screen refresh out of TC interrupt in bees, c++/juce compatibility in bees, block processing, more #254

catfact · 2016-02-11T22:45:53Z

kind of a grab bag of changes, including some experiments that haven't worked out yet, and some unfinished but promising things, and finally a few things that seem useful.

in bees, moved screen refresh out of the TC interrupt (can't believe it was in there.) this seems to make other timed events a little smoother. the tradeoff is that if the event queue is filling up, the screen can get first partially and then completely stuck. i think it's a good trade; at least it's an immediate indicator of what is happening. @rick-monster had some thoughts about changing the way we handle a full event queue, which seemed worth pursuing.
started a JUCE-based graphical bees editor. pretty early stages but some tricky compilation steps and graphics context management is worked out. hoping for contributors, otherwise it will take a while to get finished.
relatedly, all bees headers now have __cplusplus switches, and avr32_sim sources compile again, and will even link with g++.
fix a minor bug with op_gate not pickling/unpickling its inputs.
fix a potential bug where count of graphics ops could underflow and totally screw things up (not sure if it ever happened.)
some more incremental tweaks to block processing on bfin - control changes can continue to be collected during an audio block, and are processed at the end of it.
tried implementing a cycle counter on the bfin, and added SPI protocol code and a bees operator to retreive CPU usage back to the avr32. but it doesn't work; for some reason reading the CYCLES registers totally stalls parameter change handling. very mysterious. would be nice to figure this out.
cleaned out a lot of dead code, comment, and print_dbg cruft. of course there is plenty more still floating around.

… some dead code along the way.

…g some issues. remove tons of dead code. start adding human-readable module doc generation in descriptor step. show baudrate on screen in debug build.

… now properly reports initial param values and doesn't randomly crash when booted from a scene

…d sometimes overrun the event queue on startup.) also lots of dead code and comments.

…atibility checks in bees headers. fix potentiall nasty index-underflow bug on op_gfx.

Conflicts: apps/bees/src/op.h apps/bees/src/param.c apps/bees/src/scene.c

…led op-creation selection

…ing gfx util sources

…for now)

…Refresh event type

ghost · 2016-02-11T23:05:03Z

Hmm - I got interested in the event queue because all the serial-overload bees crash states I've seen seem to spew event queue-full errors, even after the overloading serial comms stimulus is removed. Maybe the change to move screen refresh out of interrupt prevents serial-overload crash already? Will test at home.

One other possibility is that bees simply cannot recover from a totally full event queue. Thinking one way to test this is rig some op or other with an event-queue-flooding payload (also experiment with different types of events for the payload), send the booby-trapped op a single bang, and see if the system subsequently recovers...

catfact · 2016-02-11T23:21:29Z

well, when events are being raised faster than they are processed, eventually nothing will happen until something slows down (a soft timer or peripheral.)

what i was thinking of was your suggestion that the event queue should be emptied instead of saturated. i tried this at once point a long time ago and can't really remember how it worked out - certainly the effects were unpredictable. but it might be more recoverable. i would definitely want to accompany some really obvious UI thing, like immediately flashing the whole screen white or something, so the user is warned that stuff isn't necessarily going to make sense until something is throttled.

ghost · 2016-02-11T23:56:53Z

What you're saying does make a lot of sense. Theoretically I would expect UI to get a bit intermittent & buggy under those overload conditions, but more or less keep on trucking even if some button-presses or screen-redraws get dropped.

However allowing the event queue circular buffer to wrap (effectively completely draining it from full) didn't actually prevent outright crash due to serial overload. Interesting problem - maybe b0761f7 provides a missing piece of the puzzle...

…screen. seems to work pretty well. possible for screen to become unrecoverably corrupted after an interrupted refresh, but timer events continue to be serviced?

ghost · 2016-02-13T18:59:18Z

26c2e47 looks like a bunch of DOS line-endings leaked into files that were previously using UNIX convention. this means I can't merge my changes to param.c into a local copy of catfact/dev . Can we look at that before merging these commits?

catfact · 2016-02-13T19:01:38Z

huh! yes of course... sorry about that

catfact · 2016-02-13T19:08:30Z

ok i blasted the whole tree with dos2unix and i have no problems merging.. does it work for you?

ghost · 2016-02-13T19:38:05Z

that's one way to skin this particular cat I guess - kind of uneasy about it gimme a sec maybe I can squish out the mistake and retain the commit history...

ghost · 2016-02-13T21:01:55Z

can't untangle the mess with rebase magic - but able to merge both my active branches into local copy of your dev (after the dos2unix commit) - just testing if anything's broken now...

EDIT: bfin module emulator build broken after the merge - once I fix that & re-test the serial comms we can merge this...

catfact · 2016-02-13T22:39:27Z

i have no problem closing this request and redoing (26c2e47), the big one that seems to have introduced CRs.

and a great many things from 3-rd party code (libfixmath, fat_io_lib, and juce)

ghost · 2016-02-13T22:52:59Z

ok other than the CR thing all ok to merge. Only 2 trivial changes required to fix emulator build. Serial overload still triggering hard crash (no more so than on the unmerged feature branch) - only difference is we now have the 'white screen of death'. I'm going to try the event-queue-bomb-payload thing see if that also triggers the crash...

catfact · 2016-02-13T23:00:33Z

how weird, i can't get a crash at all now with timers, even with several running at 1ms. i suppose serial interrupts could be coming in at >1khz though. it clearly isn't a good idea to do this though, or expect that the avr32 can keep up with that many interrupts and keep all the timers going.

if a large-throughput transactoin is required, seems mandatory to pause app processing (i.e., disable timer IRQs) until it is done. have you tried that yet?

ghost · 2016-02-13T23:08:49Z

nope haven't tried it - let's have a look around for the right incantation...

