Another efficiency improvement. This ensures that the side of the unit
is checked first before the potentially complex rest of the filter is
evaluated.
(cherry-picked from commit 6c38b8ce934ea68bd11a3aa960a91fe6f9dbfb87)
This is for efficiency reasons, as location filters do not necessarily
change throughout a side’s turn, while fewer and fewer units have moves
left.
(cherry-picked from commit 0dcbe1d0d0b1dda90a0fa649120348b7542f2644)
The old method is very slightly faster in some circumstances (e.g. for
simple filters and when all units have moves/attacks/etc. left), but we
are talking fractions of micro seconds. By contrast, depending on the
filter used, the new method can save large amounts of evaluation time
once some units have move or attacked.
(cherry-picked from commit 972ecc2f56c3cc1391db165518324b1a3757431d)
- Fix the rabbit AI
- Enable invoke_synced_command to also call (some) built-in commands
and give an error message in the case of an unknown command
- Remove some unnecessary implementation details
fixes#1649 . ai.synced_command could easily be used to implement all types of
undeteced cheats so it was removed. As a replacement this commit adds a
[custom_command] synced command that just calls
wesnoth.game_events.on_synced_command which calls a lua handler that
must first be set.
This removes the Lua deprecation_message function in favour of exposing the C++ variant to Lua instead.
It also moves all deprecation messages to a separate logdomain, making them easily enabled en masse.
This reverts commit 1d84d83c6c6c70e3c25de6e3d443a7ec17936711.
I had introduced use of the hidden= key in b2df837d8e7a95ae2e15dcbdffbd0653ab481443, but
apparently didn't add them to the valid list of non-aspect keys.
The previous version caused errors under some circumstances when an
event removed units other than the attacker during or immediately after
an AI attack.
When the AI moves a unit out of the way for another unit’s move, it is
occasionally possible for the former unit to use up all its movement
points (e.g. if it has few MP or on narrow crowded maps). If that unit
is on the list to be moved later during the same sequence, this caused
ai_helper.robust_move_and_attack() to return nil, which then made the
Fast Micro AI produce an error message.
It was left there in cfb2aa8050 for backward compatibility. However,
this causes a “duplicate AI” error at start-up in debug mode.
Furthermore, the file is not included in any current 1.12 or 1.13
add-on, so it should be safe to remove it. Using
[ai]ai_algorithm=idle_ai is the preferred option to include the Idle AI.
As indicated in the comment of the idle_ai config file, an AI needs to
contain at least one stage, otherwise the default AI will be injected
under certain circumstances. Unlike indicated in the comment, this now
also applies to loading a mid-scenario save. (I will add that to the
comment in the next commit.)
As this AI was in data/ai/dev/, it was previously only loaded when
Wesnoth was started in debug mode. The config was thus not available
when playing in no-debug mode, causing the default AI to be injected on
reload. This caused issue #1955.
While not strictly necessary, we keep a redirect in data/ai/dev for
backward compatibility, in case the file is included in add-ons.
The WFL min() function was changed from returning zero to returning an
error when an empty list is passed as argument (commits eccd3d8e and
fd9f1153). This adapts the Poisoner FAI code to deal with that
correctly.
(as proposed by @DeFender1031)
All existing deprecation messages in the Lua code have been changed to
use the new deprecation system.
The goal is to eventually use an equivalent system for ALL areas of the API.
This covers all the main higher-order functions included in WFL, except zip.
The two already implemented in ai_helper have been replaced with redirection stubs.
This is done by ignoring shroud for path finding (while still taking
hidden units into account correctly), consistent with default Wesnoth
AI behavior.
