at a time. Of course, it doesn't actually do anything yet.
Logically, the algorithm is very simple (or at least I think it is).
I just have to figure out how to execute the 6502 until a scanline
is done. In the process, the internal buffers (and hence the
external FrameBuffer) will be filled with data, and we just do
a FrameBuffer::refreshTIA() to 'see' the changes. The first part
of this is the problem. Any advice Brian??
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@656 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
However, we do leak memory: yyparse() happily allocates Expressions as
it parses, but when it hits a parse error, it doesn't return a valid
pointer to the top of the Expression tree. From what I can tell, the
so-called Expression* result is the int value of the last lexer token
cast to an Expression* (due to yacc's use of a union).
I know how to avoid the leak: we need to keep a vector of Expression
pointers in YaccParser. If there's a parse error, yyerror() can delete
all the Expressions using the vector. If not, we clear the vector (er,
calling .clear() on a vector doesn't delete all its elements, too,
does it?). Every time yacc says "$$ = new WhateverExpression", it also
should vector.push_back($$).
Will implement this tomorrow; am getting tired & flaky.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@655 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
to include spaces in arguments. "print 80" and "print {80}" are equivalent,
but "print 80 + 1" and "print {80 + 1}" are not (the former counts as 3
arguments, the latter only one).
Actually, I haven't tied the YaccParser to getArgs() yet, so both would
still be errors. I still need to fix the segfaults on parse error.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@654 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Semi-fixed a bug: formatFlags was capitalizing the wrong flags. I'm
pretty sure it's because they were being pushed into the BoolArray in
reverse order. I "fixed" formatFlags by having it ignore the BoolArray,
and call my new CpuDebug methods to get the flag states directly.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@653 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
but decided we'd be better off with just one IntMethodExpression class.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@652 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
it, then try changing the A register, then "expr". It should return the
changed value.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@651 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
a convenience: <foo is equivalent to foo&$ff (or foo%$100), and >foo is
equivalent to foo/$100, assuming foo is a 16-bit quantity.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@650 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Changed the % prefix for binary input to \ (backslash). It simplifies
the lexer/parser syntax *greatly*. Might change back at some point.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@649 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
be expanded to pop up a palette when double-clicking on a color,
to visually choose a new one.
Added ColorWidgets for the 4 TIA color registers to TiaWidget.
Since these are tied to a DataGridWidget, a nice side-effect is
being able to press '+' and have all the colors cycle.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@648 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Some work on TIADebug. Change tracking now works for the color
registers. Getting/setting the color registers works (sort of,
since the change doesn't seem to be registering in TIA).
Some work on TiaWidget. Color registers and scanline count are
shown. VSync and VBlank are shown, but can't be changed.
Made CheckboxWidget only emit a signal when interactively
clicking on the checkbox (changing state programmatically no
longer emits the signal). TODO: maybe two signals should be used??
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@647 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
At this point, you should be able to do this:
> define foo 1
> expr foo+1
result: 2
> define foo 2
> expr
result: 3
> undef foo
> expr
result: 0 <-- because undefined labels are treated as -1. Bug?
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@646 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Added a static Debugger::debugger() method, which is sort of a hack,
but is very useful when some class needs access to the debugger
and we don't want to pass the debugger object through dozens of
c'tors. Besides, there's only ever one instantiation of the
debugger anyway.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@640 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
but at least Stella doesn't crash. I have to implement dialog
resizing to take care of the remaining problem (vs. just deleting
and re-creating the debugger dialog).
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@639 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
and call evaluate() on it repeatedly... Ladies and gentlemen, we have a
compiler!
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@636 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Yes folks, it's coming, and if everything works out it will be the
fastest implementation yet for any Atari 2600 debugger.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@634 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
label => address mappings. It should be faster than the linear search,
and it's a damn sight cleaner. Still TODO is to restore the case-insensitive
string matching from the original version.
Changed a few EquateList methods that used to return char*, so they now
return string or string&. Changed other classes that call them so they
work with the new return types.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@633 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
instruction at a time, and before each instruction it disassembles the
next instruction. If the disassembly result contains the string arg,
stop executing and return.
This let us say things like "runto STA" to run until the next STA
instruction, or "runto LABEL", or whatever else. It does *not* use the
breakpoint nor trap mechanisms.
Unfortunately, as written, Debugger::disassemble is *slow*... "runto"
takes about 40-50 seconds to execute one frames' worth of 6502
instructions on an Athlon 1200. Also, as written, it's uninterruptible:
while it's slowly doing its thing, the SDL main loop isn't running,
so you can't even close the Stella window to get out...
Also, it would be nice to be able to tell "runto" which field you want
to search (address label, instruction name, operand). Splitting it out
this way could also make it faster: instead of doing a full disassemble
and searching it, we could have "runto LABEL" evaluate the label and
just stop when the PC equals the target (would be almost as fast as
just using a breakpoint). We may create separate commands for this:
"runtoaddr", "runtoinstr"...?
Being able to say "run until the next RTS instruction" is a lot more
expressive than having to disassemble until you find the RTS, set a
breakpoint after it, then "run", then remove the breakpoint...
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@632 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
different ways, and hence sometimes certain things were done
and not at other times. Now there's only one way to enter/exit
the debugger. This fixes the problem of locking/unlocking the
data bus in debugger mode; a lock is done on entry and released
on exit.
The above change may expose other bugs. I haven't tested
extensively, but if something has stopped working it's because
it relied on buggy code.
Cleaned up the Debugger class definition a little, adding some
documentation and making certain things private when they didn't
absolutely need to be public.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@631 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
System when it uses peek() to examine memory/TIA/etc.
I'm not entirely sure I put a lockDataBus() call where it will get
executed when the user presses ` to enter the debugger: I put it in
Debugger::start() (does everything call this, or does EventHandler do
something else?), and again in Debugger::state() (which gets called to
build every prompt string).
