Previously, they would sometimes use desktop dimensions, and on current versions of SDL,
this would exclude areas meant for taskbars, etc. So the resulting image was smaller than it should have been.
Map the R77 F13 key to the former.
Some refactoring of the mapping code in PKeyboardHandler and EventHandler.
Some refactoring of the VidMode handling in FrameBuffer.
Option 'tia.fsfill' is now used to select 4:3 vs. 16:9 mode in fullscreen.
- this means that 320x240 is now the default 'base' size, and zoom levels are based on that
- the TIA image is rendered into whatever sized window is active, meaning that NTSC and PAL modes now look similar
Removed some 'small screen' dead code in FrameBuffer.
- Certain parts of the code referred to int literals, when they should use named constants instead
- Different classes had enumerations for the same quantities; this is now fixed
- Basically, TIAConstants and FrameBufferConstants now contain all constants, and their relationship is clearly indicated (previously it wasn't as clear)
This is most apparent in OSX, where toggling windowed/fullscreen mode can sometimes take up to one second,
and any sounds playing from the ROM become stuck until the operation completes.
Also present on other systems to some extent.
Sound is now muted until the operation is complete.
- Only allow such changes when in emulation mode, or a few other common ones (launcher, debugger)
The idea is here that some dialogs have to be resized/repositioned when such a mode change occurs, and the UI core doesn't currently support that.
- Only render surfaces when they are visible (sounds logical enough, but the code didn't check for it).
- Now uses 'ColorId' as the datatype; this is currently mapped to uInt32, but can change in the future if required
- Eliminates needless and annoying casts in various places; all colors are now 'ColorId' type
- This addresses issue 158, and reduces CPU usage to near 0% when no changes are happening
- This returns the code to the same performance levels as version 3.x.
- It seems that textures must be destroyed *before* the renderer is destroyed
- This isn't mentioned anywhere in the SDL docs, and it works everywhere else, but in any event it is now fixed
- renamed Rewinder and associated dialog to TimeMachine and friends
- changed EventHandler state enum to stronger 'enum class' type (and associated changes to the codebase)
- moved EventHandlerState into separate class, allowing a few files to not need to include EventHandler.hxx (reduce compile dependency)
- plumbed EventHandlerState::TIMEMACHINE into the codebase; still TODO is activate it and add a usable dialog)
Clang 3.8, gcc 5, Visual Studio 2017 and Xcode 8 are now the minimum
supported compilers, and they all support C++14, so we may as well use it.
- Change all make_ptr to make_unique
- Change iterator begin(), end(), etc to const versions where appropriate
- Remove UniquePtr, since C++14 supports it natively
below the current electron beam position.
Fix long-standing bug whereby entering the debugger for the first time
and tracing/scanline advancing, the TIA image was blanked, and didn't
work correctly until you exited and re-entered the debugger.
Minor refactoring optimizations, and renaming of methods.
instead of textboxes. This allows to eliminate error checking,
since the widget will only allow sane values.
Added YStart/Display min/max variables to FrameManager, and
refactor for other parts of the codebase to use these variables.
std:: functions into BSPF namespace at all. So I removed them, and have the
calls map directly to the std:: versions.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3304 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
since when a vector is deleted it automatically calls clear on itself.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3271 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
To anyone reading this, Merry (belated) Christmas and Happy New Year!
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3239 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
of these aren't actually bugs per-se, but are to follow good programming
practices.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3234 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Added updated C++14 make_unique templates, which now allow to also
use unique_ptr for arrays.
Updated SoundSDL class to use unique_ptr, eliminating another new/delete
pair.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3171 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
on every call to FrameBuffer::update(). This will the same CPU usage for TIA mode,
and a slight increase CPU usage for launcher and debugger modes. The code to do
this was just too fragile, and not worth the extra effort.
This fixes several rendering issues, such as garbage in fullscreen mode in Linux
when using Intel GPU, triple-buffering issues in Windows, etc.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3148 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Renamed 'Vlist' to 'VarList', since it's more descriptive.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3092 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
I realized that they didn't need to be stored in a map, since the integer
ID was never actually being used. This must have been part of a proposed
API that I've since deleted??
