| Commit message (Collapse) | Author | Age |
|
|
|
| |
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
| |
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
|
|
| |
Run integration tests in an electron environment for the main process.
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
|
|
|
|
| |
Since https://github.com/electron/electron/pull/33435 has landed in
electron 19 alpha, but not in 18, moving to 19 lets us remove the
workarounds for setBackgroundColor.
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
| |
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
|
|
|
|
| |
Embed the service-inject script into the service-preload script to avoid
having to load it manually and reduce IPC communication when a service
loads.
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
| |
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
|
|
|
|
|
| |
We need mui and mobx-react-like support before we can upgrade.
See https://github.com/mobxjs/mobx/issues/2526 for discussion about the
ramifications of concurrent rendering for mobx.
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Due to `BrowserView.setBackgroundColor` not working under linux, we have
to inject styles to make sure our `BrowserView` is opaque.
We try to cover more edge cases to avoid the interference of the
injected style and the web site's own styles according to the CSS
specification:
https://www.w3.org/TR/css-backgrounds-3/#body-background
In particular, we avoid overwriting the styles for the `html` element
if `body` already has an opaque background so that the background of
`body` can extend to the whole canvas.
This might still interfere with the web site if it updates the
background color on the fly (dark mode), but a reload should solve
most of such issues.
Hopefully, after
https://github.com/electron/electron/issues/32898
is resolved, we can get rid of the hack entirely.
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
| |
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
loglevel wants to persist the current log level in localStorage, which
apparently crashes the network service. We'll need to find another way
to log in service preload scripts. The main window preload scripts seems
unaffected, even though we use loglevel (and hence localStorage) in both
the preload and the renderer script.
Maybe related: https://github.com/electron/electron/issues/31689
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
| |
|
| |
|
|
|
|
| |
Signed-off-by: Kristóf Marussy <kristof@marussy.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Now we can run with ESM at build and test time and transpile into
commonjs for electron. This greatly simplifies testing, since we treat
everything as ESM during build with esbuild anyways. Now the test
environment and the build scripts match the apparent (but not the actual
for the main, preload, and inject modules) runtime environment.
Caveats:
- We may use top-level async expressions in tests and script, but not in
code that gets transpiled into commonjs or scripts that get imported
by vite. The limitation w.r.t. commonjs seems fundamental.
- Jest only experimentally supports ESM and there are some limitations
with mocking. Most limitations (except the lack of automatic mocks)
can be worked around by async importing code that uses mocks.
- There are packages marked as modules (so that node reads any scripts
in them as ESM) that nevertheless get transpiled into commonjs
modules. However, these should be clearly marked by using a .cjs
extension as their bundle. The worst offender is the root package,
which has a .cjs as its main entry point that gets read by electron,
but is in fact marked as a module. This doesn't seem to bother electron
at all. The service-inject package is an IIFE with a .js extension,
but it outputs a fully self-contained bundle, so the choice of module
format should be irrelevant.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
We will build all packages except the frontend (where vite remains in
use) with esbuild.
For some reason, the @yarnpkg/esbuild-plugin-pnp doesn't allow esbuild
to load esm modules and we fall back to commonjs for dependencies.
Hence we had to switch back to node_modules (but still rely on yarn
hardlinking for a more efficient install).
|
| |
|
| |
|
| |
|
| |
|
|
|