UI cutoffs prevent any any UI layers with index <= the cutoff index from
being displayed. If there's no cutoff, the cutoff index should be -1;
the old value of 0 was causing the lowest UI layer to not be unhidden.

The previous implementation added a tooltip timer before blocking for
the next SDL event, and removed it before returning. If resize or move
events were received, however, the function would return early, and the
timer would never be removed and would instead timeout periodically.
Subsequent move events would add additional timers, leading to a flood
of timeouts clogging up the event loop.

This new implementation only ever adds a timer when the mouse is moved,
and timers only fire once. remove_timer() is also changed to take an
int& for convenience.

The menu already ellipsizes menu items as needed based on screen size.

Widgets may query the current scissor region to reduce render calls,
e.g. by not rendering children or other elements that are not visible.
During a regular layout and render cycle, the UI framework pushes a
scissor region of the entire screen, so this works fine. When doing
ad-hoc rendering (while we transition to full use of the UI framework),
this is not the case, which caused crashes.

m_wrapped_lines being empty caused the size request calculation to yield
a natural size (0 lines) smaller than the minimum size (1 line), causing
assertion failures.

The handle_terminal_resize() call in startup.cc (added in e03bc23)
seems to no longer be required, since the code to determine the HUD size
based on species (added in the same commit) is no longer present, and
thus the HUD should not need to be resized any more. This permits some
minor simplification of handle_terminal_resize().

Pre-wrapping to a given screen-width should be avoided in general
because the screen can be resized. In addition, the GOTO_CRT region size
is not a well-defined concept on local tiles.

The `get_tension` check is not well-defined without a player, and also
isn't used for monster attackers. Also, the singing sword was shouting
at the player's location, not the attacker (which caused another crash
in arena).

Because of an incorrect formula added in a559921, Mara and rakshasa
clones had 200% hp of their original instead of 50%.

Regular monsters copied via `cast_phantom_mirror()` had 286% hp
and max hp instead of 35%.

Debug builds unsurprisingly spew large numbers of messages, so
repeatedly showing a more prompt is very inconvenient.

The call to expand_mlist() assumes that the monster list has its initial
position and size; this is not the case if you run arena mode more than
once without quitting the game in between.

After this fix, the UI is still broken if you run arena then a regular
game. I tried using an unwind_var, but that caused crashes, and a better
fix is in the works anyway (using the widget toolkit).

A copy created via rakshasa's Phantom Mirror has only 35% of
the target's hp and max hp. If the original monster is almost dead
(1-2 HP), its copy will have 0 hp, which crashes the game.

Fix this by giving cloned monsters at least 1 hp. Also, it allows
rakshasas to clone lurking horrors, who always spawn with only 1 hp.

Apparently it was previously possible, for some definitions of the
word 'possible', to have multi-key macro binds; e.g. 'ab' -> foo. I
couldn't actually get any of these to trigger, however, even when using
xdotool to quickly deliver keypress events. This commit simplifies the
complicated macro expansion code that resulted. Saved macros that have
multi-key binds are silently truncated to the first character.

This commit also slightly refactors filtering of empty macros.

This makes the difference between getchm and getch_with_command_macros slightly
clearer.

While you could theoretically have multiple keys returned from
_getch_mul, in practice I never managed to make it happen, and it's
preferable for the code paths through the macro system to be as simple
as possible.

We only ever add a single key at a time to the macro buffer; better to
enforce that.

This also asserts in remap_key() that the macro buffer is empty before
remapping a key; this must be the case because ui::pump_events() first
tries to get a key from the macro buf, and only remaps a new key if
there is no macro key available.

Returning std::move() here is fine because the parameter is an rvalue
reference, so RVO is not relevant, and a move() call is required to
prevent a copy.

Spell range strings don't use colour codes any more. As for explore
discoveries, it's simpler to parse the tagged string and then use the
dedicated width() method.

(cherry picked from commit 710a5d0)

(cherry picked from commit 883716c)

A better fix would be to use the just-removed align_self, but I'd rather
not revert that removal, and adding a better implementation would be too
disruptive at this point (feature freeze).

(cherry picked from commit 6cf4ce5)

(cherry picked from commit 3121d2d)

Updated tiles for Sif Muna's Divine Exegesis and Fedhas' Wall of Briars,
Overgrow, and Grow Oklob abilities.