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3062 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
and bumped version # to 4.5 for the next release (whenever it
may happen).
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3057 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
the advantages of C++11 (move semantics, list initialization, etc).
I'd hoped to somehow wrap a vector behind Common::Array and not have
to change the codebase to this extent, but it didn't work out. And
I've since read that it's bad form to extend from std::vector anyway.
This is *THE LAST* bit of work I'm doing with arrays; everything is
now a proper vector.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3055 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
lists to the Array class? Completely deleting all that code and
using a std::vector directly :)
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3054 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
I've found and fixed a few memory leaks. You will need an up-to-date compiler.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3031 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
than the desktop.
Some code cleanups; add const and optimize some methods.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@3010 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
on top of the dialog surface. This is useful when the surfaces are using
different resolutions, and we don't want to draw the exact overlaying surface
pixels directly into the the dialog surface.
For now, this is most useful for rendering snapshots in the ROM launcher, and
eventually it will allow arbitrarily-sized images to be scaled (in hardware)
to the picture area of the launcher.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2978 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
For the Windows release packages, the VS 2013 runtime libraries are now installed with Stella, so the user doesn't need to manually install vcredist.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2944 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
fix issues in Ubuntu Unity, and caused CPU usage to spike to 100% in
all other tested platforms. So for now, I'm putting it back to low
CPU usage for all other platforms (Linux w/o Unity, OSX, Windows, etc),
and treating this as a Unity bug (which isn't unheard of).
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2935 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
This code was originally there for software mode, where it was very
expensive to update pixels, so it was done as little as possible.
However, it was also a bit of a hack, and sometimes interfered with
double-buffered hardware rendering. So now showing the various UI's
will burn slightly more CPU, but will be guaranteed to work under
all conditions.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2934 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
having to know about TIA, Properties, etc. Basically, it now saves
a snapshot of either the FrameBuffer or an FBSurface, and adds
text comments passed into it. The contents of the surface and comments
are no longer calculated (or known) by the PNG code.
This is in preparation for saving FBSurface from anywhere, which will
help in the debugger for taking snapshots.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2928 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
This sometimes results in a window resize being visible, but better that than
an outright crash. This issue has been reported to the SDL mailing list,
so this fix may be backed out in the future.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2927 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
which I though a just a one or two-liner equivalent to glReadPixels.
Turns out that SDL2 doesn't return 24-bit RGB data (only 16 or 32-bit),
so the PNGLibrary had to massage the data. In the process, converted
the PNG saving functions to actually use PNG library functions, which
greatly simplified converting packing from 32 -> 24-bit. Historically,
the PNG save functions were written before libpng was a requirement,
so they were hand-rolled. Once libpng was integrated into the codebase,
it didn't make much sense to not use it anymore.
So an expected two-line code fix became this commit ...
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2919 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
and (simulated) fullscreen mode.
Fullscreen mode is now simulated in that it creates a window that matches
what would appear in fullscreen. The next thing to do is actually have
the backend create this fullscreen mode correctly.
Changed 'gl_fsscale' to 'tia.fsfill' to more properly indicate what the
option does.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2894 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Changed pointers to references in c'tor calls, making things a little safer.
Removed FBSurfaceTIA, since it was tied too closely to SDL itself. Added a
class called TIASurface that is functionally very similar, but is more generic
and accessible by the FrameBuffer directly. Eventually, this class will take
responsibility for all things related to rendering the TIA image (Blargg TV
effects, phosphor mode, etc).
TIA rendering is currently borked; fixes will follow ...
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2889 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
rectangles. Now that the underlying rendering is hardware accelerated,
this API more closely matches what is actually occurring. As well, it
makes things easier comprehend: 'src' rect is the actual surface data,
and 'dst' rect is its final output onscreen (scaling, etc being applied).
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@2883 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Stephen Anthony