Sim: Use a CSPRNG #10806

mia-pi-git · 2025-01-08T18:49:20Z

Yes, it's a new dependency. No, I do not think we should roll our own crypto, nor should we maintain our own crypto.

sim/prng.ts

Slayer95 · 2025-01-08T22:12:30Z

Doesn't this break RNG accuracy in old gens?

sim/prng.ts

DaWoblefet · 2025-01-10T01:53:49Z

I am currently unable to /exportinputlog and start a battle produced by that output with /importinputlog. Here's sample output from /exportinputlog on this branch:

>start {"formatid":"gen9ou","seed":{"type":"Buffer","data":[11,69,137,245,223,243,216,165,17,20,1,110,46,168,60,11,55,93,224,7,215,192,82,15,48,248,8,164,34,225,214,225]}}
>player p1 {"name":"DaWoblefet","avatar":"steven-masters2","team":"Chandelure|||FlashFire|Inferno||252,,,,,|F|,0,,,,|||,,,,,Ghost","rating":0}
>player p2 {"name":"jlfjsadf","avatar":"101","team":"Blissey|||NaturalCure|SoftBoiled||252,,,,,|F|,0,,,,|||,,,,,Normal","rating":0}
>p1 team 1
>p2 team 1
>p1 move inferno
>p2 move softboiled
>forcelose p1

Here's sample output on main:

>start {"formatid":"gen9ou","seed":[63433,64785,32115,55087]}
>player p1 {"name":"fklsdjfas","avatar":"170","team":"Blissey|||NaturalCure|SoftBoiled||252,,,,,|F|,0,,,,|||,,,,,Normal","rating":0}
>player p2 {"name":"DaWoblefet","avatar":"#dawoblefet","team":"Chandelure|||FlashFire|Inferno||252,,,,,|F|,0,,,,|||,,,,,Ghost","rating":0}
>p1 team 1
>p2 team 1
>p1 move softboiled
>p2 move inferno
>forcelose p2

In addition, although I can import an old inputlog, it does not give the same RNG results (on master, this Inferno will miss; on this branch it hits). So we should either have it use the old RNG when using an older seed, or prevent old inputlogs from being imported, because a good chunk of the time you will get a crash or the battle will otherwise be very screwed up from trying to perform actions only possible because of previous RNG.

sim/prng.ts

test/sim/misc/state.js

This is a pretty major refactor which is mostly unrelated to the feature, but it does make the code a lot simpler.

Zarel · 2025-01-10T08:07:41Z

data/mods/gen5/conditions.ts

@@ -32,8 +32,7 @@ export const Conditions: import('../../../sim/dex-conditions').ModdedConditionDa
 			// However, just in case, use 1 if it is undefined.
 			const counter = this.effectState.counter || 1;
 			if (counter >= 256) {
-				// 2^32 - special-cased because Battle.random(n) can't handle n > 2^16 - 1


Holy shit, I traced this back to

9bbb0b8#diff-1876d2faa56ede4eb79712f3c52e03d209fbd5e4b99ad47f5d1b31d3197fbfafR8460

Back then we were using the Gen 3 RNG, which only generated 16-bit numbers. Very different times.

Zarel · 2025-01-10T08:08:34Z

data/mods/gen9ssb/moves.ts

@@ -1248,7 +1248,7 @@ export const Moves: import('../../../sim/dex-moves').ModdedMoveDataTable = {
 				target.clearBoosts();
 				this.add('-clearboost', target);
 				target.addVolatile('protect');
-				const set = Math.floor(Math.random() * 4);


Oof, this is a real bug and pretty bad.

pyuk-bot · 2025-01-10T08:14:42Z

data/mods/gen5/conditions.ts

@@ -32,8 +32,7 @@ export const Conditions: import('../../../sim/dex-conditions').ModdedConditionDa
 			// However, just in case, use 1 if it is undefined.
 			const counter = this.effectState.counter || 1;
 			if (counter >= 256) {
-				// 2^32 - special-cased because Battle.random(n) can't handle n > 2^16 - 1
-				return (this.random() * 4294967296 < 1);
+				return this.random(0x100000000) === 0; // 2^32


Wait why is this 2^32 when counter >= 256? Before that point, the chance is 1 in counter and counter doubles on each success, so the chance of each successive success would be 1/1, 1/2, 1/4, 1/8, 1/16, 1/32, 1/64, 1/128, 1/4294967296…

Obviously it was already like this before this PR I’m just noticing this now

I think this is how it actually works in gen 5? Need a mechanics expert, though.

https://bulbapedia.bulbagarden.net/wiki/Protection#Generation_V

Yep, I was right.

