The current stable branch is determined based on the current version mapped to
the npm `latest` tag (by replacing the patch version number with 'x' - e.g.
`1.2.3` --> `1.2.x`).
PRs against the stable branch will be deployed to the preview server (as long as
the rest of the requirements are met). Commits on the stable branch itself will
be deployed to production.
Fixes#16908
Lighthouse v1.6.5 treats localhost/1.2.7.0.0.1 as secure domains (i.e. as if they where HTTPS), so we need to stop handling the is-on-https audit specially.
In some cases (unclear when), traceviewer-js, used by Lighthouse under the hood,
assumes `atob`/`btoa` are defined in the global scope. This is true for browser
environments, but not on node.
As a result, some aggregations that required access to model-tracing failed to
produce results, dropping the overall PWA score.
This affected #16665 (e.g. commit 0de6eec7a).
(Coincidentally), this wasn't an issue before fdfeaaf1f, because
pre-verification was run after `test.sh`, during which `aio-builds-setup` was
built.
Now that `deploy-staging.sh` is being run before `test.sh`, we need to build
the `aio-builds-setup` scripts first.
Previously, `aio/aio-builds-setup/scripts/travis-preverify-pr.sh` was supposed
to exit with 1 if a PR did not meet the preconditions and 2 if an error occurred
during pre-verification.
It relied on the exit codes of the node script that did the actual work, but
didn't account for errors that would be thrown in the `sh` script itself (e.g.
if the node script was not available). This caused such errors to appear as
non-verified PRs, instead of real errors that should fail the build.
This commit swaps the exit codes, so that now a 2 means non-verified PR and 1
designates an error.
Previously, no previews would be deployed for PRs that didn't touch files inside
`aio/`. Now, previews will be deployed for PRs that touch non-spec files inside
either `aio/` or `packages/` (as long as other preconditions are met).
Partially addresses #16526.
Currently, running `yarn test-pwa-score` right after deploying to staging fails
with the error:
```
{ Error: Unable to load the page: timeout reached ... code: 'PAGE_LOAD_ERROR' }
```
As a temporary fix, this commit prevents the build from failing because of PWA
score errors (until we identify the cause and fix it).
Previously, when trying to upload the build artifacts for a PR/SHA that was
already successfully deployed (e.g. when re-running a Travis job), the preview
server would return a 403 and the build would fail.
Since we have other mechanisms to verify that the PR author is trusted and the
artifacts do indeed come from the specified PR and since the new artifacts
should be the same with the already deployed ones (same SHA), there is no reason
to fail the build. The preview server will reject the request with a special
HTTP status code (409 - Conflict), which the `deploy-preview` script will
recognize and exit with 0.
This reverts commit d0bc83ca27.
Protractor-based prerendering is flakey on Travis and takes several minutes to
complete, slowing down the build. Prerendering has a lower impact now that we
use a ServiceWorker. We will revisit in the future (probably using a
`PlatformServer`-based approach).
PR Close#15346
Gaining access to another PR's JWT, would allow faking that PR's author wrt to
GitHub team membership verification for as long as the JWT is valid (currently
90 mins).
Igor Minar
Georgios Kalpakas
Peter Bacon Darwin
Jesus Rodriguez