This PR allows us to code playwright integration tests that can potentially replace some of our manual tests and allow us automated testing of the app itself.
Current technical limitations:
* No app level keyboard simulation support (e.g, zoom in, zoom out, etc.)
* No code coverage support, so even though we are testing the app, the code coverage does not reflect this fact
Previous attempt failed by design of `npm pack` / `npm publish`,
as documented at https://docs.npmjs.com/cli/v6/configuring-npm/package-lock-json :
> One key detail about package-lock.json is that it cannot be published,
> and it will be ignored if found in any place other than the toplevel
> package. It shares a format with npm-shrinkwrap.json, which is
> essentially the same file, but allows publication.
>
> This is not recommended unless deploying a CLI tool or otherwise using
> the publication process for producing production packages.
, and we are a CLI tool. Switching to shrinkwrap.
They were used a long time ago, then I scrapped them for simplicity to
new contributors. I'm re-considering this and re-introducing one, for
two (maybe three) reasons:
1. Reading on supply chain attacks
2. Build broken because of a dep change (see previous commit broken
because of a change in yargs @ 17.1.0)
(3.) Performance
* Catch promise errors better
* Move subFunctions to bottom of createNewWindow
* Use parents when creating child BrowserWindow instances
* Some about:blank pages have an anchor (for some reason)
* Inject browserWindowOptions better
* Interim refactor to MainWindow object
* Split up the window functions/helpers/events some
* Further separate out window functions + tests
* Add a mock for unit testing functions that access electron
* Add unit tests for onWillPreventUnload
* Improve windowEvents tests
* Add the first test for windowHelpers
* Move WebRequest event handling to node
* insertCSS completely under test
* clearAppData completely under test
* Fix contextMenu require bug
* More tests + fixes
* Fix + add to createNewTab tests
* Convert createMainWindow back to func + work out gremlins
* Move setupWindow away from main since its shared
* Make sure contextMenu is handling promises
* Fix issues with fullscreen not working + menu refactor
* Run jest against app/dist so that we can hit app unit tests as well
* Requested PR changes
* Add strict typing; tests currently failing
* Fix failing unit tests
* Add some more eslint warnings and fixes
* More eslint fixes
* Strict typing on (still issues with the lib dir)
* Fix the package.json import/require
* Fix some funky test errors
* Warn -> Error for eslint rules
* @ts-ignore -> @ts-expect-error
* Add back the ext code I removed
Domain-ish URL test now also covers considering internal URLs with different sub-domains like
- `https://listen.tidal.com`
- `https://login.tidal.com`
Co-authored-by: Daniel Fuchs <dfuchs@multamedio.de>
Co-authored-by: Adam Weeden <adamweeden@gmail.com>
Co-authored-by: Ronan Jouchet <ronan@jouchet.fr>
As documented in #1153, for Widevine support to work properly, we need to listen for the Widevine ready event, and as well for certain sites the app must be signed.
This PR adds the events, and as well adds better documentation on the signing limitation.
This may also help resolve #1147
As documented in https://github.com/jiahaog/nativefier/issues/923#issuecomment-599300317 ,
- #923 is caused by installing placeholder app deps at nativefier
*install* time, with yarn (8.0.2) or npm (8.0.3). This is new in
Nativefier 8.x, for the motivations behind it, see
https://github.com/jiahaog/nativefier/pull/898#issuecomment-583865045
- During testing, I did test global installs, but never to a
system / non-user-writable path (my `$npm_config_prefix` is set to
`"$HOME/.node_modules"`)
- But without such a config and when installing globally to a
non-user-writable/system path with `sudo npm i -g nativefier`,
- Installation of nativefier core works...
- ... but then `postinstall` tries to do its job of installing
app deps, and fails in various OS-dependent ways, but all about
access rights.
I suspect that, although main nativefier install runs as `su` with
access rights to system paths, `postinstall` scripts are run *out*
of `su`.
That would make sense for security reasons: out of hook scripts,
npm knows exactly what will be touched in your filesystem: it's the
static contents of the published tarball; a postinstall script with
sudo rights could do nasty dynamic stuff. So, although I don't see
any mention of that in
[npm-scripts docs / hooks](https://docs.npmjs.com/misc/scripts#hook-scripts)
and I haven't dug npm/cli's code, I can understand it.
So, reverting back to `webpack`ing the placeholder app, as done pre-8.0.
## Breaking changes
- Require **Node >= 8.10.0 and npm 5.6.0**
- Move to **Electron 8.1.1**.
- That's it. Lots of care went into breaking CLI & programmatic behavior
as little as possible. **Please report regressions**.
- Known issue: build may fail behind a proxy. Get in touch if you use one:
https://github.com/jiahaog/nativefier/issues/907#issuecomment-596144768
## Changes summary
Nativefier didn't get much love recently, to the point that it's
becoming hard to run on recent Node, due to old dependencies.
Also, some past practices now seem weird, as better expressible
by modern JS/TS, discouraging contributions including mine.
Addressing this, and one thing leading to another, came a
bigger-than-expected revamp, aiming at making Nativefier more
**lean, stable, future-proof, user-friendly and dev-friendly**,
while **not changing the CLI/programmatic interfaces**. Highlights:
- **Require Node>=8**, as imposed by many of our dependencies. Node 8
is twice LTS, and easily available even in conservative Linux distros.
No reason not to demand it.
- **Default to Electron 8**.
- **Bump** all dependencies to latest version, including electron-packager.
