Fix child process killing when the parent process SIGINTs.
If you `ctrl + c` the Puppeteer parent process, we would sometimes not properly handle killing of the child processes. This would then leave child processes behind, with running Chromium instances. This in turn could block Puppeteer from launching again and results in
cryptic errors.
Instead of using the generic `process.kill` with the process id (which for some reason is negative the pid, which I don't get), we can kill the child process directly by calling `proc.kill`.
Fixes#5729.
Fixes#4796.
Fixes#4963.
Fixes#4333.
Fixes#1825.
This PR splits the logging for send and receive messages in to separate debug channels.
This way it is easier to filter and takes advantage of debug's automated coloring to make
it easier to visually parse on the command line
* chore: remove "Extracting..." log message
Fixes#5741.
* test: support extra Launcher options and skips
The extra Launcher options and skipping conditions enable
unit tests to be run more easily by third-parties, e.g.
browser vendors that are interested in Puppeteer support.
Extra Launcher options were previously removed as part of
switching away from the custom test harness.
* test: enable more tests for Firefox
* chore: move `index.js` into `src`
This is the first part of a series of pull requests that will slowly
make it possible to initialise Puppeteer for either a Node environment
or a web one. By creating the `initialisePuppeteer` function we can
inject dependencies in for the given version of Puppeteer (e.g. inject a
different debug library for Node vs the web).
This PR moves the initialisation into `src` and calls into it from the
root `index.js`.
This PR starts exploring the Page class and how to best document it. It explores how best to document events in the system, and I think pulling them out into an `enum` is the best solution here. It lets us end up with a page of docs that explicitly lists all the events the page class can ever emit.
* chore: ensure new-docs are up to date
This adds a Travis check that if we re-generate the docs we get the
exact output (by checking if the Git tree is dirty).
We do the dirty check by using `git status --porcelain`, seeing how many
lines that outputs, and using that as the exit code (taking only the
first 255 lines to avoid invalid exit codes). `--porcelain` makes the
output be empty if the repo is not dirty in anyway which translates into
an exit code of 0.
We can't use `git diff-index --quiet HEAD` as it exits with 0 if there are
untracked files in the repo but we want that to cause a failure.
Just one was used externally and I wrapped that up in a method. I think
it's a useful method to provide (I can imagine wanting to know if JS is
enabled on a page) so I think there's no harm here (I'd rather that then
have JSHandle reach into a private variable).
I've had misleading type errors due to left over builds or tests
behaving oddly. We should just strip the entire folder out before
building again so there's no left over artefacts that could cause
issues.
The `|| true` in the command is so if `rm` errors because the folder
doesn't exist, it doesn't exit with an error code.
This adds the new docs that are generated by API Extractor via TSDoc.
They are sparse because we do not yet use TSDoc in the codebase. This is
the starting point and from here we can work through all the different
classes and improve the documentation.
Replacing the Node EventEmitter with Mitt caused more problems than
anticipated for end users due to the API differences and the amount of
people who relied on the EventEmitter API. In hindsight this clearly
should have been explored more and then released as a breaking v4.
This commit rolls us back to the built in Node EventEmitter library
which we can release to get everyone back on stable builds. We can then
consider our approach to migrating to Mitt and when we do do that we can
release it as a breaking change and properly document the migration
strategy and approach.
We deferred this initially because our Windows CI built wasn't stable
and so debugging this was hard. It's now much more stable so let's push
this back a month but at the same time I'll reach out to the Moz folks
as it should be easier to debug reliably now CI is stable on Windows.
* chore: migrate to Mitt as the EventEmitter
This commit moves us to using Mitt [1] for the event emitter in
Puppeteer. This removes our dependency to Node's EventEmitter which is
part of a larger stream of work to enable a Puppeteer-web version that
doesn't depend on Node.
There are no large breaking changes as we support the main methods that
EventEmitter had, but it also provides some methods that Puppeteer
didn't use. Technically end users could depend on this but it's
unlikely.
[1]: https://github.com/developit/mitt
It conflicts with an inbuilt TypeScript `Request` type so can cause
confusion when in TS land. Note: `Response.ts` and `Worker.ts` also
suffer from this; PRs to rename them are incoming.
This script generated an `index.d.ts` file but that file was never
commited to git nor included in Puppeteer when we ship. As of right now
people who want TS types can install from the DefinitelyTyped repo and
we are working on shipping types from Puppeteer itself.
Therefore this script is not adding any value and can be removed.
* Don't use expect within Promises (#5466)
If a call to expect fails within a Promise it will not
be resolved, and causing the test to crash.
The patch aligns the code similar to what is used by all
the other tests.
We should import them just like any other module. This commit makes that
change. It does not change any behaviours or the types themselves.
EXPECTED_PROTOCOL_DIFF as we're updating the structure of it.
* chore: fix invalid SSL assertion on Catalina
The error Chrome gives with an invalid cert changes between older Mac
versions and Catalina as detailed here:
https://support.google.com/chrome/thread/18125056?hl=en.
This PR changes Travis to run Catalina (and we think most devs run up to
date OS versions) so this fix ensures the test behaviour is consistent
locally and on Travis.
For those on older Mac versions I've left a comment by the tests to
hopefully save them debugging!
Co-authored-by: Mathias Bynens <mathias@qiwi.be>
The coverage utils depend on `src/api.ts` being up to date and pointing to the right modules. If they aren't, you would get a cryptic error on CI:
```
1) "before all" hook in "{root}":
TypeError: Cannot read property 'prototype' of undefined
at traceAPICoverage (test/coverage-utils.js:40:54)
at Context.before (test/coverage-utils.js:103:7)
2) "after all" hook in "{root}":
TypeError: Cannot read property 'stop' of undefined
at Context.after (test/mocha-utils.js:168:22)
```
This change logs a clearer error that highlights the missing class and exits, so it's much easier to realise what's gone wrong.
Ideally the coverage wouldn't need a hardcoded list of sources, but until then this will help spot this error in the future.
This corresponds to Chromium 83.0.4103.0.
This roll includes:
- Enable SameSiteByDefaultCookies and CookiesWithoutSameSiteMustBeSecure https://crrev.com/c/2122809
When this test was failing, it would cause no future tests to run. This
was because the `expect` call within the `page.on` callback would throw
an error, and that would trigger a unhandled promise rejection that
caused the test framework to stop.
The fundamental issue here is making `expect` calls within callbacks.
They are brittle due to the fact that they throw, and the test framework
won't catch it, but also because you have no guarantee that they will
run. If the callback is never executed you dont' know about it.
Although it's slightly more code, using a stub is the way to do this.
Not only can we assert that the stub was called, we can make synchronous
`expect` calls that Mocha will pick up properly if they fail.
Before this change, running the tests (and making it fail on purpose)
would cause all test execution to stop:
```
> puppeteer@3.0.4-post unit /Users/jacktfranklin/src/puppeteer
> mocha --config mocha-config/puppeteer-unit-tests.js
.(node:69580) UnhandledPromiseRejectionWarning: Error: expect(received).toBe(expected) // Object.is equality
Expected: "yes."
Received: ""
at Page.<anonymous> (/Users/jacktfranklin/src/puppeteer/test/dialog.spec.js:42:37)
[snip]
(node:69580) UnhandledPromiseRejectionWarning: Unhandled promise rejection ... [snip]
```
But with this change, the rest of the tests run:
```
> puppeteer@3.0.4-post unit /Users/jacktfranklin/src/puppeteer
> mocha --config mocha-config/puppeteer-unit-tests.js
Page.Events.Dialog
✓ should fire
1) should allow accepting prompts
✓ should dismiss the prompt
2 passing (2s)
1 failing
1) Page.Events.Dialog
should allow accepting prompts:
Error: expect(received).toBe(expected) // Object.is equality
Expected: "yes."
Received: ""
at Context.<anonymous> (test/dialog.spec.js:53:35)
at processTicksAndRejections (internal/process/task_queues.js:94:5)
```
This is much better because one failing test now doesn't stop the rest
of the test suite.
This probably isn't the only instance of this in the codebase so I
propose as we encounter them we fix them usng this commit as the
template.