(Also removes the Buffer dependency from PRNG, and slightly improves comments.)

mia-pi-git changed the title ~~PRNG: Use a CSPRNG~~ Sim: Use a CSPRNG Jan 8, 2025

mia-pi-git force-pushed the csprng branch 2 times, most recently from d82a9cc to 5a070aa Compare January 8, 2025 18:57

mia-pi-git marked this pull request as draft January 8, 2025 19:04

mia-pi-git force-pushed the csprng branch from 5a070aa to b9151e5 Compare January 8, 2025 19:05

mia-pi-git marked this pull request as ready for review January 8, 2025 19:10

DaWoblefet self-requested a review January 8, 2025 19:28

Sim: Use a CSPRNG

9d901b5

mia-pi-git force-pushed the csprng branch 2 times, most recently from b9a93e8 to d5a34be Compare January 8, 2025 19:45

Add test

0c2c4a8

mia-pi-git force-pushed the csprng branch from d5a34be to 0c2c4a8 Compare January 8, 2025 19:50

mia-pi-git added 2 commits January 8, 2025 14:12

fix test prng

fed037c

move prng test to others

9a4ed18

larry-the-table-guy reviewed Jan 8, 2025

View reviewed changes

sim/prng.ts Outdated Show resolved Hide resolved

larry-the-table-guy reviewed Jan 8, 2025

View reviewed changes

sim/prng.ts Show resolved Hide resolved

mia-pi-git added 2 commits January 8, 2025 14:22

fix slight hack

c9320b3

tf?

789cde0

Karthik99999 reviewed Jan 8, 2025

View reviewed changes

sim/prng.ts Outdated Show resolved Hide resolved

Fuck this

ed3d04a

mia-pi-git force-pushed the csprng branch from c5ef777 to ed3d04a Compare January 8, 2025 21:36

mia-pi-git added 2 commits January 8, 2025 15:40

fucking lol

c866037

fix crap

3b387b3

mia-pi-git force-pushed the csprng branch from d48623a to 3b387b3 Compare January 8, 2025 22:06

Slayer95 reviewed Jan 8, 2025

View reviewed changes

sim/prng.ts Show resolved Hide resolved

mia-pi-git added 3 commits January 8, 2025 17:02

i'm going to kill someone

27344a3

i hate state

06507f4

fix test

9323324

larry-the-table-guy reviewed Jan 9, 2025

View reviewed changes

sim/prng.ts Outdated Show resolved Hide resolved

mia-pi-git added 2 commits January 8, 2025 21:03

Good work genius

eac4cab

typo

808d520

Fix exportinputlog

39983c5

mia-pi-git force-pushed the csprng branch from b9ee0a2 to 39983c5 Compare January 10, 2025 03:11

Zarel reviewed Jan 10, 2025

View reviewed changes

sim/prng.ts Outdated Show resolved Hide resolved

Zarel reviewed Jan 10, 2025

View reviewed changes

test/sim/misc/state.js Outdated Show resolved Hide resolved

Refactor for inputlog backwards compatibility

848b63e

This is a pretty major refactor which is mostly unrelated to the feature, but it does make the code a lot simpler.

Zarel force-pushed the csprng branch from 5617534 to 848b63e Compare January 10, 2025 06:20

Zarel added 2 commits January 10, 2025 02:46

Readability pass

40cbb50

Readability (again)

f05a128

Zarel requested review from HoeenCoder and KrisXV as code owners January 10, 2025 08:03

Zarel reviewed Jan 10, 2025

View reviewed changes

pyuk-bot reviewed Jan 10, 2025

View reviewed changes

Zarel added 4 commits January 10, 2025 13:08

Remove sodium-native dependency

70411e3

Refactor to serialize seeds in hex strings

53ef066

(Also removes the Buffer dependency from PRNG, and slightly improves comments.)

Apparently << is 32-bit signed

59145ba

Readability

5b40442

mia-pi-git merged commit d3e60b3 into smogon:master Jan 11, 2025
1 check passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Sim: Use a CSPRNG #10806

Sim: Use a CSPRNG #10806

mia-pi-git commented Jan 8, 2025

Slayer95 commented Jan 8, 2025

DaWoblefet commented Jan 10, 2025 •

edited

Loading

Zarel Jan 10, 2025

Zarel Jan 10, 2025

pyuk-bot Jan 10, 2025

pyuk-bot Jan 10, 2025

Zarel Jan 10, 2025

Zarel Jan 10, 2025

Sim: Use a CSPRNG #10806

Sim: Use a CSPRNG #10806

Conversation

mia-pi-git commented Jan 8, 2025

Slayer95 commented Jan 8, 2025

DaWoblefet commented Jan 10, 2025 • edited Loading

Zarel Jan 10, 2025

Choose a reason for hiding this comment

Zarel Jan 10, 2025

Choose a reason for hiding this comment

pyuk-bot Jan 10, 2025

Choose a reason for hiding this comment

pyuk-bot Jan 10, 2025

Choose a reason for hiding this comment

Zarel Jan 10, 2025

Choose a reason for hiding this comment

Zarel Jan 10, 2025

Choose a reason for hiding this comment

DaWoblefet commented Jan 10, 2025 •

edited

Loading