i haven't posted in, like, a month- probably the longest period of time that i've gone without posting since the start of this whole optimization fuckfest.
this post is a little difficult for me to write because i really don't feel like i'm done. over the course of this summer, i've probably spent more time coding than i've spent sleeping ("definitely. not even a contest." ame says to me as i write this) and i still do not remotely feel like i'm done. there is just so much work left and so much more that could be done on all of this that stopping now feels wrong.
but i also, like, can't anymore.
ame and i have kind of taken to calling this little bouts of mine, my Apeshit of the Month . it's basically just rips on the fact that when i usually work on the game i just go nonstop for a long period of time before i burn out and do nothing for a little bit before just going off on another Apeshit of the Month. i think i'm now reaching the end of my energy for that.
originally, this month was just supposed to be a mini-continuation of last month- little bit more optimization in the AI to speed up how long a turn takes to process. it then kinda blew up. some other section of the code would end up catching my attention and i'd end up getting distracted, or i'd see something and it'd annoy me so i'd start working on it- not to mention that 90% of my coding experience has come in the past few months, so i'd occasionally see something i did a few weeks ago and think "wow, this is shit" and then i'd end up redoing it.... and i think i've just reached the point where i've coded my heart out and need to do something else now.
this post is a mix of a catalog of what i've done for the sake of actually making a record of it and also a deep-ish dive into what the code does and how it does it for those of you who are interested in coding things yourselves.
(and likely some angy for those who enjoy that too.)
gonna be a loooooooooong fuckin post.
How many levels in a meech?'
for my opening act, we have something that i actually forgot about until right before i started writing this post. this fix comes courtesy of our dear mr. perrence, who did this way back in june,
which i then immediately forgot about because there's just been so much happening. this came up when i was running a profiler to see what functions were being called during the AI phase of a four-person double battle. four-person doubles are great for these. since you have 3 AI trainers in the fight, as well as multiple targets that each of them need to choose from, it creates a situation where the scripts are basically being stressed-tested to the greatest extent possible during normal gameplay.
so i ran this profile, and saw that there were 244265 calls to the pbGetExpInternal function.
that's a lot of calls!
so let's take a peek at what's going on here.
(pictures with a white background are from essentials v17.2, and pictures with a black background are reborn code, for reference.)
what you are seeing is the calculation for EXP. most of this isn't actually relevant here, since levels under 100 just need to check an exp table. it's also fairly inoccuous- if this function is being called a fuckload of times, the true culprit is the function that is calling it:
HAHA, i've got you NOW, essentials!!!!! just wait until i tell everyone about how BAD you a- wait, no, actually, this function seems fine. the for loop explains why it's getting called so much too, i mean, look at meech:
meech so stronk it's gotta run 100 times for her.
...but then why is it running 244265 time?
and that brings me to the TRUE culprit:
the problem here is not about what the code is doing, but rather, what it is not. and the thing that it is not doing is saving your level.
this means that your level is manually calculated up based on exp (like it says right there) every time it's needed.
this can be fixed by making a class variable:
and just like that: poof. the game has obtained new knowledge. never again will unsuspecting pokmans be level checked more than once.
The Compil - nope, nope, not yet, too scary
uhhhh let me see what else i've got in here...
*blows dust off the code
so all the way back in june again, we noticed that a bulky sprite refresh function would be called every time a sprite's attributes were changed.
here it is now, the privrefresh function:
there it is!
it's 350 lines long, quite big, does a lot of math (you know how much i hate math).
now, i'm not going to sit here and tell you to remove this function and replace it with sharktale.
(pictured here: me, working on the code)
that would be silly.
but there are certainly times when the function doesn't need to be called, like when you're setting a bunch of variables before the sprite is even displayed on-screen. those calls aren't necessary.
which brings me to the setHW_XYZ function.
(pay no mind to the variable names. hopefully no one will throw rocks at me for this.)
this neat little function only runs privrefresh once for five whole variables wow crazy.
this is boring. like, this is definitely an optimization, but...god i just hate graphics. i think we ended up just putting it in the initialization functions and left it at that.
let's get to the good stuff.
grab your hazmat suit, remember the buddy system, and send some prayers to The Exalted Lord Arceus (praise be) and get ready to dive into
(ok. i can do this.)
the compiler is-
the compiler is just...
don't forget about move memory. we'll come back to that. also shout out again to the Fairy of Toothpaste for patching up the battle factory. lad's killin it.
the compiler is just....... terrifying.
i mean, look at this!
those aren't even words!
how does anyone manage to look at this and not immediately have a brain aneurysm??
the compiler is the one place
that no one should go
it is such a terrible place
or, at least,
now? i am beeg cass. i have fought the AI. i have fought the battle system.
