In addition, they should not activate for a spread move that ends the game.
(This contrasts with Soul-Heart, which will activate until the last target.)
PS wasn't correctly detecting `disabled: 'hidden'` moves as disabled
for the purposes of detecting Struggle activation. This has been fixed.
Thanks to DaWoblefet for unit tests!
Fixes#6620
Not having prefer-const on the JS side makes JS -> TS refactors really
unreadable. This commit just auto-fixes it so we're using
`prefer-const` everywhere.
Previously, format bans/unbans were processed in order of "most specific
to least specific".
So `+Giratina-Origin` trumps `-Giratina` trumps `+Uber` trumps
`-All Pokemon`.
And nonstandard reasons (`Unobtainable`, `Past`, `CAP`, etc) were
slotted between `Uber` and `All Pokemon`.
This has the unfortunate effect that if a base ruleset allows a certain
normally-banned category of Pokémon, you can't use `-All Pokemon` to
whitelist a list of Pokémon.
Moving nonstandard reasons to lowest precedence once again allows
`-All Pokemon` to be used as intended.
Regression caused by: 47b55f96bc
`Battle#dragIn` used to run the `SwitchOut` event among other functions
but were accidentally left off in the refactor. This commit adds them back.
The main bug caused by this is Natural Cure not curing status aliments when
phased out by Roar.
`activeTurns` was previously a horrible hack, used for "first full turn
only" effects like Speed Boost as well as "first move action" effects
like Fake Out.
In addition to being a huge hazard for API users such as OMs, this also
means weird bugs such as Speed Boost not working if you get hit by
Sky Drop on your first turn.
This commit fixes them by splitting these counters into two - an
`activeTurns` counter for Speed Boost, and an `activeMoveActions`
counter for Fake Out.
Move targets are now intended to be +1 +2 +3 for foes. The old syntax
of using 1 2 3 is still supported, but is not recommended.
(The old syntax will still be used in the old client, but the Preact
client will support the new syntax going forward.)
This makes the difference between move number and move target
clearer.
This also fixes the Conversion 2 ambiguity (although for backwards
compatibility, we do still need to special-case it).
Dragon Darts has a variety of mechanics that I don't care to fully
write unit tests for, but this commit should track its in-game
implementation much more closely.
If anyone still has a Dragon Darts interaction that this implementation
handles incorrectly, they should confront me about it.
Fixes#6279
This is a really hacky implementation of Emergency Exit, but Emergency
Exit itself is a huge mess on cart, too.
Our previous implementation:
- activated Emergency Exit at AfterMoveSecondary timing for move damage
- activated Emergency Exit immediately after dealing any other damage
This new one:
- activates Emergency Exit only in three situations:
- right after AfterMoveSecondary timing, for move damage
- right after DamagingHit timing, for DamagingHit residual damage
(Rough Skin, Iron Barbs, Rocky Helmet)
- right after the switch update, for switch-hazard residual damage
(Stealth Rock, Spikes)
- does not otherwise activate (so Substitute, Hail, Toxic, etc no
longer activate Emergency Exit)
This should much accurately simulate Emergency Exit behavior, including
most famously timing it after healing berries after hazards, as
documented in:
https://www.smogon.com/forums/threads/pokemon-sun-moon-battle-mechanics-research.3586701/#post-7075354Fixes#6309
By adding a `getGame` function of type:
```
// null is returned if the gameids don't match
// or the game doesn't exist
getGame<T extends RoomGame>(constructor: new (...args: any[]) => T) => T | null
```
(Credits @urkerab and @whalemer for the function signature.)
It allows refactoring previous code of:
```
if (room.game && room.game.gameid !== 'hangman') return;
const game = room.game as Hangman;
```
to:
```
const game = room.getGame(Hangman);
if (!game) return;
```
This has a couple of advantages:
- TypeScript will throw an error if the if condition is not present.
- In the new code, the template must extends `RoomGame` and be assignable to the same ID, so it's 100% typesafe
U-turn is the only switch effect that occurs after Pickpocket resolves during an attack segment (even though the switch message and animation appear to happen before it)
- Healing Wish is not consumed if the Pokemon switching in
has full health and no status condition.
- Being statused with full health will consume the Healing Wish.
- It does not trigger on damage after switching in.
- The effect will be consumed if an injured Pokemon is swapped
into a slot with an active Healing Wish by Ally Switch.
Check whether the move is a max move. If it isn't, then resolve the choice as usual. Otherwise, do not add the move to the queue, and let the runDynamax mechanic handle move resolution.
Added regression test to check for focus punch message in both dynamaxed and regular use
Custom rules can now use whitelists instead of blacklists.
Whitelist rules look like:
`- all moves, + Metronome`
(Metronome Battle does not use this ruleset currently, because its
scripted check displays a better error message than a whitelist rule,
which can't actually display the whitelist as of the current
implementation.)
It turns out not creating a format on-the-fly breaks deserialization.
We now just use the same custom-rule system that tournaments use.
Some hacks are currently necessary (many tests assume that we're
playing in Anything Goes rather than Custom Game) but we'll work them
out in time.
`Dex.installFormat` has been deprecated and removed. Formats are now
directly created and cached by our unit test framework. This should
lead to fewer weird bugs.
