. Update the Vapor support plan with completed work and remaining gaps. Verification — passed. — 77 test files and 460 tests passed. — passed. — passed. The new regression tests fail without their corresponding fixes. Removing the graph changes fails the Vapor case while the VDOM case still passes. Notes Vue 3.6.0-rc.14 has no , so no scan of that field is added. The computed collector serves Options API computed lookup; Vapor does not support Options API computed. Nested Teleport picking remains pending a Vue release containing vuejs/core#16142. 🤖 Generated with Claude Code

webfansplz · 5d ago
webfansplz

Problem A 10,000-component tree can produce a response of about 2.9 MB, exceeding the default 2 MiB transport budget. The panel cannot finish initialization and may reconnect and retry the same failing snapshot. Repeated absolute paths account for about 36% of the measured payload. Changes Without component-tree hooks, plan a bounded snapshot during collection and reduce its depth only when needed to fit the budget. With hooks, collect the original tree and invoke hooks once in the original order, then fit the already-decorated result. Hooks are not rerun for the reduced subset. Return a connected, shallower skeleton with real child counts. Keep visibility and expansion tracking aligned with the returned nodes, and load remaining branches through the existing pagination. Add optional only to reduced responses. The client preserves this marker, and the agent result schema accepts it. Add an internal prop to , defaulting to false. Only the components page enables it, and only for a reduced snapshot, so unloaded branches offer an expand action without changing custom inspector, Pinia, or normal snapshot defaults. Add read-only hook-presence detection through , including hooks registered during deferred plugin setup. Behavior Snapshots that fit the budget retain their previous serialized content byte for byte, including hook-decorated results. The 2 MiB limit, existing semantics, generic RPC oversize rejection, and the server's final size check remain unchanged. A root node that alone exceeds the budget still fails that check. Runtime child-count mutations continue to invalidate the event-buffer measurement cache, and nodes exposed to plugin hooks have their trust revoked before mutation. Reducing a snapshot changes the response envelope and node array without mutating cached patch nodes. Verification Fresh real Chromium runs against this branch's built packages opened the panel and used the actual UI expand actions to load every node: Both initial responses carried . All child pages stayed below 2 MiB, with no browser errors or page reloads. The response sizes were captured from the panel's actual RPC requests; node totals include the App root and, for the branched tree, its 100 branches. Limitations The hook path still collects the full tree before fitting it. Hooks observe the original nodes, including nodes later omitted from an oversized response; they are called once, not again for the retained subset. The client still lacks a snapshot-version lower-bound filter for component patches. The pre-existing race with delayed patches is outside this change; regression coverage verifies that constructing a reduced hook snapshot does not itself dispatch events. Absolute file paths are not deduplicated. Tests Validated in an isolated worktree based on current , including the merged deferred-setup and bulk-update changes: — 78 files, 506 tests passed Formatting check for all 17 changed files Normal pre-commit hooks Real Chromium panel opening and complete expansion for both 10,000-leaf shapes Regression coverage includes byte-identical 1,000/5,000-node snapshots, exact budget boundaries, hook order and non-repetition, deferred hook registration, real child counts, hidden updates, deeper initial-depth settings, client marker resets, and the opt-in collapsed state. Browser artifacts remain outside the repository.