(re-do of #254) screen refresh event (with WSOD), beekeep_juce, block process tweaks

catfact added 30 commits October 28, 2014 15:11

fix conflicts from upstream dev merge

79b6e24

resolve conflict in upstream pull

d2af65c

Update README.md

c5a8ad9

Merge branch 'dev'

d366d7e

Merge branch 'dev'

656636c

Merge remote-tracking branch 'upstream/dev' into dev

6d5c075

Merge branch 'dev' of https://github.com/catfact/aleph into dev

9e884ac

move some blackfin init stuff out of main.c. add a cycle counter macro

6047d7c

add cpu-usage-requests to SPI state machine. not really tested. clean…

01dbba4

… some dead code along the way.

fix SPI state machine stuff for CPU usage reporting. more dead code

411dffb

add cycle counter to report CPU usage from bfin. unfortunately causin…

fb7c20c

…g some issues. remove tons of dead code. start adding human-readable module doc generation in descriptor step. show baudrate on screen in debug build.

don't disallow param changes during module block process. seems ok

a4ce531

don't stop interrupts during screen refresh. seems ok

1f154ef

fix incredibly dumb problem with rawsc init that was driving me nuts.…

97db6f5

… now properly reports initial param values and doesn't randomly crash when booted from a scene

remove a lot of print_dbgs (there were so many that debug builds woul…

3a1d743

…d sometimes overrun the event queue on startup.) also lots of dead code and comments.

fix bees JSON schema

52618c2

Merge remote-tracking branch 'upstream/dev' into dev

8fbf7bf

add pickling for op_gate!

047b3a4

add beginnings of juce-based graphical beekeep frontend. add cpp comp…

26c2e47

…atibility checks in bees headers. fix potentiall nasty index-underflow bug on op_gfx.

Merge remote-tracking branch 'origin/dev' into dev

909de49

Conflicts: apps/bees/src/op.h apps/bees/src/param.c apps/bees/src/scene.c

fix op.c to ignore op_cpu, which was removed

86afc7c

beekep_juce: remove osc build intermediates. fix crash bug with cance…

1af83f5

…led op-creation selection

somewhat flawed attempt at screen-independent positioning

12ec05b

fix centering for drag position. actually looks ok now

0c18042

don't track xcode junk

305b328

use child components for op ins and outs

00c6246

beekeep: visual feedback on hover

9d1f845

beekeep: start working on connector component. also, whoops, add miss…

a395749

…ing gfx util sources

Merge branch 'dev' of https://github.com/catfact/aleph into dev

bf8b621

remove op_cpu from the makefile since it is also not in the op list (…

afb044f

…for now)

catfact added 2 commits February 11, 2016 03:58

fix worlds stupidest c/c++ conflict

6940814

bees: move screen refresh out of tc timer interrupt, add kEventScreen…

b0761f7

…Refresh event type

catfact added 2 commits February 11, 2016 20:44

strange, waves compile was broken

c145cf5

allow event queue to wrap after some number of stalls, but flash the …

a2c36b9

…screen. seems to work pretty well. possible for screen to become unrecoverably corrupted after an interrupted refresh, but timer events continue to be serviced?

catfact changed the title ~~Dev~~ screen refresh out of TC interrupt in bees, c++/juce compatibility in bees, block processing, more Feb 13, 2016

ran dos2unix on the whole tree...

f0f8f53

catfact closed this Feb 13, 2016

catfact added a commit that referenced this pull request Feb 14, 2016

Merge pull request #255 from catfact/dev

1304922

(re-do of #254) screen refresh event (with WSOD), beekeep_juce, block process tweaks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

screen refresh out of TC interrupt in bees, c++/juce compatibility in bees, block processing, more #254

screen refresh out of TC interrupt in bees, c++/juce compatibility in bees, block processing, more #254

catfact commented Feb 11, 2016

ghost commented Feb 11, 2016

catfact commented Feb 11, 2016

ghost commented Feb 11, 2016

ghost commented Feb 13, 2016

catfact commented Feb 13, 2016

catfact commented Feb 13, 2016

ghost commented Feb 13, 2016

ghost commented Feb 13, 2016

catfact commented Feb 13, 2016

ghost commented Feb 13, 2016

catfact commented Feb 13, 2016

ghost commented Feb 13, 2016

screen refresh out of TC interrupt in bees, c++/juce compatibility in bees, block processing, more #254

screen refresh out of TC interrupt in bees, c++/juce compatibility in bees, block processing, more #254

Conversation

catfact commented Feb 11, 2016

ghost commented Feb 11, 2016

catfact commented Feb 11, 2016

ghost commented Feb 11, 2016

ghost commented Feb 13, 2016

catfact commented Feb 13, 2016

catfact commented Feb 13, 2016

ghost commented Feb 13, 2016

ghost commented Feb 13, 2016

catfact commented Feb 13, 2016

ghost commented Feb 13, 2016

catfact commented Feb 13, 2016

ghost commented Feb 13, 2016