Sleeping negative delay #2478

Ruzihm · 2025-12-20T08:01:36Z

I wanted to take a shot at continuing the work by @Cyberboss in #1539

After making adjustments to fit the latest codebase, I also made additional changes:

Changed for to accept expressions OR certain statements as its incrementor. (apparently you can have for(var/i=0, i<1, sleep(0)). neat!)
Put back in the background attribute check for putting in the background sleep (which is now an opcode)
Changes proc/delay scheduling to track number of negative sleep attempts per thread/proc to approximate sharing in byond based on some very basic testing (see below)
Changed the YieldOrder unit test to match the behavior of this byond script (based on @Cyberboss 's version ) :

#define MODE_INLINE 0 // spawn
#define MODE_BACKGROUND 1 // set waitfor = FALSE + sleep
#define MODE_RAND 2 // random seeded

#define TestSleep(delay, expect) if(mode == MODE_INLINE || (mode == MODE_RAND && prob(50))){ spawn(##delay) { ExpectOrder(##expect); } } else { BackgroundSleep(##delay, ##expect); }
#define ExpectOrder(n) ASSERT(++counter == ##n)

/world/New()
	world.log << "Inline:"
	TestSequence(MODE_INLINE)

	world.log << "Background:"
	TestSequence(MODE_BACKGROUND)

	rand_seed(22475)
	for(var/i in 1 to 10000)
		world.log << "Rand-[i]:"
		TestSequence(MODE_RAND)


/proc/BackgroundSleep(delay, expect)
	set waitfor = FALSE
	sleep(delay)
	world.log << "Expect: [expect]"
	ExpectOrder(expect)

var/counter
/proc/TestSequence(mode)
	counter = 0
	var/start_tick = world.time

	TestSleep(0, 2)
	ExpectOrder(1)
	sleep(0)
	ExpectOrder(3)

	TestSleep(-1, 4)
	ExpectOrder(5)

	TestSleep(0, 11)
	sleep(-1)
	ExpectOrder(6)

	TestSleep(-1, 7)
	ExpectOrder(8)
	sleep(-1)
	ExpectOrder(9)

	TestSleep(1, 13)
	sleep(-1)
	ExpectOrder(10)
	sleep(0)
	ExpectOrder(12)

	ASSERT(world.time == start_tick)

	sleep(1)
	ExpectOrder(14)

(^ Which breaks in current OD, running as a server)

I based the 10/20 sleep scheduling proc/thread thresholds on this test:

proc/background_proc_a()
	set background = TRUE
	for(var/o = 1 to 200)
		//sleep(-1)
		world.log << "a at [o]"
	world.log << "a done"
	
proc/background_proc_b()
	set background = TRUE
	for(var/o = 1 to 200)
		//sleep(-1)
		world.log << "b at [o]"
		if (o % 5 == 0) 
			background_proc_c()
	world.log << "b done"
	
proc/background_proc_c()
	set background = TRUE
	for(var/o = 1 to 5)
		//sleep(-1)
		world.log << "c at [o]"
	world.log << "c done"


/world/New()
	spawn() background_proc_a()
	spawn() background_proc_b()

Which produces results like this in Byond:

a at 1
a at 2
a at 3
a at 4
a at 5
a at 6
a at 7
a at 8
a at 9
a at 10
b at 1
b at 2
b at 3
b at 4
b at 5
c at 1
c at 2
c at 3
c at 4
c at 5
c done
b at 6
b at 7
b at 8
b at 9
b at 10
c at 1
c at 2
c at 3
c at 4
c at 5
c done
a at 11
a at 12
a at 13
a at 14
a at 15
a at 16
a at 17
a at 18
a at 19
a at 20
b at 11
b at 12
b at 13
b at 14
b at 15
c at 1
c at 2
c at 3
c at 4
c at 5
c done
b at 16
...
b at 195
c at 1
c at 2
c at 3
c at 4
c at 5
c done
b at 196
b at 197
b at 198
b at 199
b at 200
c at 1
c at 2
c at 3
c at 4
c at 5
c done
a done
b done

And with this patch you get:

