When installing the module, instead of copying the metainfo file to the
destination, merge the available translations for it. Since this is done
at install time, add gettext as BuildRequires for the RPM packaging;
add libappstream-glib-devel as well only on RHEL/CentOS 8, since it
contains the metainfo.its for gettext in order to handle .metainfo.xml
files.
This requires a LINGUAS file with the list of translations currently
available: this is the way msgfmt picks the available languages to
merge. Generate it dynamically from the list of available .po files.
Recent versions of gettext know how to extract messages also from XML
files such as AppStream metadata. Hence, extract their messages, and
append them together with the global message catalog.
The generated RPM is noarch, so by definition has no binaries to extract
debug symbols from. Hence, drop the %define that makes RPM not error out
in case there are no debug symbols around (usually in archful binaries).
This ensures that we stay independent from any changes that go on in
online repositories, and our mock/pbuilders are sufficiently pre-cached.
Drop the explicit install of `cockpit-ws` -- all our bots images have it
pre-installed.
This started giving errors like:
22:10 error 'Alert' is defined but never used no-unused-vars
Use eslint-plugin-react to improve the JSX usage, not just to silence this error.
On developer systems, bots is commonly a symlink to a shared checkout,
accept non-directories as well. Also, only ignore it at the top level of
the project directory.
This centralizes/factorizes the rpm package builds, and builds
RPM packages in the VM instead of on the host, which is cleaner. It also
paves the way for supporting Debian and arch builds.
Drop the `make srpm` rule, as it's not very useful. Keep the `make rpm`
rule, as sometimes developers do this manually. This might be replaced
later on with another image-customize feature which copies the built rpm
out of the VM.
Use the CockpitRsyncWebpack plugin to optionally rsync build code to a
remote machine, the destination is determined by looking at the
package.json's name field.
Upgrading glibc in Rawhide now (semi-intentionally) removes
/etc/nsswitch.conf, but does not enforce installing a new enough
pam/authselect to ensure that it gets re-added. This breaks resolving
"localhost".
Reported as https://bugzilla.redhat.com/show_bug.cgi?id=2033020
These work without any source changes.
Don't update copy-webpack-plugin to the latest 10 major release, but
keep it at 9. 10 breaks compatibility with node.js 10, but that is still
the default in CentOS 8 stream (and we can't select the newer modules in
the spec or packit.yml).
npm 7 changed how it resolves dependencies, and starter-kit fails to
build with lots of unresolved peer dependencies of PatternFly.
With an absolute path, `resolve.modules` will only look in that
directory; the default is a relative path "node_modules" that just
works [1]. Use that default, as we don't use `$SRCDIR` in this project
anyway.
[1] https://webpack.js.org/configuration/resolve/#resolvemodules
Sync some improvements from cockpit-machines:
* Unbreak running `tmt` in local trees with an already existing ./bots
symlink.
* Don't install all NPM packages, only chrome-remote-interface and sizzle.
* Disable detection of affected tests, taking too long for real
projects.
`var` has broken/unexpected scoping in JavaScript, and can lead to
subtle errors. It's preferable and safer to use `const` whenever
possible, and `let` where necessary.
There are no usages of `var` in starter-kit right now, so just enable
the rule to ensure that it stays that way.
CentOS/RHEL 9 switched to OpenSSL 3, which does not support the `md4`
hash any more. webpack 5 hardcodes that in no less than 21 places, so
monkey-patch `crypto.createHash()` to substitute sha256 for md4.
Hack to work around https://github.com/webpack/webpack/issues/13572
