RandomElement for structured RNG #2868
Conversation
mrgriffin
force-pushed
the
rhh-battle-random-element
branch
from
dee9830
to
660ef91
Compare
Converts Tri Attack and Dire Claw to use structured RNG.
mrgriffin
force-pushed
the
rhh-battle-random-element
branch
from
660ef91
to
7b306b6
Compare
| u32 RandomUniformDefault(enum RandomTag tag, u32 lo, u32 hi) | ||
| { | ||
| return lo + (((hi - lo) * Random()) >> 16); | ||
| return lo + (((hi - lo + 1) * Random()) >> 16); |
There was a problem hiding this comment.
This is a very serious bug I discovered when testing RandomElementArrayDefault—it meant that any calls to RandomUniform (outside of tests) were exclusive of hi instead of inclusive.
mrgriffin
added
bug
| @@ -758,6 +766,8 @@ enum { TURN_CLOSED, TURN_OPEN, TURN_CLOSING }; | |||
| #define SKIP_TURN(battler) SkipTurn(__LINE__, battler) | |||
| #define SEND_OUT(battler, partyIndex) SendOut(__LINE__, battler, partyIndex) | |||
|
|
|||
| #define WITH_RNG(tag, value) rng: ((struct TurnRNG) { tag, value }) | |||
There was a problem hiding this comment.
Because of the way the macros work out, the seemingly-unnecessary outer parentheses here are required to keep this whole thing being treated as a single token, which prevents APPEND_TRUE from inserting a TRUE between tag and value.
We could have instead chosen to have a 1-bit field that doesn't do anything in TurnRNG* to absorb this value. This would mean we can write something more like rng: { tag, value } directly in MOVE, which might be nicer.
* Or use a different struct in MoveContext which has that field.
| if (t <= 0) \ | ||
| { \ | ||
| r = FALSE; \ | ||
| } \ | ||
| else if (t >= 100) \ | ||
| { \ | ||
| r = TRUE; \ | ||
| } \ |
There was a problem hiding this comment.
This fixes abilities like Compound Eyes which produce a "percentage" that is out of bounds.
Converts Tri Attack and Dire Claw to use structured RNG via
RandomElement, so that theirKNOWN_FAILINGs can be removed.Also introduces
WITH_RNGwhich allows tests to specify an exact RNG value for any arbitrary RNG tag (instead of just the ones supported byhit:,criticalHit:, andsecondaryEffect:), I suspect this is mostly only useful with tags used inRandomElementcalls.NOTE:
WITH_RNGcurrently only works if the element type is 32-bit or smaller.