- **Move to TS**. TS is great. As of today, I see no reason not to use it,
and fight interface bugs at runtime rather than at compile time.
With that, get rid of everything Babel/Webpack.
- **Move away from Gulp**. Gulp's selling point is perf via streaming,
but for small builds like Nativefier, npm tasks are plenty good
and less dependency bloat. Gulp was the driver for this PR: broken
on Node 12, and I didn't feel like just upgrading and keeping it.
- Add tons of **verbose logs** everywhere it makes sense, to have a
fine & clear trace of the program flow. This will be helpful to
debug user-reported issues, and already helped me fix a few bugs.
- With better simple logging, get rid of the quirky and buggy
progress bar based on package `progress`. Nice logging (minimal
by default, the verbose logging mentioned above is only used
when passing `--verbose`) is better and one less dependency.
- **Dump `async` package**, a relic from old callback-hell early Node.
Also dump a few other micro-packages unnecessary now.
- A first pass of code **cleanup** thanks to modern JS/TS features:
fixes, simplifications, jsdoc type annotations to types, etc.
- **Remove GitHub integrations Hound & CodeClimate**, which are more
exotic than good'ol'linters, and whose signal-to-noise ratio is too low.
- Quality: **Add tests** and add **Windows + macOS CI builds**.
Also, add a **manual test script**, helping to quickly verify the
hard-to-programatically-test stuff before releases, and limit regressions.
- **Fix a very small number of existing bugs**. The goal of this PR was
*not* to fix bugs, but to get Nativefier in better shape to do so.
Bugfixes will come later. Still, these got addressed:
- Add common `Alt`+`Left`/`Right` for previous/next navigation.
- Improve #379: fix zoom with `Ctrl` + numpad `+`/`-`
- Fix pinch-to-zoom (see https://github.com/jiahaog/nativefier/issues/379#issuecomment-598612128 )
* Update deps except eslint
* Update eslint and lint:fix (WIP, needs manual fixing for remaining 44 problems)
* Manually fix remaining eslint errors
* Document deprecation of `version-string` as of electron-packager 9.0.0
* Upgrade to Electron 1.7.9 (chrome-58, node-7.9.0, v8-5.8)
* npm: Disable generation of package-lock.json and gitignore it
--Trying this, package-lock is a pain in PRs. May not be a good idea
(obviously we lose deps pinning), will revert if necessary.--
* npm tasks: add dev-up-win for Windows developers,
and e2e for end-to-end tests. Update docs.
* Move normalizeUrl test to a jest unit test, makes no sense to be in the mocha e2e tests
* Switch from babel-preset-es2015 to babel-preset-env,
with target.node=4.0. Seem like it's today's most convenient
way to support the latest ES and let babel transpile to what
makes sense for our currently minimal node version
This implements the following cases:
* Basic runs for a specific version/platform/arch with default options
(testing for existence of expected output files/folders)
* Runs for each target platform with `out` set
* Runs for each target platform with `asar` set
* Runs for each target platform with `prune` set
* Runs for each target platform with `ignore` set
* This includes testing `ignore` including a path separator
* This also includes testing that `ignore` entries are applied
with the app folder trimmed
* Runs for each target platform with `overwrite` set / not set
* Runs with `all` set, with `arch` or `platform` each set to `all`,
and with `platform` set to multiple platforms (comma-delimited or
array)
* Tests specifically for the darwin target platform, to test
`icon` and `sign`
The test logic is organized such that the version of Electron being
tested can be configured once centrally (in config.json), and the tests
will not run until after downloading any necessary electron versions
(to avoid timing out due to long downloads).
Resolves #77.
This adds support for --all, --platform=all, and --arch=all.
In order to accommodate outputting multiple directories for multiple
platforms and architectures, this also implements a new directory
structure under the output folder (distinguished by both platform and
arch). This structure is applied even to OS X distributions, which
formerly were output directly to an .app folder. This could be considered
a backwards-incompatible change.
One other backwards-incompatible change is the value that the packager
function passes to the callback, which is now always an array of paths,
rather than just a single path.
The behavior of the icon option has also been modified to use its
basename and apply .ico or .icns depending on platform, to make it
usable with --all and --platform=all. This attempts to maximize
backwards compatibility, by allowing a full filename to be specified,
but replacing the filename's extension with what is appropriate for
each target platform. Alternatively, the extension can now be omitted.
In the process of implementing this, it became evident that some things
were being done in 3 different places, and weren't always being done
consistently, so I've deduplicated everything I could.
This also includes a few other changes to improve stability for
multi-target runs, and other fixes:
* Avoid targeting darwin if the environment doesn't support symlinks,
to avoid the process bailing out on Windows
* Implement --overwrite centrally in index.js such that it explicitly
skips if an output directory already exists, for consistency with
all target platforms and to avoid any possible errors that would halt
operation during one target of a multi-target run
* Use ncp instead of mv to move to finalPath, which avoids flakiness
I noticed when testing on Windows 8 especially with multi-target runs
* Simplify temp directory logic by using a nested structure, so there
is only one top-level directory to clean up
* Reinstate fix from #55 which seems to have been clobbered by a later
merge
* linux.createApp now resolves to the final output directory;
it was formerly resolving to the executable path
* mac.createApp now replaces space with underscore in bundle IDs
* Only the platform modules that are needed are loaded
* The win32 module only loads rcedit if needed
This also fixes a couple of missing updates to docs (readme/usage).
This commit addresses the following issues:
* Resolves #40
* Resolves #38
* Resolves #70
* Works around #71
* Resolves #84 by reinstating #55
