On Windows, relative folders don't work so well. So we need to
lookup the full path. This is best done in DetermineBasePaths()
and as a bonus that only sets SP_WORKING_DIR once.
Also, while at it, make it more like the upcoming release-workflow,
so they look a lot more similar.
Functional it should be the same, except that Windows should
now also output when a test fails.
Basically, with '-c' you now create a sandbox. It will still use
your personal-dir and global-dir to find files you installed there,
but all new files are stored with a base folder identical to the
folder the configuration is in.
This is a bit of an old bug, that we many have tried to solve in
various of different ways. The code has grown sufficiently complex
that it is hard to see what consequences of actions are. This is
in my opinion the most harmless solution, while increasing the
usefulness of the '-c' flag.
In essence, the problem was that empty folders were always created
in the directory where the configuration was, but as that directory
wasn't added to any searchpath, files weren't stored there, unless
by accident it was a folder already on the searchpath. For example,
if you do './openttd -c local.cfg', it did work as expected. But
in the more generic variant, it did not.
With this patch, you can run './openttd -c /new/folder/local.cfg',
and it will create and prepare that folder to receive new files.
'content_download' is also stored in the directory the
configuration is in; this was already the case. Important to
note that there is only one search-path for 'content_download'.
In other words, when using '-c', it will not look in '~/.openttd'
inside the 'content_download' folder.
This was just weird. With XDG _personal_dir was created already,
but later on it was checked if it was different from config_dir,
and the creation was skipped. All this checking and validation
makes my head spin .. let's make it a bit more simple.
Formally it was only done on exit. This means that if it crashes
changes in settings were not stored. This is often rather
frustrating. Additionally, targets (like emscripten) where people
are unlike to use "Exit Game", will never see their configuration
stored.
The drawback is that on every setting change there is some minor
I/O of writing the ini file to disk again.
If I explicitly tell the system I do not want sound, I still get
presented a nice message I do not have any BaseSounds available
on my system, and that I should download one to enjoy sound. Well,
let me tell you, with "-snull" that is really really not going to
help. So please, be quiet, and let me enjoy the game without
"boooooo" and "DING DING DING".
Thank you.
The intention of this function was that you could download
everything after a filter was applied; but this never really
took off. Instead, a select few people used this functionality
to download every available package on BaNaNaS. This is not in
the spirit of this service. Additionally, these few people were
good for 70% of the consumed bandwidth of BaNaNaS.
LZO was used before the first version we track in our version
control system, which dates back to Aug 2004. Somewhere before
that time a few savegames / scenarios exist which use LZO. No
other savegame / scenario does since then. Let's not encourage
people to install something that ancient.
There are no scenarios on BaNaNaS that require LZO.
This is, by specs, undefined behaviour. See
https://reviews.llvm.org/D67122
In cases where this is done, optimizations done by LLVM can
generate code that causes crashes.
GetVariableAddress() had two (legit) ways this could happen:
- For SaveLoad set to global
- For SaveLoad set to SLE_VAR_NULL, where sld->address is always
a nullptr, and object could or could not be a nullptr.
By default gzip embeds a timestamps which makes building it twice
not reproducible, passing -n skips this embedding behaviour.
Motivation: https://reproducible-builds.org