i fite the compiler.
last time, on Dragon Blog Z:
we had THIS.
hooooooo i was mad. i was all out here like thats how mad i was.
also it turns out this wasn't the real culprit either.
like, it was a culprit, but arresting this guy would be like giving someone a parking ticket when there's a murderer on the loose.
the true real final ACTUAL culprit times infinity
this rat bastard.
this piece of shit was behind it all.
lemme tell you about what this function does.
it opens a file.
wow. i'm so smart.
it reads the file.
incredible. how do i do it
while this is, of course, a perfectly acceptable way to save and load move memory, it wasn't good enough for me.
there are lots of ways to load and process data. this function here uses a bitstring.
imagine a very, very, very long line of numbers, stretching off into space.
that's what moves.dat looks like.
the actual data that you want from moves.dat is located somewhere in that string. to find it, this function shifts to a certain section of the string (determined based on the move ID), reads the data out piece by piece, and then ends the read.
this isn't an inherently bad way to process memory. it does work, after all. but it's definitely not efficient. the series of function calls, coupled with the frequency at which this function itself is called, means that string processing becomes a lot of what the battle system does in the background.
so i changed it.
feast your eyes.
now this is how to read memory.
moves.rxdata is basically identical to moves.dat, but instead of having data in a bitstring, it has it in an array. not only is this a lot faster to process, but it's also really easy to read.
want the function code? it's in [id]. there ya go. ez.
if, for some reason, you just randomly want the effect of move 482 in some oddball script, you can just grab it with $pkmn_move. simple, speedy, clean.
abilities wasn't off the hook, though.
ok compiler. let's go and have a ch-
nope nope nope im out byeeeeeee
so the compiler as a whole is a little bit too large to post here; while this is a blog post about coding and script changes, i don't really want to blind you with more blind text than my lawyers recommend.
but i also think i spent a few hours staring at this trying to understand what it did, and i don't want anyone to have to go through that.
so i'll take pieces and explain them and we'll go from there.
let's start with this.
this is from essentials 17, so it won't be what our code looks like. i just don't want to get hurt this time.
this section defines how the PBS data gets processed (pokemon.txt in particular for this). it defines: 1.) what section this pertains to, 2.) what byte the data gets written to, and 3.) what type of data gets written.
take, for example, "Color". whatever is entered in the "Color" field in the PBS will be written to byte 6 and processed as type "e". (it also requires PBColors in order to function.)
the problem i have with this process is pretty much the same problem that i had with moves.dat- it's a lot of data processing for very little (if any) gain. the difference here is that patching move compilation was easy.
for moves, you just take this:
and tell it to stop.
"why is it moves.rxdata?" you might be wondering. and, well, idk. i'm not sure that the file extension matters a whole lot for this and it seemed like a simple way to differentiate between the two.
dexdata compilation, on the other hand, is like what if the move compilation was tied into a knot. and written in dothraki. and you had to read it with your chin.
instead of rewriting it, i managed to get away with just repurposing the current code to throw everything into an array rather than a bit string:
pg-13 warningL there is a swear word here. if you find such language offensive, then how did you make it on a fangame for pokemon reborn
(as you can see i was getting rather frustrated with the code.)
this array is all of your pokemon data. you'll notice that there are a handful of nested arrays inside of it- they're placed where they are to match up with the old order that dexdata was stored in. it's super arbitrary.
the order that data gets stored in is determined by the same parameter that determined what byte data was stored in.
color would be the first item, type 1 would be the third item, base stats would be the fifth (and is, itself, an array) and so on. it's not really in a very intuitive order, but that's mostly because i'm not really sure an intuitive order exists for this.
the section that allocates data to specific bytes is a little spread out, so i'm going to skip the explanation there (i didn't really do a whole lot to it and also holy shit this post is massive already) and jump to implementing this.
because it's a fuckin pain.
i'll walk you through it.
this is an example of a call to dexdata. the calls are all structurally similar: dexdata is opened, the bitstring is offset to the data you want, the data is pulled, and dexdata is closed.
pbDexDataOffset is what's really running the show here. the number at the end of the function corresponds to the specific data it needs- in this case, the 8 refers to Type1, which you can confirm by checking the original code.
fgetb and fgetw read a specific number of bytes of data. it reads 1 byte for the type and 2 bytes for height/weight.
with the new system, that code becomes this:
$pkmn_dex is preloaded at game start, so there isn't a check to see if it exists or not. if you somehow manage to get in-game without it preloaded, then, well, i'm impressed! also please don't do that.
you can see that it works pretty much the same as move initialization did earlier. you put in the pokemon, you put in the index of data you want, and boom. you have it.
we are moving off some antiquated data structures one step at a time.
