It was reported on Dec 22 2014 in irc that the game is broken if
this is missing, and a search shows that bigmaps and many campaign
story screens use jpg images.
It makes more sense imo to document this as an official dependency
now, and if we decide to change everything to png's then we can go
back and disable the jpg check then.
Conflicts:
SConstruct
Backported from master, commit bf75c9d062.
As I said on IRC, whether unit type variations use variation_id or
variation_name to define the variation id is wholly immaterial for
syncing purposes because serialized singular units:
1) Do not include [variation] definitions nowadays;
2) Have a single variation=<id> attribute, where the source for the id
of variations is determined internally during initialization.
One situation that could result in collateral OOS errors is
[store_unit_type]/wesnoth.unit_types retrieving different information
for 1.12.0 and 1.12.1 game clients running the same WML/Lua script. I
suspect this should be relatively rare and it's possible to guard
against synchronization issues by checking for both variation_id and
variation_name for WC/Soulless variations and picking one accordingly.
This fixes many instances of unit types with "technical" variations in
my campaign After the Storm being assigned help sections of their own.
Required adding a new method to unit_type that's actually cheaper than
checking for variations().empty(), as it doesn't involve allocating new
objects.
The password was sent under the client message's root node, rather than
the [create_game] subnode where the MP server expects to find the
password. This caused games created by affected clients to always lack a
join password.
This bug was presumably introduced by commit
13e22fd6f1 (1.11.6+dev).
The guard placement randomization is practically impossible to write so
that it'd guarantee a route through the map without making it too
trivial.
Now, there are 3 separate hardcoded guard layouts, one of which will be
chosen at random.
This is done to keep the difficulty of the scenario more consistent. We
now always unload one level 2 unit and two level 1 units (assuming that
at least 3 landing hexes are available).
If the lurker move is interrupted by an ambush, the attack action
previously sometimes caused an error, because the lurker might not get
to its target.
I don't know if this commit is related to or fixes any bugs, but
there are many bugs reported about fog / shroud / mp campaigns it
seems, and these lines always looked a bit squirrely to me. Since
boolean config values can be indicated by "yes", "no", "true" or
"false", testing for the presence of "yes" / "no" only seems like
it won't work in general, even if it works oftentimes. This commit
replaces it with a test for whether the to_bool function is
selecting the default boolean value.
