microlinkhq / browserless / 36680925732

80%

Build Type

push

github

Committed by web-flow

Commit Message

fix(function): resolve the supplied page without a CDP round trip (#952)

`resolvePage` opened a CDP session on every candidate page and asked it
`Target.getTargetInfo`. On the `getPage` path `strictTarget` is always
set, so the scan always ran, and `browser.pages()` is every page on that
browser rather than only ours.

Measured in production on `({ page }) => page.title()`, correlating the
page count the isolate sees against the time the snippet itself takes:

  pages=1  fn.run 80-118ms   (12 samples)
  pages=2  fn.run 473ms
  pages=2  fn.run 1747ms

One neighbour is enough. The cost is not the length of the list, it is
opening a session on somebody else's page and waiting on a target that
may be mid-navigation. Worst observed was 15055ms of `fn.run` for a
snippet whose only statement is `page.title()`, with the fetch, spawn and
build phases all normal.

The isolate already has the id locally. `page.target()._targetId` is
maintained from target events, needs no round trip, and matches
`page.mainFrame()._id`; both were confirmed against a production isolate
on `@cloudflare/puppeteer` 1.4.0.

The scan now reads that first and only falls back to asking over a
session when no page carries it, so a build where the private field
disappears still resolves rather than failing with PAGE_NOT_FOUND.

The three existing tests describe pages with no local id, so they now
cover the fallback. Three more cover the local path, a neighbour that
refuses a session, and the mixed case.

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

Coverage Stats

1377 of 1619 branches covered (85.05%)

Branch coverage included in aggregate %.

23 of 23 new or added lines in 1 file covered. (100.0%)

8 existing lines in 1 file now uncovered.

5916 of 7432 relevant lines covered (79.6%)

67773.04 hits per line