upstream uses a hotkey toggle instead but vanilla didn't have it. even "always run" itself was just a vanilla bug. I don't want to have a special on-the-fly input for autorun toggle, hopefully user either wants to always run + occasional mouse movement, or only run using the proper button
this allows to properly set available range which makes demos sync again (aside from weapon switching that I broke)
when importing demos we now force shorttics (even tho some source ports can record demos with longtics depending on compatibility level but importing those is for the later time)
There's 3 approaches here, of which only 1 appears to work more or less correctly
1. Use XFixes "pointer barriers." Introduced in XFixes version 5.0 (end of 2010), these effectively allow 1:1 mapping with Windows ClipCursor, giving behavior BizHawk wants.
2. Use XGrabPointer, with the confine_to argument set to the presentation panel. This mostly works, however, it changes how some events operate and for whatever reason prevents Mono from responding to mouse buttons. This approach can be used with 1. too for a simple warp of the mouse cursor over to wherever the window is.
3. Use Mono's internal CaptureWithConfine function. This is just internally using XGrabPointer with the confine_to argument. Somehow it makes mouse buttons work, but it ends up working too well as it can respond to the menu bar somehow, and interacting with such or the right click menu cancels the grab (seems to be internal Mono code doing such in this case here?).
Note a lot of weirdness with previous code came down to testing being in a VM and mouse integration being enabled. That apparently just prevents any kind of mouse warping from happening so nothing appeared to work correctly. Disabling such allows these approaches to work as above, also makes relative mouse values sane (previous values with mouse integration were completely bonkers).
TODO: Need to check if this works on XWayland (maybe it does?). Also need to refresh mouse capture on window resize and move (needed in general for both Windows and Linux)
currently requires manually binding RMouse X/Y in config.ini (and zero deadzone I guess)
empirically match dsda default settings with sensitivity syncsettings (defctrl.json wasn't touched yet)
TODO: look at mouse calc inside the core
that way we don't have to rebuild the core for every commandline option that it already supports and we decide to use, also more meaningful presentation of those options on the managed side
native api side done. trying to map it probably interferes with absolute inputs, needs some global relative mouse sensitivity setting for input translation purposes. hopefully everything else doesnt break
attempts to resolve#3053 by shuffling some bools around.
Specifically, `_runloopFrameProgress` is now set when either (continuously) frame advancing or rewinding.
removed the NETCOREAPP3_0 code because it didn't even compile; lots of internals missing and we probably wouldn't need the 1% speedup anyway (the function using it is unused currently too)
by immediately throwing out the current movie. This should prevent this message box from coming up repeatedly, not providing any additional useful information and just blocking all other windows.
should resolve#2332
fixes 18078c876
and some others, maybe as far back as the commits introducing the lists,
I haven't checked
also I left a marker so it's easier to find these later
fixes#4204
RANT TIME
visible area is largely nonsense when it comes to CRT TVs, because they may cover different amounts of the screen based on model, and you can adjust some of them also based on model to show more or less of it.
4:3 aspect ratio is also somewhat nonsense because TVs don't automatically rescale your video input to the entire screen like modern computers do. they just stretch or shrink it based on pixel frequency of the input and how it relates to whatever the given TV is configured to.
Rec601 was created in order to somewhat standardize this but it only contained RECOMMENDATIONS (hence the name) on how to digitize analog video, not how analog devices should be configured. so a lot of TVs looked different from what it would be digitized to based on this spec.
in gaming it's a common misconception that console output is meant to be resized to 4:3 DAR of the TV. in reality it's stretched according to PAR - pixel shape which comes from differences between color frequency of the input and standard NTSC or PAL color frequency.
if the console updates color more often than the standard, analog signal with higher pixel density will be put on the screen as is, and pixel will look "squished", sometimes up to 2x for 512px resolution modes (PS1). if the console updates color rarer, pixels will look stretched (A2600).
Amiga PAL mode has 1:1 PAR, no stretching is needed, so whatever aspect we're outputting will just go to encodes as is. NTSC Amiga PAR is 6:7, so we're shrinking a bit in encodes and when hawk is configured to device aspect.
this all means that it doesn't matter how many pixels we take of the rendered image, all we care about is stretching it to proper PAR. whether it contains overscan only affects fullscreen because extra blank area means less of useful data. and since some exceptional games decide to render AT THE VERY BOTTOM there's little to no harm in showing all 574 rendered scanlines. who hates it and wants 568 can crop it, but I don't expect anyone to care (or even notice).
so while nominal internal res for Amiga is 576 and canonical UAE default is 568 (tho it lets you adjust visible area and we don't), I think we can safely output our heretical number at all times.