now, for me, patching dexdata was about me looking at how this worked and going "hhhhhhhhhhhh i hate this", thus sparking a strong desire to fix
but this code is also, like, actually bad.
infinite fusion had to weigh adding more pokemon (its main feature) against the game grinding to a halt whenever dipping into dexdata
rejuv can't add gen 8 without compiler edits because the compiler effectively limits the total number of abilities to 256. gen 8 has more than that.
it's kinda bad.
that's enough on the compiler- i've been writing this post for five hours and we're still not even on the ai.
so i spent, like, three days on the compiler?
the rest of the apeshit of the month went in to this fuckin thing.
there is just... so much to cover here.
over the course of the last month, we trimmed the ai from 42,000 lines to just a bit under 25,000.
some of this was just code cleanup
some of this involved offloading work off into individual functions
some of this was to fix some bugs.
the rest..... i don't even remember. it was so long ago.
first let's cover some background stuff.
Importing Hidden Classes aka "i hope you liked talking about the compiler earlier"
during compilation, hidden classes are made that hold the IDs of everything that you've compiled. they're hiding off to the side, in that constants.dat (or rxdata. one of them.) since they're off to the side, though, they aren't always accessible to the main script classes, which occasionally means that direct calls to them will fail. i manually imported them by modifying the compiler to yeet them out entirely instead of adding them to the constants file and then copying them into the main scripts myself.
is this a good way to do it? no. i already know that there's a better way, but i just have really burned out hard on this and don't plan to get to it immediately. but it involves having the scripts import themselves. 's fun.
this one was me. i did this.
i made PBStuff to store large arrays of constants that frequently get called/checked together. this was mostly for the sake of the AI, which tended to have these very long lists of moves that stretched out well beyond the incredibly small rmxp script window, but it also came in handy for making sure that ability-changed moves like role play and entrainment would, like, actually fuckin work correctly.
wow! comments! that explain things!
so this was pretty important for code cleanup for two reasons.
the first is that it makes the code much more readable if you're just calling an array of things rather than just rattling off a massive list of moves/abilities.
the second is that it ended up fixing a lot of issues with role play/etc, which had some abilities missing because those moves are really picky.
in the end, it allows for stuff to go from this:
i do not know for sure if it correctly gets called as an array with this syntax.
which is just so much clearer.
okay. let's get started with the real stuff.
isConst? and hasWorkingAbility
this'll get me some hate mail.
i opened this post by talking about how exp/leveling functions work, and that they were called 60 billion times.
well, these two functions were the next two worst offenders.
(this is from one AI phase.)
so what do these functions do?
this one you've already seen from the Role Play screencap. isConst? takes 3 parameters: an object, the class you want to check, and the value you're checking for. isConst? puts them together like so:
if you're still with me, this far into my gigantic post, then you might see my issue here.
isConst? checks to see if a value exists before it gets that value.
it has a rescue statement if the value does not exist.
so why not just directly get the value?
the rescue statement would catch it the same way.
but then, later on, I realized something.
why does this function exist?
you can see in PBStuff that i have arrays of values that are all called like this: [CLASS]::[VALUE], which is functionally identical to what isConst? does.
and the reason it doesn't work is because it can't directly access the hidden classes.
fixing this is what originally prompted me to import the hidden classes into the scripts; that's what enables the use of the class/value call. nearly every single instance of isConst?(val,mod,constant) can just be replaced with
[val] == [mod]::[constant]
and skip out on several lookups and function calls.
hasWorkingAbility? faces a similar situation:
it's effectively a glorified isConst? call.
the checks that are made here are necessary, but they are also being run a lot.
these calls can be replaced by running a single check at the start of a large function, and then replacing the remainder of the calls with (abilityworks && object.ability == PBAbilities::Ability), which can then be condensed further with some organizing. (i'm also considering just making a class variable, but i only thought of it this morning. work is never done.)
AI Move Memory Utilities
remember this from earlier? the AI move memory system is part of what helps the AI just kick your ass.
it also takes up a lot of space in the AI. i mentioned that there were about 17k lines of code that are gone now. 10k of that was through ai memory functions.
every AI move memory check has two basic components: it first pulls the memory, and then it pulls certain information from it. there's a lot of things that are done with the memory, but the majority of it falls into four categories.
the memory puller is here:
and all of the utility functions are here:
astute readers will notice that there are five functions here. however, we all know that in ruby, the 5th element in an array is index 4. so thus i'm still correct and not just too exhausted to change that number.
this, in combination with the PBStuff arrays allows these massive chunks of code to be condensed into single lines.
imagine that happening hundreds of times and you know what's happened to the AI code.
i'm almost 8 hours into this post.
last big thing and y'all can go home.
