Someone seems to be trying to use up all our backtrace.io quota by
submitting the same bug over and over again which was fixed in
https://github.com/OpenRCT2/OpenRCT2/pull/8533, 17 days ago now.
To prevent such behaviour, I've disabled the old token and replaced it
with a new one, causing dumps from old builds to be rejected now and not
counting towards our quota.
This was a regression from #7435 which introduced threads and caused JNI
to misbehave and fail to load our expected classes. Provide a workaround
based on the description in https://stackoverflow.com/a/16302771 which
stores a main thread's class loader and uses that in neighbouring
threads.
Enable gzip, but only for local save, as GitHub will accept
gzip-compressed files even if they don't advertise it explicitly. Update
changelog and GitHub issue template to reflect that.
It should now cover all use cases supported previously, with added
benefit of automatically checking and prioritising installed directories
(including DESTDIR=.) as well as path relative to $PWD.
Note that this currently, this isn't implemented in the C++ standard
library. This means that date_order() will always return no_order,
causing this function to always return the default
DATE_FORMAT_DAY_MONTH_YEAR, as is the case before this commit too.
Once date_order() will be properly implemented in the standard library
someday, it will however start returning the right order for the current
locale, causing this function to start working as intended.
Previously when the arguments of a function wouldn't fit on a single line, clang-format would put each argument on its own line instead. By enabling the binpack parameter setting, it tried to fit as many on one line as possible instead.
Co-authored-by: Hielke Morsink <hielke.morsink@gmail.com>
