[Copilot]

Motivation and measured win

Core Heft startup validates configuration and plugin metadata with AJV. Precompiling built-in validators moves that work from each invocation into the build. A paired comparison of actual builds of this PR and its main base across all 182 runnable downstream Heft consumers in the repository found faster boot on every measured project. The median per-project cold-start saving was 32.5 ms (15.6%), with a range of 18.7–61.3 ms. This supersedes the earlier single-project synthetic estimate.

Changes

• Generation: Add a built-in Heft task plugin that emits standalone CommonJS and ESM validators from *.schema.json files. Generated modules resolve AJV runtime helpers through the deep @rushstack/node-core-library/lib/JsonSchemaRuntime import, so consuming projects do not need to resolve AJV themselves.
• Runtime: Add jsonSchemaValidator to heft-config-file. Core Heft config, manifest, and built-in plugin option validation use generated validators. Other schema-publishing projects are not yet migrated.

new ProjectConfigurationFile({
  projectRelativeFilePath: 'config/my-plugin.json',
  jsonSchemaValidator: require('./schemas/my-plugin.validator.cjs')
});

Downstream startup profiling

Revisions: main base f11aaa2 vs PR 724c10d, each separately installed and built from source in its own directory on the same Linux host (Node 24.22.1). For each project declaring @rushstack/heft as a dependency (excluding Heft itself), launch that revision's self-built apps/heft/lib-commonjs/start.js from the project's folder with run --only _profiling_nonexistent_. This deliberately initializes real project configuration and plugin schemas, then stops at the same expected missing-phase error, before any build tasks execute. Both revisions produced that expected result for every measured project. Run one cold-process warmup per revision/project, then five paired cold-process measurements, alternating revision order; report the median of those five runs for each project. Wall time includes process launch; CPU and peak RSS are child-process resource usage. No synthetic validator hook or historical-code approximation was used.

Metric	Main base (median project)	PR (median project)	Median paired saving
Startup wall time	207.7 ms	175.6 ms	32.5 ms (15.6%)
Process CPU time	281.9 ms	231.4 ms	51.5 ms (18.2%)
Peak RSS	88.0 MiB	81.7 MiB	6.3 MiB (7.1%)

Project coverage and median paired wall-time savings by repository area:

Project area	Projects	Median saved	Faster on PR
apps	13	34.0 ms	13
build-tests	61	32.3 ms	61
build-tests-samples	14	30.7 ms	14
build-tests-subspace	4	33.3 ms	4
eslint	7	29.9 ms	7
heft-plugins	17	36.9 ms	17
libraries	34	31.8 ms	34
repo-scripts	2	29.9 ms	2
rush-plugins	11	33.2 ms	11
vscode-extensions	5	34.1 ms	5
webpack	14	29.4 ms	14
Total	182	32.5 ms	182

Of the 188 repository projects declaring a Heft dependency besides Heft itself, six are rig packages, not standalone Heft build projects: @rushstack/heft-node-rig, @rushstack/heft-vscode-extension-rig, @rushstack/heft-web-rig, decoupled-local-node-rig, local-node-rig, and local-web-rig. They have no root config/heft.json; starting Heft from their package folders fails on both revisions, so they are excluded rather than counted as successful boots. Their consuming projects are included. These results measure initialization through command selection, not build-task execution or performance of external Heft consumers.

[David Michon (dmichon-msft)]

Copilot Fix the comment

[David Michon (dmichon-msft)]

Always use withFileTypes: true for readdir calls

[Copilot]

Copilot Fix the comment

Fixed in 724c10d: the schema test now calls readdirSync with withFileTypes: true and filters file entries. The targeted test could not start because the configured package-feed hostname returned ENOTFOUND.