This means we have RTL support again with ICU 58+. It makes use of:
- ICU for bidi-itemization
- ICU for script-itemization
- OpenTTD for style-itemization
- harfbuzz for shaping
By default, GitHub adds all arguments of the matrix between ().
This is fine sometimes, but in other times it becomes a very
lengthy line.
With this commit, we decide what is between those (), making it
a lot more readable.
Lately we had a few times that people pushed to their PR branch
a few times to make small changes. Sadly, this triggers all CIs
every time, which takes ~20 minutes. As we are limited in the
amount of runners we get assigned to us, this means all other CI,
even for other repositories within OpenTTD, are delayed too.
We can avoid this by simply cancelling old runs when a new PR is
pushed. There is a downside: sometimes people already push a new
commit, but still want to know if the old one passed. That will
no longer be possible with this change.
This requires the use of WinHTTP (for Windows) or libcurl (for all
others except Emscripten). Emscripten does not support http(s)
calls currently.
On Linux it requires ca-certificates to be installed, so the HTTPS
certificate can be validated. It is really likely this is installed
on any modern machine, as most connections these days are HTTPS.
(On MacOS and Windows the certificate store is filled by default)
Reminder: in case the http(s):// connection cannot be established,
OpenTTD falls back to a custom TCP-based connection to fetch the
content from the content-service. Emscripten will always do this.
Currently they had a name that the rest of our system cannot
deal with correctly. "cert.pfx" is also not very descriptive from
a system as a whole.
As such, we now name it like any other file, so it can be published
safely to the CDN.
It turns out that having "-g" in the compile-statement causes
Emscripten to pick -g3, which makes for very big binaries. This
is very likely not your intention when building Emscripten, as
smaller really is better.
For comparison, with RelWithDebInfo the binary is ~80MB. With
Release it is ~7.4MB.
With std::variant all memory can be figured out at compile time, so the compiler needs to keep track of fewer elements. It also saves out a unique_ptr and its memory management, over a slight impact for resolving a setting.