webfansplz · 1h ago
webfansplz

Problem With the v9 panel open, replacing a keyed tree of about 5,000 components can take roughly 5.4 seconds from the app update to . This is elapsed update latency, not one synchronous main-thread task. Two quadratic paths dominate bulk mount and unmount: the event buffer serializes the entire accumulated patch list for each incoming event, and the DOM-order cache repeatedly invalidates and rescans the whole parent. Changes Accumulate newly added patch bytes and the event envelope instead of reserializing the growing patch prefix. Invisible runtime trust markers enable this path; child-count mutations invalidate measurements, and exposing nodes to plugin hooks revokes trust. Untrusted objects retain the original serialization fallback. Retain the valid prefix of each DOM-order cache and rebuild only as much of the suffix as a lookup needs. Mutation records invalidate the affected prefix, removed nodes are deleted, and stale cached positions are checked before use. Behavior For runtime-generated events, serialized wire content, event order, budget decisions, and backpressure transitions match the previous implementation byte for byte. Legacy-oracle tests cover insert floods, later child-count mutations, hook changes and retained references, and DOM reorder sequences. A real Chromium shadow comparison also checked DOM-order output and final tree order on the 5,000-component workload, including keyed replacement. Fragment, Teleport, KeepAlive, and Vapor cases are covered by the unit oracle. This PR contains only the event-buffer and DOM-order optimizations. Initial-snapshot budget fallback and client changes are excluded. Benchmark Five-run medians on the same Apple M4 machine with Chromium 156: The broader optimization series reduced v9-open render elapsed time from 26,965 ms to 8,777 ms, and update p99 from 5,040 ms to 244 ms. Those before/after figures include separately scoped changes and are not an isolated estimate of this PR. A fresh, interleaved experiment isolates the steps: A = main with deferred plugin setup: 30.90 s; B = A plus incremental event buffering: 15.22 s; C = B plus DOM-order caching: 9.16 s. Adding the separate snapshot-budget work produced D = 9.22 s. In that experiment, A-to-D update p95 rose about 11% amid substantial run-to-run noise, while frame p95 stayed approximately unchanged. Paired p95 differences did not establish a stable regression. Diagnostic traces showed major GC landing in ordinary-update windows and accounting for most of the sampled p95 difference; this does not establish GC timing as the sole cause. B-to-C ordinary-update p50 had a small, unresolved increase of about 1 ms. Limitations The public and APIs expose internal objects. Arbitrary in-place mutations by third-party code in the same process can invalidate cached byte measurements; this is not a supported mutation path. Runtime child-count changes and plugin-hook exposure are handled explicitly. Tests Validated this commit in an isolated worktree based on current : — 78 files, 486 tests passed Formatting check for all six changed files Normal pre-commit hooks Earlier validation includes legacy-oracle comparisons, structural-patch and DOM-insertion linear-growth checks, and real Chromium shadow comparisons. Benchmark and trace artifacts remain outside the repository.

webfansplz · 2h ago
webfansplz

Problem v9 runs plugin setup before DevTools is opened. Pinia installs synchronous deep watchers during setup, making 5,000 Map writes stall the app for about 14 seconds even with the panel closed. v8 delays plugin setup by default. Changes Defer setup until the first panel connection or a public runtime query, command, or custom request. Route plugin API internals through captured raw requests so internal work does not activate plugins. Allow setup to reenter public requests without waiting on its own activation. Emit after the initial setup flush and refresh client discovery after the first snapshot, preserving inspector and timeline-layer notifications. Behavior change Setup timing returns to v8 semantics. The Chrome extension performs one additional detection refresh on first activation. Concurrent requests during the activation window do not wait for setup to finish; the request that initiates activation does wait. Benchmark For 5,000 Pinia Map writes with DevTools closed, the median action duration dropped from 14,334 ms to 3.6 ms; v8 measured 3.4 ms. Results are five-run medians on the same Apple M4 machine with Chromium 156. Tests Validated this commit alone on a fresh worktree based on : — 78 files, 468 tests passed Build ran before lint and typecheck to generate the workspace packages' exported declarations. Regression tests cover lazy activation, setup reentry, lifecycle cleanup, and client refresh after asynchronous plugin discovery.

webfansplz · 3h ago
Structured data for AI agents

Repository: vuejs/devtools. Description: ⚙️ Devtools for debugging Vue.js applications. Stars: 2948, Forks: 292. Primary language: TypeScript. Languages: TypeScript (78.2%), Vue (20.3%), CSS (0.9%), HTML (0.3%), JavaScript (0.3%). License: MIT. Homepage: https://devtools.vuejs.org/ Latest release: v8.2.1 (3mo ago). Open PRs: 31, open issues: 112. Last activity: 1h ago. Community health: 37%. Top contributors: webfansplz, alexzhang1030, renovate[bot], skirtles-code, Azurewarth0920, LoTwT, JianJroh, xuzuodong, btea, yyx990803 and others.