ai

Importing constructs validation schemas even when their model or tool is never used. Defer the remaining eligible schema construction with the existing / utilities and deferred Zod helpers. Validators are cached on first use; inferred public option and response types are preserved. Most schema-body changes are indentation. Follow-up to #23636. This PR only changes and is independent of the other provider follow-ups. Load-time measurements Median improvement: 0.25 ms (1.1%). The median paired saving is 0.18 ms, with a 95% bootstrap interval of 0.09–0.26 ms (10,000 resamples); the candidate was faster in 125/200 pairs. Measured locally on Node v24.18.0, Apple M4 Pro, darwin/arm64. Both revisions were built with the repository's tsdown config. Alternated before/after order across 200 fresh Node-process pairs after five discarded pairs, timing only with . Filesystem caches were warm and Node's compile cache was disabled. Baseline: , which includes #23636. All provider dependencies were held at the baseline revision for both builds, so this result isolates this provider's own changes. This measures package import latency. Schema construction moves to first use, so the figures do not claim an equivalent first-request or end-to-end latency improvement. No live provider requests were made. Reproduce the import measurement Build the baseline and this PR, preserve separate copies of their directories with resolvable external dependencies, and hold the other dependency builds at the baseline. Save the following as and run . The existing can also measure each installed revision separately. Validation Full provider suites: 18 Node tests and 18 Edge tests passed. Added an import-initialization regression; it fails against the original provider source. Package and declaration builds passed. passed with the provider candidates together, including examples. and passed. Patch changeset included.

Importing constructs validation schemas even when their model or tool is never used. Defer the remaining eligible schema construction with the existing / utilities and deferred Zod helpers. Validators are cached on first use; inferred public option and response types are preserved. Most schema-body changes are indentation. Follow-up to #23636. This PR only changes and is independent of the other provider follow-ups. Load-time measurements Median improvement: 7.24 ms (26.2%). The median paired saving is 6.82 ms, with a 95% bootstrap interval of 6.46–7.16 ms (10,000 resamples); the candidate was faster in 50/50 pairs. Measured locally on Node v24.18.0, Apple M4 Pro, darwin/arm64. Both revisions were built with the repository's tsdown config. Alternated before/after order across 50 fresh Node-process pairs after five discarded pairs, timing only with . Filesystem caches were warm and Node's compile cache was disabled. Baseline: , which includes #23636. All provider dependencies were held at the baseline revision for both builds, so this result isolates this provider's own changes. This measures package import latency. Schema construction moves to first use, so the figures do not claim an equivalent first-request or end-to-end latency improvement. No live provider requests were made. Reproduce the import measurement Build the baseline and this PR, preserve separate copies of their directories with resolvable external dependencies, and hold the other dependency builds at the baseline. Save the following as and run . The existing can also measure each installed revision separately. Validation Full provider suites: 407 Node tests and 407 Edge tests passed. Added an import-initialization regression; it fails against the original provider source. Package and declaration builds passed. passed with the provider candidates together, including examples. and passed. Patch changeset included.

Importing constructs validation schemas even when their model or tool is never used. Defer the remaining eligible schema construction with the existing / utilities and deferred Zod helpers. Validators are cached on first use; inferred public option and response types are preserved. Most schema-body changes are indentation. Follow-up to #23636. This PR only changes and is independent of the other provider follow-ups. Load-time measurements Median improvement: 1.45 ms (7.2%). The median paired saving is 1.62 ms, with a 95% bootstrap interval of 1.35–1.82 ms (10,000 resamples); the candidate was faster in 48/50 pairs. Measured locally on Node v24.18.0, Apple M4 Pro, darwin/arm64. Both revisions were built with the repository's tsdown config. Alternated before/after order across 50 fresh Node-process pairs after five discarded pairs, timing only with . Filesystem caches were warm and Node's compile cache was disabled. Baseline: , which includes #23636. All provider dependencies were held at the baseline revision for both builds, so this result isolates this provider's own changes. This measures package import latency. Schema construction moves to first use, so the figures do not claim an equivalent first-request or end-to-end latency improvement. No live provider requests were made. Reproduce the import measurement Build the baseline and this PR, preserve separate copies of their directories with resolvable external dependencies, and hold the other dependency builds at the baseline. Save the following as and run . The existing can also measure each installed revision separately. Validation Full provider suites: 26 Node tests and 26 Edge tests passed. Added an import-initialization regression; it fails against the original provider source. Package and declaration builds passed. passed with the provider candidates together, including examples. and passed. Patch changeset included.