Script::_locker is supposed to track the number of classes and objects
that use a script in order to determine when it can be safely unloaded.
But for system script 999, _locker rapidly increases throughout the game
and eventually overflows. This causes script 999 to unload and crash.
Other scripts may also be affected.
We have been provided QFG1 SCI0 saves where this occurred over the
course of hours within only three days. Logging shows this value
skyrocketing in many games even when they are just polling input or
animating the screen. Alarmingly, this suggests that many SCI games
become unplayable in ScummVM after an achievable amount of game time.
This may be a large and complex problem within the script locker system.
For now, we can mitigate this by making Script::_locker unsigned, which
its callers expected it to be, and by logging overflow attempts and
lowering the locker. This fixes existing save games, even if they have
already overflowed the signed maximum.
fadeSetVolume and fadeCompleted are used by the fade timer to signal
SoundCommand::processUpdateCues() that there is work to be done.
By not syncing them, if a save occurs after a fade completes but before
it's processed then the game won't see the completed fade when loading.
This can happen when the game is paused because the user has brought up
the game's menu bar or control panel in order to save a game. The fade
completes while the UI is displayed, then a save occurs, and then the
fade is processed after the UI is dismissed.
Now completed fades are processed when loading.
Fixes bug #7073 in KQ6 where saving in the Land of the Dead while
Alexander dismounts Night Mare results in a softlocked save.
Fixes bug #13546 in LSL6HIRES where saving while music fades from
a previous room results in a save that doesn't play the new music.
Saving from the GMM should now work for the following games as well:
GK2, Lighthouse, LSL7, Phant2, Shivers and Torin
Many thanks to @sluicebox for his help checking the game scripts of
these games
The existing workarounds for this game's unique way of handling saves
assumed that the player restored a game at some point before the end.
Starting a new game and playing to the end would still cause an error
in the SCI1.1 floppy version because the game's save global would
contain an SCI id instead of our virtual id; we were only patching the
global when restoring.
Now we also patch the global on kSaveGame so that it always contains a
virtual id when there is a saved game.
Note that the existing uninitialized read workaround had no effect since
the read occurred in a local procedure but the workaround entry didn't
include a local procedure signature. Updating that workaround wouldn't
fix this; there would still be scenarios where the wrong save would be
auto-deleted.
Fixes bug #5294
This mainly concerns restoring sounds after loading savefiles, but it should make the whole relationship between playing and paused sounds more accurate.
The test case which I was told about was KQ4, room 21, picking up the golden ball under the bridge, saving during playback of the pickup sound and then loading that savegame. It would result in hanging note due toe the sound being triggered multiple times by reconstructPlaylist() and updateSci0Cues(). Now, the sound should only start once.
I've changed the code to be more in line with disasm and tested some situations that sluicebox told me about or that I found in the comments (ICEMAN room 14, LSL3 start scene). I got rid of isQueued, since the original doesn't have that, it has caused some confusion and doesn't even get saved with the savegames.
I cleaned up updateSci0Cues(), so that it (together with processUpdateCues()) does a bit more what the original Midi timer proc does there. An exception is the sound fade out code in processUpdateCues(). It seems that we need that, as we don't have the fading code in the drivers like the original.
The original SCI0 code is actually much simpler than our code. It relies on a correctly sorted playlist (based on priority), but my impression is that we got that right, even if we do it slightly differently. I added a sortPlayList() to the sound init, since the original inserts the node at the right position, too.
- Common: add wordWrap function to ustr.cpp
- Bladerunner: Explicitly state we have a U32String in subs (same as Subtitles::loadOuttakeSubsText)
- Don't use translations for engine specific "put strings", because they might not support.
- SCI: Use const references for showScummVMDialog
- SCUMM:
-- Don't use translation in md5 warning. left comments with the translated version.
-- Remove some redundant headers in help.cpp
-- Don't use translation in handleSaveload when printing to console
-- Also, display success transaction correctly via u32::format
- TESTBED: Use fake constructor when setting label of button
- SKY: Correctly use translation when using SaveStateDescription
- ULTIMA: Don't use translations when display_string
- ENGINES:
-- GenerateUnknownGameReport correctly, with proper translations.
-- There was an error, where a function had been declared twice, in a header file. Correct this.
- Where necessary as in, where translated messages are used.
- GUIErrorMessage now takes in U32String
- error messages across some engines use U32Strings. they are changed because they show a message dialog.
