reported by Crilith.
To elaborate a bit, the engine no longer accesses resource data through
packed structs. Instead it uses memory streams and the READ/WRITE
functions.
If data is mainly read, not written, I have replaced the old struct with a
new one with a read() function to read the whole thing from memory into the
struct's variables, and a write() function to dump the struct's variables
to memory. In fact, most of these write() functions remain unused.
If data is both read and written, I have replaced the struct with a class
with individual get/set functions to replace the old variables. This
manipulates memory directly.
Since I'm fairly sure that these structs are frequently stored as local
variables for a script, all script variables (both local and global) are
stored as little-endian and accessed through the READ/WRITE functions,
rather than being treated as arrays of 32-bit integers.
On a positive note, the functions for doing endian conversion of resources
and save games have been removed, and some general cleanups have been made
to assist in the rewrite.
Initial reports indicate that this patch indeed fixes alignment issues, and
that I have not - surprisingly - broken the game on big-endian platforms.
At least not in any immediately obvious way. And there's still plenty of
time to fix regressions before 0.9.0, too.
svn-id: r19366
more or less serious bugs here:
* The fnResetGlobals() function returned IR_CONT. Since it potentially
kills its own script resource, this can lead to illegal read accesses.
Not it returns IR_STOP instead. This was probably a bug in the original
interpreter as well, but it handled memory allocation quite differently
so it was probably never an issue.
* Since fnResetGlobals() forcibly closes resources, I've changed the
closeResource() function to silently ignore requests to close resources
where the data pointer is NULL. While it could signify an error, it isn't
necessarily so.
* Don't force the screen to fade up after the credits -- let the script do
it instead. This prevents it from fading up the wrong image.
svn-id: r19127
that:
* Re-worked the elevator script bug workaround so that it's more consistent
with the other two script bug workarounds.
* Some renamings to make it more clear that game events and input events
are two completely different things.
* Added function for clearing pending input events, and used that to fix an
annoying keyboard repeat bug when closing the debug console. (The console
would keep re-opening because the key press to open it kept repeating
even though the key had been released.)
svn-id: r18522
Screen. I've also added an unused "splash screen" function that displays
the image that used to be shown by CacheNewCluster() while copying a data
file from CD to hard disk. ScummVM doesn't do that, but it's a nice image
that I wish we could do something useful with, and it's easier to have the
code here for reference than having to dig through old revisions of the
resource manager.
svn-id: r17894
"restore" other times.) The save/restore dialog now has two separate
classes, though they both inherit from the old combined class of course.
svn-id: r16848
Also, the caching of datafiles depends on the memory usage now, not on the number of screens that the player entered in the meantime.
The old behaviour made the engine run out of memory on the PS2.
svn-id: r16843
The various game settings are no longer stored in the Gui class. They are
stored in the class that use them.
Code that doesn't belong in the Gui class, e.g. the "restart" code, has
been moved out of it.
Afterwards, the Gui class had been reduced to nothing more than a handful
of trivial methods for invoking the in-game dialogs. So the entire Gui
class has been removed.
svn-id: r16827
classes: Screen and Mouse. Screen handles most of the drawing, except the
mouse cursor and in-game menus.
The old Graphics class is no more.
I've also fixed some "reverse stereo" regressions from the first part of
the restructuring.
I'm not sure what the next step will be, but hopefully it will be smaller
than this one was.
svn-id: r16812
In this first step, I have moved all opcode functions into functions.cpp,
instead of having them scattered all over the place.
To get things to compile again, I had to rewrite the overly complicated
sound effects handling. It's much simpler now.
The next step will be to move any non-trivial code out of the opcode
functions and into the appropriate object. This, I hope, will make it
easier to create well-separated objects, instead of the current mess.
I also want to tear down the artificial boundary between the main directory
and the "driver" directory. We already have a cross-platform layer; there's
no need to have yet another one. (Actually, the rewriting of the sound
effects code took one first step in this direction.)
At the final stage, I'd like to get rid of the "drivers" directory
completely, but I'll probably need some help with that if I want to
preserve the CVS history of the code.
Things will probably be a bit bumpy along the way, but I seem to have
reached a point of relative stability again, which is why I'm commiting
this now.
svn-id: r16668
in his mail to scummvm-devel. (Though "a discussed a while ago change"
sounds like sort of thing Robert Jordan writes whenever there is danger of
anything actually happening in any of his more recent books. Tantalizing,
yet non-informative. ;-)
It's still rather messy. I'll look into cleaning it up later.
svn-id: r15818
command, would close the global script variables and player object
resources, without reopening them again. This made them fair game for the
resource expiration mechanism. The player object is probably referenced
often enough to stay alive, but the variables died on me pretty quickly,
causing ScummVM to crash.
I've also added a "reslist" debug command to make this sort of things
easier to spot. By default it only lists resources with refCount > 0. Use
"reslist 0" to see all the cached resources as well.
svn-id: r14958
our other engines do this, so there is little reason for BS2 to. I did add
a filtering mechanism so that mouse button releases and scroll wheeling is
ignored during normal gameplay, but I don't know if that was necessary
either.
Since this left little more than an empty husk where the Input class used
to be, I've eliminated that class and buried its remains in Sword2Engine.
svn-id: r13812
Part of this cleanup involved removing _unpauseZone. It was only used by
fnISpeak(), and as far as I could tell it was just because the original
code didn't trust amISpeaking() and getSpeechStatus() to return sensible
values directly after unpausing the game.
svn-id: r13781
This removes a bunch of debugging code/commands that either didn't do
anything useful under ScummVM (e.g. "soft" and "hard"), or which did things
that was already easily avaiable elsewhere (e.g. "save" and "restore").
I didn't have the heart to remove the "tony" command, though. :-)
svn-id: r13422
of how the savegame is loaded. (ScummVM adds two alternative methods: the
-x command-line parameter, and the restart/restore dialog at the beginning
of the game, which is only shown when there are savegames available.)
svn-id: r13386
it's Logic::_scriptVars[ID] instead of just ID. Apart from looking cool, it
makes it much easier to tell the difference between variables and constants
when looking at the code.
Of course, this sort of sweeping changes is jolly good for introducing
truly weird regressions, which is why I waited until after 0.6.0.
svn-id: r13331
Torbjörn Andersson
Max Horn
Eugene Sandulenko
Robert Göffringmann
Paweł Kołodziejski
Chris Apers