so i'm sitting here talking up a big game about the ai. but did you know, that our perfect ai, that even it has flaws? it's true! i was also surprised.
when making decisions, the AI first decides if it should switch, then it decides if it should use an item, and finally it picks a move.
every single step, the AI checks the scores of all of its moves.
pictured above: death.
to fix this, i cut pbChooseMoves in half and put the actual movescore grabbing bit in front.
this is a much simpler change than the others, but it slices a large chuck of the AI processing time off.
ok. we did it. actual dev content over. y'all can go home.
imma end this out on a more personal note.
frequent readers of my posts (thank you! and also: why?) will recall that last time i posted, it consisted of direct screencaps of the dev server where i was Very Angry about code inefficiency.
frequent readers of the comments (why? and also: no seriously why would you do that) will also recall that a Big Fan of essentials showed up to call bullshit on the work i've done over the past few months (despite not really understanding what i do), before calling me a memer rather than a programmer.
this hurt me.
likewise, this post was kind of a chore to write.
that's never happened before.
these posts have always been written in a certain style that i hope conveys a feeling that i am speaking to you, rather than simply putting words on a page that you then read and process. it's a stylistic choice on my part that i understand some people appreciate and some people don't.
writing in an informal, open way, only to have someone- a rival community admin- show up on your devblog to tell you that the work you've put into your game is shit kinda fucks you up.
i've spent a lot of time checking over my shoulder for everything i've written. "oh hey, is this code absolutely perfect?" "if i say this, is someone from relic castle going to tell me that i'm stupid and should just upgrade my version of essentials?" (the latter here is why i've been very careful to only include code that is in essentials 17)
but what is most troubling about this is the fact that i was bashed by someone who claimed to represent essentials for offering optimizations for other devs to use in their own games.
i was mocked for using essentials 15. but insurgence uses 15. deso uses 15. rejuv uses 16. games that have been built on older versions often can't just upgrade to newer ones because it would require reimplementing everything they've already done. meanwhile, the changes between versions are often opaque, with little information to help developers who are stuck on older versions to get any of the benefits of upgrading.
to blame someone who is trying to improve an older version of the engine for being stuck on an older version is not only callous, but it also shows that there is a massive disconnect between the people making tools to make games and the people who actually make the games.
and i should be clear here- i'm not upset with the actual developer of essentials at all. when we talked he seemed pretty open to some of the things that i mentioned we had implemented.
the problem that i see is two-fold.
the first part of it is that there is a community that seems to be entitled to essentials to the point where they will bash people who speak out negatively regarding it. i have spent the better part of three months grinding away at the scripts. i spent more time coding than i've spent sleeping. and i want nothing to do with that place. this is a unique circumstance for me because i'm generally really happy to offer my work to people. i like making these posts where i talk about what we've been doing. i like handing out my work to people because it means that everyone can make higher quality stuff without also needing the technical knowledge to tinker with the engine. it's practically a form of elitism- we have the new version, and that makes us better than you with your old one. i suspect this matters less to newer devs, as they can just pick up the latest version and start working with it, but i feel like the fact that this situation has happened at all should still be concerning.
the second part of it is just that...this is so much work. i have been going non-stop for months and i still don't feel like i'm anywhere close to having the code be in a state where i can consider it to be "done". and this has only included work on parts of the engine; there are still entire sections that i haven't even begun to think about, let alone look at.
my point is that this is all just too much work for any one person to do, and it doesn't make sense for multiple people to end up doing the same work over and over again. it doesn't make sense for me to fix the bugs in my version that are still present in the current version, meanwhile there's another version that's being developed and i sure dunno if it's being fixed in there or not. and then, even if it is being fixed, there's no telling if anyone else would be able to use it in their games since there's not a lot of guidance on what the actual changes are.
it also means that, as a result of everyone doing the same things independently of everyone else, that there's a lot of wasted scripting talent being spent on re-doing things that have already been done. for example: how many gen 8 mods are there in development right now? you've seen my scripts. gen 8 is already there. we helped out with rejuv on developing it. i've assisted deso with optimizations, and am helping rejuv implement some of this as well. when marcello heard that infinite fusion was having slowdowns due to the sheer number of pokemon in their game, i sent the compiler changes to see if that would help solve things. even aside from the major fangames, time and time again there have been fans and community members who will show up out of nowhere with their own fixes for things. i've been getting some additional mkxp support from aeodyn, who just messaged me one day saying "hey you know you could be using this better tilemap, right?" and then fuckin waynolt is out here making the entire modding system for reborn/rejuv, and then fuckin toothpastefairy fixed the goddamn battle factory before he was even on the team!
my point here, is this:
essentials needs to be open source.
thanks for coming to my ted talk i'm going to go do anything else now because i have been typing for nine hours and i can't anymore