Add a saved-page corpus and a Chromium test harness - #6
Open
EnesYilmazcode wants to merge 15 commits into
Open
EnesYilmazcode wants to merge 15 commits into
EnesYilmazcode wants to merge 15 commits into
Conversation
This was referenced Sep 24, 2026
Open
Strips scripts, frames and inline handlers, and scrubs session and tracking tokens, so captured pages can live in the repo as fixtures.
Real captures from the audit (a yoga mat search page, the Akamai interstitial, an aodAjaxMain fragment and a product page), hand-written pages for a captcha, title-recipe cards and a missing pagination strip, and copies of the old fixtures. Each page has an expected.json written from the page, not from the current code.
A characterization test, bugs included, so extracting the parsers can prove it changed nothing. Real pages carry CSS jsdom cannot parse, so the content-script loader now keeps that error out of the test output.
The content scripts, the popup and the tests now share one pure module (global Parsers, CommonJS in Node). scraper.js and offer-fetcher.js keep their function names as thin delegates, and Analyzer's coercions live there too. No behavior change: the corpus snapshot is unchanged.
Adds a parseDoc helper that builds a jsdom Document with the same innerText fallback the content-script loader uses.
Each check a page's expected.json makes (page kind, ASIN dedupe, sponsored count, ad links, next page, total, fill rate, spot products, seller prices) runs per page. The ones today's code fails run as test.failing with their finding id, so the fix that makes one pass has to take it off the list.
At most 5 amazon.com fetches, 3 seconds apart. Each page is sanitized and compared with the previous capture before it is written, and only after a y. It refuses to run in CI.
global-setup builds the prod extension, lib/extension.mjs loads it into Chromium with a fresh profile and drives the real popup, and lib/amazon.mjs answers every request from generated pages or the corpus, so no request reaches Amazon. One Chromium at a time.
A clean run, the saved yoga mat page against the parser, a captcha, duplicates, Stop, a second search tab, a product tab, a slow page, a full storage quota, two runs without an export and a stopped service worker. Scenarios that fail on today's code call bug() with the finding id right before the failing checks, so they run as expected failures.
Target.closeTarget returns before the worker is gone, which made the service worker scenario flaky.
The harness checks out commit 7c2ba1c (the store build) with a temporary index, loads it unpacked, scrapes, then swaps in the current build and reloads, like an auto-update. Both scenarios are known failures: nothing migrates v2.0 data (F-100), and a run in flight is resumed with no run id (F-101).
A separate job installs the full Chromium build for the pinned Playwright version and runs it in the new headless mode, which loads extensions. Full history is fetched for the upgrade scenarios.
A test.failing passes on any error, so a crash in the offer parser left the aod-pinned-only page green. Each page now also checks that the parsers run without throwing and return the right shape. The header count check was tagged F-25, which is the dashboard schema finding; it is now NEW-PARSE-1, since no audit finding covers it.
EnesYilmazcode
force-pushed
the
rebuild/p1-tests
branch
from
September 24, 2026 21:05
a6aa58e to
f035b71
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stack, merge in order: #5 > #6 (this one) > #7 > #8 > #9. Based on #5, so the diff here is only this step.
The proscan-web stack: EnesYilmazcode/proscan-web#1, EnesYilmazcode/proscan-web#2.
Before changing the scraper I wanted tests that run against real Amazon markup instead of hand-written snippets, which is how F-27 and F-16 went unnoticed.
Bugs the corpus found are marked as expected failures with their F-id in the name (F-11, F-12, F-13, F-15, F-16, F-17, F-20, F-26, F-27, F-30, F-37, F-100, F-101). The header result count ("of over 10,000 results" read as 0) has no audit finding, so it is tagged NEW-PARSE-1. Later PRs in the stack flip them to passing. A test.failing passes on any error, so every corpus page also has plain checks that must pass today: the parsers run without throwing and return the right shape. A crash in the offer parser now fails the corpus instead of hiding behind a known failure. Each Playwright scenario checks its setup before the buggy assertion, so a broken harness still fails loudly. (F-81, F-83)
Tested: npm test is 377 jest (19 of them test.failing) and 35 tool tests. npx playwright test is 13/13, 10 of them expected failures. The suite runs on Windows locally and on Linux (ubuntu-latest) in CI.
Does not change permissions. The manifest diff only adds parsers.js to the content script list: