…port)

Cherry-picked from 9fed286 (3.0.0-dev):

> Core: Exclude or grey internal frames from stack traces in TAP reporter
> Internal frames are those from qunit.js, or Node.js runtime.
>
> * Remove any internal frames from the top of the stack.
> * Grey out later internal frames anywhere in the stack.
>
> This change is applied to the TAP reporter, which the QUnit CLI uses
> by default.
>
> Ref #1789.

Cherry-picked from 95105aa (3.0.0-dev):

> Fix fragile code in stracktrace.js that previously worked only because
> of Babel transformations masking a violation of the Temporal Dead Zone
> between `const fileName` and the functions it uses to compute that
> value.

…x backport)

Cherry-picked from 276e09d (3.0.0-dev).
> Ref #1795.

These have been included by watch mode since QUnit 2.18, but so far
only benefited you there if you manually passed these as individual
files, or loaded them indirectly at runtime, or used your own glob
that includes mjs and/or cjs.

If you currently run qunit by passing a directory and have matching
files that you want to skip, opt-out by passing `test/*.js` or
`test/**/*.js` explicitly instead of `test/`.

Follows-up #1676.

Cherry-picked from 6f72eeb (3.0.0-dev).

Cherry-picked from 53ee0a7 (3.0.0-dev).

Cherry-picked from 05e15ba (3.0.0-dev).

Ref #1711.

`#qunit-tests li strong` also applied to `#qunit-tests li .qunit-source strong`

Cherry-picked from 52bfa69 (3.0.0-dev).

Cherry-picked from 58d93b9 (3.0.0-dev).

Optimize the hidepassed click handler for large test suites by
avoiding internal function call overhead for the iterator, as well
as copying cost for the array. Shaves off ~100ms milliseconds on the
q4000 demo 1.65s to 1.5s (Chrome, CPU throttle 6x), and seems to defer
some unattributed costs to outside the critical path.

Cherry-picked from a729421 (3.0.0-dev).

While this is not in any way a documented API, it is quite common for
QUnit runners that are based on Selenium/Webdriver to use HTML scraping
to obtain a simple summary of whether the run passed.

Intruce memory for `runEnd` event so that anything JS-based has a
stable and documented way to obtain the result in a machine-readable
format, without needing to resort to HTML scraping.

One common reason sometimes avoid the event emitter is that, depending
on various circumstances, a custom test runner might end up injecting
code slightly too late, at which point JS code to listen for our event
might not fire, whereas the DOM is still available. This is rare,
but can happen:

* if the file is served as-is.

  browserstack-runner and QTap solves this by proxying the HTML file
  and inject an inline script to reliable listen early.

  Karma solves this by proxing the test framework instead (qunit.js)
  and adapting it to include relevant event listeners upfront.

* and, if the browser is driven without early JS control.
  When WebDriver v1 is used, something like webdriver "execute" is
  not guruanteed to run before DOM-ready or window.onload. If the
  Node.js process is slow or far away from the browser (e.g. cloud),
  and if the test suite is relatively small/fast, then the injected
  code might arrive after the tests have already finished.

  grunt-contrib-qunit avoids this by using Puppeteer and
  its `Page.evaluateOnNewDocument()` method, to reliably run a script
  before any others.

* and, if the test is relatively small/fast.

Example at mauriciolauffer/wdio-qunit-service#13

Solve this once and for all by adding memory to the `runEnd` event.
This allows late event listers to handle the event retroactively.

Cherry-picked from 8f25f26 (3.0.0-dev).

Cherry-picked from adc4493 (3.0.0-dev).

…ile)

Cherry-picked from 62ca15a (3.0.0-dev).
> Core: Convert "No tests were run." from fake test to error

Early errors produced output like so:

```
not ok 1 global failure
  ---
  message: ReferenceError: boom is not defined
  stack: |
    ReferenceError: boom is not defined
        at /example.test.js
  ...
Bail out! ReferenceError: boom is not defined
TAP version 13
1..1
```

Instead of:

```
TAP version 13
not ok 1 global failure
  ---
  message: ReferenceError: boom is not defined
  stack: |
    ReferenceError: boom is not defined
        at /example.test.js
  ...
Bail out! ReferenceError: boom is not defined
1..1
```

Because prior to the introduction of the "error" event in 2.17.0,
the only failures were test failures, and those would not begin until
after QUnit.begin() / "runStart" event.

Cherry-picked from 15acb36 (3.0.0-dev).
> HTML Reporter: Add support for displaying early errors
> Ref #1786.

…rage

Cherry-picked from 497748b (3.0.0-dev).

Cherry-picked from abeac18 (3.0.0-dev).

> Tests should pass even when running in a subshell or otherwise
> without TTY or FORCE_COLOR=1.

Cherry-picked from f0af617 (3.0.0-dev).

> When an assertion message contained a colon followed by a space,
> and otherwise is a single line of ASCII characters, we formatted it
> as an unquoted string. This is invalid in YAML 1.2, and thus tap-parser
> would ignore the diagnostic block.
>
> This does not affect the TAP result itself (the test name and failure
> status are separate from this), and it does not affect the QUnit CLI
> output where the information is shown either way.
>
> When passing the CLI output to tap-parser (such as in QTap), it
> silently ignored the "diag" block containing the assertion
> message/actual/expected value, because it is not valid YAML.
>
> Note that YAML 1.2 does allow literal colons and hash tags in unquoted
> strings, when they are not followed by or preceded by a space, but
> I'm keeping our version simpler by erring toward quoted strings when
> these appear. Just because we could squeeze one more edge case with
> plain unquoted strings, doesn't mean we have to. This way keeps our
> format a bit more intuitive by making it predictable and easy to
> deduce why a string is quoted or not, based solely on the chars used
> and not based on obscure YAML internals.
>
> Ref https://yaml.org/spec/1.2.2/#733-plain-style