There are unlockDataBus() calls everywhere in the Debugger where we
execute 6502 code, including Debugger::quit(). Does the EventHandler
call quit() when it exits the debugger (if the user presses ` I mean)?
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@630 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
so it's just one test (search for STA $3E in the image) to distinguish
between the two. I still only have 3 3E images (notBD PAL and NTSC,
and Armin's test cart), but all 3 are detected, at least. Autodetection
isn't as important in Stella as it is in some other emulators, due to
the stella.pro database, so I won't try to make this too smart for its
own good...
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@629 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
if this is needed for anything other than 3E type, but since it
doesn't cause any overhead, I went ahead and did it anyway (after
reading your messages on Stella list).
Please test to make sure it still works, but I don't think there'll
be any problems.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@628 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
the TIA. For 3E, this means game authors don't need to access the TIA
through a mirror (and there's no reason they should on real hardware,
either). It also means Armin's test cart image works.
3F ROMs still need to use a mirror: poking to $00-$2C writes to the
TIA *and* switches banks, which is not that you'd want to do. Stella's
behaviour now matches a real Tigervision 3F cart, for what it's worth.
Unfortunately I had to do something nasty to allow the Cart classes to
directly access the TIA: see System.hxx and System.cxx (search for
the tiaPoke() method.)
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@627 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
CpuDebug. This is similar to the last commit, which moved all RAM
stuff to RamDebug.
Removed D6502, since it's functionality is provided by CpuDebug and is
no longer required.
The only user-visible change is that the CPU registers in CPU tab now
track changes. It may seem like a lot of work to just accomplish that,
but the infrastructure will make it much easier to add similar
functionality to new tabs. Besides, we needed to step back a little
and clean up the codebase.
Still TODO is turn TIADebug into a real DebuggerSystem object, and
finish the TIA tab.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@624 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
I'll be doing small commits after each subsystem is ported, to make sure
I don't do anything too destructive.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@623 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Cart3E::patch() now can patch RAM (before, it would have segfaulted if
you'd tried)
Cart3E constructor now initializes RAM to random junk instead of leaving
it uninitialized (which still means full of junk...)
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@621 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
with Andrew Davies' early notBoulderDash demo.
Added entry to stella.pro for the demo
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@619 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
a reimplementation of D6502. So D6502 will disappear soon. The reason
Debugger class has been getting so big is that there was only D6502 defined,
when there should have been also DTia, DRam, etc. The DebuggerSystem
classes are now providing that functionality.
The codebase may be broken over the next few days because of this, but
it's better to do it properly (I get the feeling it may never be fixed if
we don't do it now).
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@616 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Debugger class. There will be a subsystem for each GUI debugger tab,
and it will have responsibility for all thing related to that tab,
eliminating a lot of code from Debugger class.
Added RamDebug class, and changed the GUI RAM tab to use the new RamDebug
subsystem. The prompt still uses the old method, which will have to be
fixed. At that point, Debugger::(readRAM, writeRAM, ramChanged) can be
eliminated. If the Debugger ever wants to access RAM related things,
it should refer to RamDebug.
Still TODO is create a CpuDebug subsystem, and turn the current TIADebug
into a real subsystem.
Cleaned up some redundant typedefs (changed AddrList/ValueList to
IntArray). As well, we're moving away from having bspf stuff seeded
throughout the system. Integer values will be stored/passed around as
'int', and truncated where necessary as close to the given method as
possible.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@615 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
even starts, so you can enter the debugger before the first frame is
even drawn. This will be useful for everyone, but I particularly need it
for testing per-scanline TIA update code (which doesn't even exist yet).
I should eventually be able to run "stella -break kernel poker.bin" and
then advance one scanline at a time, watching the first frame get drawn.
Not even close yet :)
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@614 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
Increased debugger window width to 639 pixels. We've already passed the
512x384 mark, so it's no use restricting width to 512 anymore. Besides,
the extra width will come in handy for the TiaWidget, and maybe the
PromptWidget can now use one less line in the state() display.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@613 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
of RAM in the TIA, but it isn't tied to the TIA yet (so changes have no
effect). I suspect that development on this tab will now proceed quickly,
since I think I've found a way to lay everything out in one tab.
Removed the 2 second delay from configure when an invalid option is
given. There's no need to delay; printing an error is sufficient.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@612 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
for 2 seconds instead of silently ignoring wrong options. This seems
like a good compromise to me (I've typed "./configure --disable-opengl"
about 50 times now, and of course it doesn't work because it's supposed
to say "--disable-gl")
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@611 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
system, and if that doesn't work for someone, we'll fix it at that level.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@610 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
are part of the class itself. That means all DataGridWidgets (currently
RAM contents and CPU registers) get access to the key shortcuts.
Modified DataGridWidget to accept the number of bits in its cells, and
made sure the above listed operations are restricted to that number
of bits. That means when we add DataGridWidgets for Tia stuff with
weird bit lengths, changing data will work correctly.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@609 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
has sent a signal (required when there are more than one of the same type of
widget in a widget/container).
Still TODO is modify each handleCommand() method that cares about 'id' to
actually detect and act on it.
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@608 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
a row/column is reached, it goes to the next row/column. The reason for
this will become apparent over the next day or so.
Still TODO is revamp the base widget to accept and return an identifier
for which widget it is, since every widget/dialog in the debugger
currently assumes that there's only one type of each widget (so if there
are two DataGridWidgets, the returned signal doesn't know which one
to use). Sorry if that sounds confusing ;)
git-svn-id: svn://svn.code.sf.net/p/stella/code/trunk@607 8b62c5a3-ac7e-4cc8-8f21-d9a121418aba
stephena