Semicolons are perfectly valid for path names, but windows disallows dll paths to contain such (presumingly internally using them as path separators like with PATH env var)
A workaround for this is to simply use the "short name path" instead, which does not have a semicolon (but these names aren't guaranteed to exist, and in certain cases if they don't,GetShortPathNameW will just return the long name back, so semicolon still needs to be rechecked)
[C64] Disk: Use raw track capacity values per the G64 file format specification, should fix some disk loaders that are expecting data not to be so sparse (Spindle demos in D64 format particularly)
[C64] Disk: fix 1541 drive address decodes
[C64] VIA: 6522 core facelift, shift register and use of low-order timer latches implemented, should fix some disk loaders ("Sprite B*****e 2" demo plays, yes they called it that, and it's not what it sounds like)
[C64] CIA: fix PB6/PB7 outputs when enabled on CRA/CRB
* [6502] Pass Lorenz C64 tests
* [C64] make sure the 1541 drive uses the same 6502 undocumented behavior as the main CPU
* [6502] Use field instead of delegate for ANE/LXA system constants
doesn't matter for ASCII-only strings, and `InvariantCultureIgnoreCase`
is arguably correct in some circumstances, but IMO it's more foolproof
to simply ban it
This reverts commit b2f3bb3cba.
Revert "fix removing everything"
This reverts commit a0da874431.
YEP. this "feature" is COMPLETELY, ABSOLUTELY, UNIVERSALLY fucked. sure #4184 can be "fixed" by flipping some bools, like setting MainForm.HoldFrameAdvance back to false in TasView_MouseWheel. but then there's still the problem of removing actual input if farther lag picture changed after initial removal, and I'm not going to debug that. taseditor's ONLY bug was related to erroneous detection of lag change that we were never able to consistently reproduce or figure out, and it's completely impossible to replicate identical behavior in tastudio due to crazy overhead. but even SIMULATING that behavior would involve touching that minefield of bools that control everything in insane ways. given zero requests for this feature during tastudio's decade of existence, I'm considering it too useless for all the chaos it introduces.
so instead I'm closing #4184 by disabling the "feature" that caused it
* BizHawk#4187 Add a method to the comm LUA library to allow setting ExpectContinue
* Instead of exposing ExpectContinue set it automatically based on a threshold
* [C64] Fix loading Domark cartridges that don't have a bank count that is a power of 2, and add a memory domain for the ROM image
* [C64] Fix bracket styling
- Implement the AM29F040B flash device as its own component
- Fix save data for games that use the standard EasyAPI driver
- Fix "Briley Witch Chronicles 2": the flash driver checks AutoSelect register 02 to see if the cartridge is write-protected
- Rom is stored as bytes instead of ints, saves a lot of memory
- EasyFlash now has proper hard reset and clock operations
- SaveRam deltas fixed
- closes#4157
Not sure this is the best way to fix the linked issue, but considering this is a hard crash I'd rather fix it somehow than not at all for now.
controls that don't belong to any indexed player are put at index 0, which makes them appear in front of everything that's used all the time. when there's only a few of those, it's fine. but for systems with keyboards, you have to fullscreen the emulator window to even see regular inputs
I moved index 0 to the end for input display, but didn't touch the original generator, since it may affect other things I'm not sure about (and I'm not smart enough to properly change it)
from be0736be4 (+ cherry-picked ccba93440) to this commit:
- `Database` reported runtime 0.64 s --> 0.68 s
- profiled `ParseCGIRecord` mean runtime 0.190 ms --> 0.179 ms
- `GameDBHelper` 1M lookups reported runtime 1.03 s --> 0.69 s
- managed allocations (w/o the 1M lookups) 131 MiB --> 123 MiB (fewer
strings but more char[]s?)
since `CompactGameInfo` obviously can't store `Span`s, I tried with
`StringSegment`, but it made both lookups *and init* slower, thanks MS
Includes: `Range`/`Index` and other polyfilled types are now `public` in
BizHawk.Common rather than `internal` in each assembly. Not for dedup
reasons--though that's probably good too--just because it was necessary
for exposing these new `public` methods with `Range` in their params.
fixes 8645ed3bb
so the speedup isn't as big as with the broken code, but it's close
I think I'll be changing this again to use Span and I'll reprofile then
while libretro initially ported original puae (which is long dead), they switched to just pulling updates from winuae every once in a while (and renamed to libretro-uae), because it's still active and is kinda considered golden standard these days
The fallback is intended in cases where there is a misdetection for MAME ROMs, since it just goes off the archive's filename against a db. MAME ROMs will not have any sensical ROM extensions in them, and will most likely just have multiple files in it. As such, if this is actually a MAME ROM, they're pretty much guaranteed to show the choose file in archive dialog, which would confuse users. If it's not a MAME ROM, it most likely will not show this dialog and will just be loaded by the appropriate core (of course in practice this likely doesn't occur anyways, since most users likely just have their ROMs named with dat names like No-Intro etc which never match MAME's names).