After the initial changes just to scummvm/gui for u32, this commit includes the whole project
- Widget creations now always have u32 descriptions, labels, or tooltips
- Message dialogs make use of default arguments instead of providing the same argument explicitly
- encode String::format properly before passing on as argument where necessary
- Modify hugo utils (yesNoBox and notify box) to use u32
- Also provide fake constructors for the above which redirect to the u32 constructor
- Convert all keymap descriptions to u32 across all engines
- showConfirmationDialog in mohawk now uses u32
- showScummVMDialog also uses u32
- Scumm engine has dialogs now which use u32
- General fixes and wrapping convertToU32String for setLabels and related functions
- Add a fake constructor to MesssageDialog which redirects to the u32 constructor
Factors some common save game code into functions and creates an
overload of GuestAdditions::runSaveRestore that doesn't rely on
SCI memory for returning the description string, all of which will
be used by the custom save/restore kMacPlatform32 subops.
Loading is not normally allowed while audio is being played in SCI games.
Stopping sounds is needed in ScummVM, as the player may load via
Control-F5 at any point, even while a sound is playing.
Scaling works correctly with the Windows variant, which uses AVI files,
but the DOS variant uses Robot videos, and the way scaling is done
there is different, and is not working yet with KQ7 DOS. Nonetheless,
both versions are included in the game, so it's not a major issue.
This removes the unnecessary Buffer subclass and stops most places
where the output buffer was being interrogated about dimensions
instead of GfxFrameout.
* Replace raw pointers with smart pointers
* Use references instead of const pointers where appropriate
* Tweak initialisation
* Tweak palette copies to the stack
RAMA has its own custom save game format that game scripts write,
but we still want to be able to use these save game files from the
ScummVM launcher, so the metadata has to be able to be written
separately from the rest of the game saving.
To enable the optional content censoring mode, Phant2 looks
for a RESDUK.PAT file, which is normally placed by the game's
installer if the user chose to enable censorship. If the file
exists, the game reads an unlock password out of the file and
asks the user to enter the password when starting a new game to
create an uncensored game, or to click a "less intense" button
to start the game with censoring. The censorship state of the
game is then persisted in the save game file, and installations
with the RESDUK.PAT file need to enter the password again in
order to restore any of the uncensored saves.
Since we do not have an installer that can enable this feature,
add a game option toggle to enable/disable censoring (for the
releases that have the optional censorship mode) instead so the
censored content feature is available for anyone that wants to use
it. This flag is restored from ScummVM whenever a save game is
loaded, so it can be toggled on or off at any point without
needing a separate save game, unlike in the original interpreter.
These objects should have been initialized only during the first
pass. Double-initialization does not cause any visible problem
problem during normal operation (mostly it just causes memory
waste by making Object::_baseVars/_baseMethod double up their
data), but could have silently allowed games to receive bogus data
for an out-of-bounds property or method index, instead of raising
an error.
Upon investigation of Sound code across SCI32 games, it was
determined that there are actually (at least) 3 different
revisions, not just a single SCI2.1 version. This patch only
changes the parts of Sound code that are relevant to the correct
use of Audio32.
Fixes Trac#9736, Trac#9756, Trac#9767, Trac#9791.
1. Unlocking all resources of a type using a resource ID of -1 is
gone in SCI32;
2. Audio locks need to be serialized starting in GK2 for the game's
modified kDoAudio(1) call;
3. Audio locks in SCI3 must work more like SSCI, since at least
Lighthouse's `BackMusic::fade` method will attempt to unlock
audio that was never locked by a script. In SSCI (and now in
ScummVM too) this is a no-op; previously in ScummVM, it would
remove Audio32's own lock on the audio resource, resulting in a
use-after-free;
4. kDoAudio(1) starting in GK2 returns the number of active
*not-in-memory* channels being played, not the total number of
active channels.
Fixes Trac#9675.
This situation occurs rarely, but normally, when unreachable but
not yet GC'd objects use a superclass which has already been GC'd.
Thanks to @wjp for looking at this with me and clarifying what
was going on.
ObjMap owns Objects, so every time this map gets copied instead of
referenced, it creates a copy of every single object in the
associated script. This is expensive, and it breaks things like
the `Object::syncBaseObject` call in savegame.cpp, which hasn't
actually been doing anything since
58190c36b4 because it has been
operating on copies.
Save game metadata validity checks in SCI32 should all exist within
kCheckSaveGame32 since this allows most games to recover
successfully from an attempt to load an invalid save game. If
gamestate_restore fails, the game will usually crash because the
engine is left in an inconsistent state (game scripts have cleaned
up objects in preparation for a game load that is no longer
happening).
Script buffer data is modified after a script is loaded by
savegame operations, and, in SCI16, by string operations. Casting
away const to allow these mutations to happen is not a very good
design, so this patch just changes the privately held reference
to data to be mutable. (Public accessors still return immutable
data.)
This information comes directly from script data and is not
modified at runtime, so it does not need to be persisted in save
games, but does need to be set when reconstructing clones.
sluicebox
Filippos Karapetis
Eugene Sandulenko
Orgad Shaneh
athrxx
Cameron Cawley
aryanrawlani28
Zvika Haramaty
Colin Snover