[INFO] world.log: a at 1
[INFO] world.log: a at 2
[INFO] world.log: a at 3
[INFO] world.log: a at 4
[INFO] world.log: a at 5
[INFO] world.log: a at 6
[INFO] world.log: a at 7
[INFO] world.log: a at 8
[INFO] world.log: a at 9
[INFO] world.log: a at 10
[INFO] world.log: b at 1
[INFO] world.log: b at 2
[INFO] world.log: b at 3
[INFO] world.log: b at 4
[INFO] world.log: b at 5
[INFO] world.log: c at 1
[INFO] world.log: c at 2
[INFO] world.log: c at 3
[INFO] world.log: c at 4
[INFO] world.log: c at 5
[INFO] world.log: c done
[INFO] world.log: b at 6
[INFO] world.log: b at 7
[INFO] world.log: b at 8
[INFO] world.log: b at 9
[INFO] world.log: b at 10
[INFO] world.log: c at 1
[INFO] world.log: c at 2
[INFO] world.log: c at 3
[INFO] world.log: c at 4
[INFO] world.log: c at 5
[INFO] world.log: a at 11
[INFO] world.log: a at 12
[INFO] world.log: a at 13
[INFO] world.log: a at 14
[INFO] world.log: a at 15
[INFO] world.log: a at 16
[INFO] world.log: a at 17
[INFO] world.log: a at 18
[INFO] world.log: a at 19
[INFO] world.log: a at 20
[INFO] world.log: c done
[INFO] world.log: b at 11
[INFO] world.log: b at 12
[INFO] world.log: b at 13
[INFO] world.log: b at 14
[INFO] world.log: b at 15
[INFO] world.log: c at 1
[INFO] world.log: c at 2
[INFO] world.log: c at 3
[INFO] world.log: c at 4
[INFO] world.log: c at 5
[INFO] world.log: c done
[INFO] world.log: b at 16
...
[INFO] world.log: b at 195
[INFO] world.log: c at 1
[INFO] world.log: c at 2
[INFO] world.log: c at 3
[INFO] world.log: c at 4
[INFO] world.log: c at 5
[INFO] world.log: c done
[INFO] world.log: b at 196
[INFO] world.log: b at 197
[INFO] world.log: b at 198
[INFO] world.log: b at 199
[INFO] world.log: b at 200
[INFO] world.log: c at 1
[INFO] world.log: c at 2
[INFO] world.log: c at 3
[INFO] world.log: c at 4
[INFO] world.log: c at 5
[INFO] world.log: c done
[INFO] world.log: b done

Which is a bit different around when the c loop ends the 2nd time in a row, but is overall more similar compared to the version on master:

[INFO] world.log: a at 1
[INFO] world.log: b at 1
[INFO] world.log: a at 2
[INFO] world.log: b at 2
[INFO] world.log: a at 3
[INFO] world.log: b at 3
[INFO] world.log: a at 4
[INFO] world.log: b at 4
[INFO] world.log: a at 5
[INFO] world.log: b at 5
[INFO] world.log: c at 1
[INFO] world.log: a at 6
[INFO] world.log: c at 2
[INFO] world.log: a at 7
[INFO] world.log: c at 3
[INFO] world.log: a at 8
[INFO] world.log: c at 4
[INFO] world.log: a at 9
[INFO] world.log: c at 5
[INFO] world.log: a at 10
[INFO] world.log: c done
[INFO] world.log: a at 11
[INFO] world.log: b at 6
[INFO] world.log: a at 12
[INFO] world.log: b at 7
[INFO] world.log: a at 13
[INFO] world.log: b at 8
[INFO] world.log: a at 14
[INFO] world.log: b at 9
[INFO] world.log: a at 15
[INFO] world.log: b at 10
[INFO] world.log: c at 1
[INFO] world.log: a at 16
[INFO] world.log: c at 2
[INFO] world.log: a at 17
[INFO] world.log: c at 3
[INFO] world.log: a at 18
[INFO] world.log: c at 4
[INFO] world.log: a at 19
[INFO] world.log: c at 5
[INFO] world.log: a at 20
[INFO] world.log: c done
[INFO] world.log: a at 21
[INFO] world.log: b at 11
...
[INFO] world.log: b at 195
[INFO] world.log: c at 1
[INFO] world.log: c at 2
[INFO] world.log: c at 3
[INFO] world.log: c at 4
[INFO] world.log: c at 5
[INFO] world.log: c done
[INFO] world.log: b at 196
[INFO] world.log: b at 197
[INFO] world.log: b at 198
[INFO] world.log: b at 199
[INFO] world.log: b at 200
[INFO] world.log: c at 1
[INFO] world.log: c at 2
[INFO] world.log: c at 3
[INFO] world.log: c at 4
[INFO] world.log: c at 5
[INFO] world.log: c done
[INFO] world.log: b done

I'm not terribly happy with using hashmaps to lookup the thread/proc info by thread/proc id but maybe it's better than storing the negative sleep counts in the procs and thread classes. As for speed, I wasn't able to reproduce the slow startup as described in the OP of #1262 but the idea is to close that issue. Maybe it helps Goon?

- Remove `sleep` from DMStandard. - Two new opcodes Sleep and BackgroundSleep (which is just `sleep -1`, much faster than popping a `float` off for background sleeps). - Create new lightweight async proc state for sleeping. - Compiler/runtime adjustments to handle new opcodes.

…r than `-1.0f`

…g-negative-delay

…ive-delay sleeps are done in the same thread

…-delay

github-advanced-security

InspectCode found more than 20 potential problems in the proposed changes. Check the Files changed tab for more details.

…ntor.