- closes#4133.
This has a 50% chance of breaking something else (it either happens or doesn't), but should be more correct. Mainly, RecalculateScrollbars was called based on visible rows which depend on whether scrollbars are visible or not, but the NeedsVScrollbar calculation did not take the existence of the HScrollbar into account, leading to an off-by-one. I think.
it has to rewind to remove a frame, which makes it pause on first removed frame when we hit restore, instead of going all the way to LastPositionFrame. I tried a ton of things to fix it, and they break one of the 3 scenarios of going from the edited frame, emulating through removed was-lag frames until LastPositionFrame, and pausing there:
- autorestore
- restore manually via middle-click or the || button
- step frame by frame using wheel+RMB
wheel+RMB steps used to unpause in wrong places because paused state was disabled when it fires, so we couldn't decide whether to unpause or not based on it at the end of AutoAdjustInput(). so I'm enabling """paused""" state inside TasView_MouseWheel(), much like < and > buttons do, except they can be held to force unpause, but here you can't let go of mouse wheel since it already only fires once anyway. I'm disabling this state once tastudio has reached its INTERIM seek frame, and based on it I now can decide to unpause after AutoAdjustInput().
AutoAdjustInput() still has a tiny problem of not advancing the first removed was-lag frame when you go step by step, because it can only go one direction at a time, and it goes up as explained above, but we need it to also go down. this can be forced if we allow tastudio to seek to the CURRENT frame by loading its state and reemulating it, but I don't know if this will cause issues, and nobody is stepping through autoadjusted frames anyway, since it's meant to happen quickly and on autorestore
TODO: removing consecutive was-lag frames seems to remove them ALL!!!
inb4 all this causes other wild bugs...
remove all consecutive was-lag frames in batch like taseditor does. current frame has no lag so they will all have to go anyway
still need to fix wrong pausing
move tasview context menu to the other side of the cursor if it's not fitting into the screen fully. of course if there's not enough space there EITHER, the user must do something about it, meanwhile best we can do is clamp location to 0. not using MouseEventArgs because we need absolute on-screen coords
other tastudio menus are small so probably fine?
- the only dialog that's meant to be used during emulation is undo history, so let it spawn at its old location in the middle of the screen. user will move it regardless of where we spawn it
- positioning of input prompts will happen later
- all config/metadata dialogs now block tastudio and spawn at the top just like open/save. user is clicking tastudio menu so it's distracting if those dialogs spawn too far away from where they're already looking. and the middle of tastudio can be indefinitely far away because it can easily occupy the entire screen (we have cores with tons of buttons, and some users like to make it as tall as the screen). so while in the middle is """consistent""", it's very annoying if you have to do a lot of consecutive edits (for example to greenzone caps, or colors). overall, top priority of smooth tas workflow, so """consistency""" with unrelated tools made for unrelated tasks is not helpful if it makes the workflow more annoying
- ensuring the dialogs spawn fully on-screen will happen later
removed unused tastudio menus and config params
- branch hover interval was added before I replicated taseditor logic, which makes branch screenshots only appear if you keep the pointer over branch number, and not if you're just moving it around. making it configurable doesn't look useful
- seek cutoff interval (how far target frame has to be to spawn seek progress bar) feels redundant because any onscreen edits won't spawn the thing anyway, and for farther seek points it's not annoying enough to change
- EmptyNewMarkerNotes is not relevant to tastudio because it never copied previous marker notes like taseditor does. currently it does nothing and nobody ever requested it even tho it's in the menu, so there's no evidence anyone ever needed it
- fun fact: I was sure autoadjust input to lag never worked but it mostly does! so it wasn't removed, instead I plan to fix it (it pauses on every added frame). it's also not saved.
- other dialogs use InputRoll so I had to remove those options from them too (tastudio is the only one that they were meant for anyway)
bump undo history to 1k. it may not be needed often but when it is, some actions are not combined (for example it adds every newly emulated frame) so you may end up unable to undo some dumb mistake than was only a few actions ago
use normal control BG color for tastudio color editor. I tried it with dark theme on linux and got no difference. if this causes issues we'll see how to fix it without randomly giving one dialog unique BG color. also removed todo since labels are good now
feos