(cherry picked from commit 29a1220)

This fix drops mouseup events if the UI layout has been changed since
the corresponding (well, previous) mousedown event. This fixes a bug
where the loading screen would transition to the main menu on mousedown,
and the mouseup event would then activate the button underneath the
cursor.

I'm not sure this is the best way to fix this. Some alternatives:
- have the loading screen transition on mouseup
- have the loading screen "swallow" the mouseup event explicitly via
  some new API
- have menu buttons only activate on mouseup if they also received a
  corresponding mousedown

(cherry picked from commit 47823c4)

(cherry picked from commit 6afad29)

This is necessary if scoring (or perhaps sequell) is to do anything
interesting with seeded games. This commit adds the seed for all game
types, since even for normal games the seed is revealed on game end.
Milestones do not have this field for any game type.

(cherry picked from commit 6c54e25)

(cherry picked from commit 3d4e485)

(cherry picked from commit 1393849)

It is tactically useful to know how much damage is due to flay. For
example, it matters in deciding whether to quaff cancellation or who to
attack.

This information is already revealed to the player, but not in a
convenient way. At the moment the careful player must pay attention to
what the sources of damage are as the fight proceeds in order to know
this; and, if this is neglected, it is difficult to recover the
information from the message log later in the fight.

This commit reveals that information more straightforwardly by adding it
to the flay status light. I show a number instead of a more
coarse-grained indicator. Note that otherwise the conscientious player
will still want to keep track manually of damage since they can get
better information that way.

The damage is displayed as a negative number for consistency with the
similar corrosion status light.

(cherry picked from commit f18fdeb)

It turns out that multi-key macros were used for keybinds for
multi-character sequences sent by some terminals corresponding to
special keys. For example, under the VT100 standard, \{27}Ov is sent by
numpad 6. Another example: on OS X, option+arrow keys send
multi-character escape sequences as well (maybe even outside of
terminal). This sequence of commits broke such old macro files, and even
worse, tended to break ESC, since nearly all of these sequences are ESC
sequences. We'll need to rethink how to simplify this.

Some of these simplifications might be independent of the current revert
project and could be restored...

This reverts commit 049a785.
    (Note: also a partial revert of d158fba, 0b1efcb, 0e779d2)

This reverts commit 678ee33.

This reverts commit 5a8975c.

This reverts commit 3ea5ddb.

(cherry picked from commit 72dcfe0)

Specific problem and solution found by @Wise Wolf on discord yesterday.

(cherry picked from commit dc1bd94)

At first glance, this may appear to be a feature, but it is in fact a
bug: the bug is that paste doesn't work for crt-emulated input.
Implementing this as an actual textbox is the only way I've found to get
this to work in webtiles.

This doesn't materially change the implementation of the seed selection
ui for console/local tiles; it is just as bad as ever.

(cherry picked from commit 74baf03)

Not sure what the conditions are that trigger this, but probably at
least game_exit::abort.

(cherry picked from commit 8315797)

(cherry picked from commit b8d28a4)

(cherry picked from commit 1381ef2)

(cherry picked from commit 2ac9325)

853fe2e fixed this for harming allies, but not yourself.

(cherry picked from commit 6843c08)

(cherry picked from commit 75acf02)

The vault cheibrodos_wizard_tower placed on-average 3 8-tier monsters at
Lair depth, which is fairly nasty. This commit reworks the monster and
loot placement to better scale with depth, starting with 1.25 monsters
at D:7-9. The monster set now includes a possible deep elf annihilator
in Dungeon, but only at depths where those are possible 8 spawns, and
with a small chance. The monster count in Elf has been increased as
well.

The vault range is now increased to include D:13-D:$. The loot now has
some thematic aux armor, randbooks, and staves mixed in, and the loot
scales in count with depth.

(cherry picked from commit f0389da)

This has been broken for a while; it was a mid/mindex confusion. I've
also refactored the call so that it will better handle the case where
your summon does before the fineff fires.

(cherry picked from commit c466198)

We'll see if players think this is too spammy or too obscure, but for
0.24.0 at least this will be helpful for testing any seed issues that
come up.

(cherry picked from commit 9c56bcb)

(cherry picked from commit 72b0ba0)

0.24 "Plants vs. Vampires"

…ild.