ZeWaka · 2025-12-21T03:22:33Z

Haven't tested, but goon definitely uses a lot of background/negative/0 sleeps etc if anyone needs a actual codebase testbench

DMCompiler/DM/Builders/DMProcBuilder.cs

OpenDreamRuntime/Procs/AsyncProcState.cs

OpenDreamRuntime/Procs/DMOpcodeHandlers.cs

OpenDreamRuntime/Procs/IOpenDreamGameTiming.cs

OpenDreamRuntime/Procs/OpenDreamGameTiming.cs

Ruzihm · 2025-12-25T02:46:23Z

Tested with this code:

proc/background_proc_a()
	set background = TRUE
	var/t = world.timeofday
	var/x = 0
	for(var/o = 1 to 9900000)
		x = o
	var/elapsed = world.timeofday - t
	world.log << "a done after time: [elapsed]"
	return x

proc/background_proc_b()
	set background = TRUE
	var/t = world.timeofday
	var/x = 0
	for(var/o = 1 to 99000)
		x = o
		if (o % 5 == 0)
			background_proc_c()
	var/elapsed = world.timeofday - t
	world.log << "b done after time: [elapsed]"
	return x

proc/background_proc_c()
	set background = TRUE
	var/t = world.timeofday
	var/x = 0
	for(var/o = 1 to 500)
		x = o
	var/elapsed = world.timeofday - t
	return x

/world/New()
	spawn() background_proc_a()
	spawn() background_proc_b()

Byond:

Welcome BYOND! (5.0 Public Version 516.1658)
a done after time: 6
b done after time: 5

OD master:

[INFO] world.log: a done after time: 65
[INFO] world.log: b done after time: 65

OD with PR:

[INFO] world.log: a done after time: 19
[INFO] world.log: b done after time: 19

Cyberboss and others added 22 commits November 26, 2023 21:21

Add YieldOrder tests

0e13edb

Fix yield scheduling

4f0b534

Fix ProcScheduler state poisoning between tests

0911a5e

Up DMTests timeout x10 for slow Windows runners

f0e4f27

Emit constant folded BackgroundSleep opcode on negative values othe…

8d7ebeb

…r than `-1.0f`

Merge master

8c66a1d

Clear ProcScheduler state each test run

1e4dd81

Merge remote-tracking branch 'upstream/master' into 1262-SmartYield

7dce998

Create IOpenDreamGameTiming

38e1ff4

Merge remote-tracking branch 'upstream/master' into 1262-SmartYield

e3cecd2

Correct tick timing for DMTests

dfc4de4

Add a failing test

591eb09

Fix ProcStatus returned from SleepState

cb18a4e

Merge remote-tracking branch 'upstream/master' into 1262-SmartYield

eebf85b

attempt at continuing work from @Cyberboss OpenDreamProject#1539

447db24

Merge remote-tracking branch 'cyberboss/1262-SmartYield' into sleepin…

29dce91

…g-negative-delay

implements pseudo scheduling by doing a same-tick yield when 10 negat…

20e439f

…ive-delay sleeps are done in the same thread

Merge remote-tracking branch 'upstream/master' into sleeping-negative…

16e02b5

…-delay

track threads and procs for negative sleep interruption timing

62e8eb5

add back the background attribute check

6ae008e

update delay comment

b05a0aa

boring-cyborg bot added Compiler Involves the OpenDream compiler Runtime Involves the OpenDream server/runtime labels Dec 20, 2025

github-actions bot added the size/L label Dec 20, 2025

github-advanced-security bot found potential problems Dec 20, 2025

View reviewed changes

Ruzihm marked this pull request as draft December 20, 2025 08:07

adds support for certain statements (so far only sleep) as or increme…

25f4801

…ntor.

github-actions bot added size/XL and removed size/L labels Dec 21, 2025

Ruzihm added 2 commits December 20, 2025 22:57

remove requirement for sleep delay expression

de96c6a

Fixes disassembly view of sleep, remove accidental using

3a636bd

Ruzihm marked this pull request as ready for review December 21, 2025 06:06

lint

881b84f

github-advanced-security bot found potential problems Dec 21, 2025

View reviewed changes

lint

fb391fd

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Sleeping negative delay #2478

Sleeping negative delay #2478

Uh oh!

Ruzihm commented Dec 20, 2025 •

edited

Loading

Uh oh!

github-advanced-security bot left a comment

Uh oh!

ZeWaka commented Dec 21, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Ruzihm commented Dec 25, 2025 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Sleeping negative delay #2478

Are you sure you want to change the base?

Sleeping negative delay #2478

Uh oh!

Conversation

Ruzihm commented Dec 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-advanced-security bot left a comment

Choose a reason for hiding this comment

Uh oh!

ZeWaka commented Dec 21, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Ruzihm commented Dec 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Ruzihm commented Dec 20, 2025 •

edited

Loading

Ruzihm commented Dec 25, 2025 •

edited

Loading