Instrument editor undo

[alederer]

Based roughly on the sample editor's undo feature --

• separate undo/redo stacks per DivInstrument
• DivInstrument refactored to inherit from an underlying plain-old-data DivInstrumentPOD (String name was the only non-POD member)
• undo stack consists of "diffs/patches", which consist of a MemPatch (direct memory diff, only as large as the actual changed region of memory) and a String name (for name changes)
• DivInstrument clears DivInstrumentPOD to zero and then reruns the constructor, ensuring no garbage in uninitialized areas/padding (necessary because we're generating diffs directly from memory)
• diffs are generated by caching a copy of the current DivInstrument in FurnaceGUI and comparing the latest current DivInstrument to the older copy. this happens at the end of drawInsEdit
• (notably, DivInstrument can be modified outside drawInsEdit, e.g. via processDrags/pointMotion, otherwise we could just cache the DivInstrument copy on the stack at the start of drawInsEdit)
• to avoid generating diffs for every single change to memory (which would be hundreds when e.g. drawing macros, dragging sliders, keystrokes, etc.), we delay generating a diff as long as the mouse left/right are down or a text field is open

Potential future issues/work

• the current approach to avoiding generating loads of diffs wouldn't work if we support other input methods like mousewheel or MIDI learn. my thought was to use elapsed process time and adjacency/overlap of memory patches
• haven't looked at memory usage
• haven't tested a whole lot... concerned slightly about DivInstrument fields that reference outside resources like samples (or sample indices?)
• haven't profiled my FixedQueue fix, you said that modulus would be a problem so i just used the same more limited if (idx>=cap) idx-=cap; and figured it was ok, i'm no low-level pro by any means!

[alederer]

thanks for approving the CI workflow @freq-mod --

"fixed" the class-memaccess error on Linux by casting to unsigned char* (it's a reasonable warning but using memset on non-trivial class is intentional here to avoid garbage in the padding)

[alederer]

very low-hanging options:

• recordUndoStepIfChanged can run a memcmp on the data to see if it can skip its byte-by-byte check (presuming a memcmp would be more optimized)
• skip copying into cachedCurIns if we already know it hasn't changed

other possibilities:

• we could only run the comparison on mouse-up and key-up, or something like that, but it makes me a bit nervous to assume that much about possible sources of modification (maybe since i already got bitten assuming all edits were from inside drawInsEdit!). if you feel comfortable with it i can implement it that way though.
• i'd be hesitant to put it on a timer ("every 10 frames" or whatever), since it'd create uneven load (harder to profile, if it really was slow would cause jitters), and it could end up grouping together modifications that should have separate undo states

do we have an established minimum specification to run on? like hopefully we could profile and find a memcmp of 10k or whatever is trivial?

[tildearrow]

very low-hanging options:

• recordUndoStepIfChanged can run a memcmp on the data to see if it can skip its byte-by-byte check (presuming a memcmp would be more optimized)
• skip copying into cachedCurIns if we already know it hasn't changed

other possibilities:

• we could only run the comparison on mouse-up and key-up, or something like that, but it makes me a bit nervous to assume that much about possible sources of modification (maybe since i already got bitten assuming all edits were from inside drawInsEdit!). if you feel comfortable with it i can implement it that way though.
• i'd be hesitant to put it on a timer ("every 10 frames" or whatever), since it'd create uneven load (harder to profile, if it really was slow would cause jitters), and it could end up grouping together modifications that should have separate undo states

do we have an established minimum specification to run on? like hopefully we could profile and find a memcmp of 10k or whatever is trivial?

On mouse up and key up sound good. User input is the only way to change instrument parameters.