In Generation 8, a Pokemon's speed updates dynamically. Meaning that it can move sooner than expected if its speed is modified by something such as tailwind or swift swim. This is different from past generations where the Pokemon's updated speed would only take effect the next turn.
This is until pre gets online and we can figure out what's wrong with
it. The bug doesn't seem to be on our end, and I can't figure out why
it's generating Pokemon with species "".
- Centiskorch was misspelled in formats-data, causing a crash in the
egg validator
- A few validation errors were due to Gen 6 not inheriting from Gen 7,
Gen 7 not having a scripts file, and Gen 8 having a gen of 7
- Intimidate (Gen 7) wasn't inheriting from Intimidate (Gen 8), giving
it no name, causing a few validation errors
(Technically not a build error, but I also added Keen Eye to the list
of Intimidate immunities, as reported by SadisticMystic.)
- A lot of tests relied on Teleport always failing; these have been
switched to Gen 7 or swapped Teleport for Celebrate
- Inverse Mod suddenly stopped working; its implementation was a huge
hack and I can't figure out what went wrong, so I've switched it to
using the same system the other mod tests use. It's still a huge
hack, but I don't have the free time to fix it right now.
PS already tracks which evolution level an egg move or event pokemon
learns a move from, but apparently this was not tracked for DW moves.
This fixes a compatibility interaction between DW Marill and Azurill
moves.
This happened because the father checker was simplistic enough
that it didn't think to check Duskull for Pain Split, since
Dusclops learns it in future gens.
- Tyranitar egg moves are fixed
- Staravia egg moves are now properly supported
(9742ecf62f was a hack that caused the Tyranitar problem)
- Some Pokemon evolve by having a move in their learnset (like Piloswine
with Ancient Power). These can only carry three other moves from their
prevo, because the fourth move must be the evo move. This restriction
doesn't apply to gen 6+ eggs, which can get around the restriction with
the relearner.
Fixes#2287
(Except Shiftry and friends, but those are already tracked in
`rulesets:obtainablemoves`.)
Previously, we split gen 2-5 egg move validation into two phases,
`checkLearnset` where we searched for a valid father, and
`reconcileLearnset` where we made sure the father could learn the move
combination.
Egg move validation has now been completely moved out of these
functions and into `validateSource`, which calls `findEggMoveFathers`.
The new algorithm no longer requires `C` moves to be hardcoded into
`learnsets.js`, now doing a more thorough check validation for the
father's move combination. This should be slower than before, but
net performance should be massively improved due to two other
optimizations in this refactor:
- We no longer do any father-searching if the moveset can be obtained
any other way - in particular, this means no more father validation
in gen 6+ where all egg move combinations are legal anyway.
- We only check fully-evolved pokemon as fathers (because anything a
prevo can learn, its evos can learn, too - yes, we remember to make
exceptions for Salazzle, Combee, and future-gen evos)
In addition, `/learn` should now provide significantly better
information for egg move breeding, since it uses a more thorough check
instead of the validator's short-circuiting the moment it finds a valid
father.
This also improves some baby-only move validation, specifically fixing
the issue with Seismic Toss Charm Chansey.
- Repealing rules now always works, regardless of rule order
(Fixes AAA validation)
- Fix a check for egg move hidden ability validation
(Fixes a Gen 4 Dragon Dance Charizard set)
- Always ban AG when banning Uber
(Fixes allowing Rayquaza-Mega in lower tiers)
- Fix ability validation in Let's Go
- Fix valid-move-combo dexsearch
In most other similar systems, like TeamValidator, we use `thing.dex` instead of having it extend `ModdedDex`. Battle has always extended `ModdedDex`, though. This changes Battle to match the others.
This should fix an issue with `Battle.data` not being cached.
This also frees up Battle to extend ObjectReadWriteStream<string> in a future update.
* Refactor validator
This is a major refactor intended to make the default rules easier to
understand, and also easier for OMs to bypass.
Removed rules:
- `Pokemon`: This is now `-Nonexistent`. Its previous name was intended
to be interpreted as "simulate the Pokémon games exactly, and only
allow what they allow". The new name should make it clearer that it
mainly bans CAPs and other nonexistent Pokémon and functionality.
- `-Illegal`: This is now `Obtainable` (see below).
- `Allow CAP`: This is now `+CAP`. Instead of having a hardcoded rule,
OMs can now be manually whitelist any pokemon/item/etc or group of
them, overriding `-Nonexistent`.
- `Ignore Illegal Abilities`: This is now `!Obtainable Abilities` (see
below).
`Obtainable` was previously `-Illegal`, and does the same thing: Makes
sure you have a regular Pokémon game with only Pokémon that can be
obtained without hacking.
But instead of being a ban, it's now a rule that does nothing by
itself, except contain more rules:
- `Obtainable Moves`
- `Obtainable Abilities`
- `Obtainable Formes`
- `Obtainable Misc`
- `-Nonexistent`
- `-Unreleased`
This allows OMs to piecemeal repeal and unban any of these individual
rules, instead of the previous approach of unbanning them all and
manually reimplementing every single validation you wanted to keep.
* Refactor PokemonSources into a class
This mostly just makes a lot of the weirder checks in the validator
substantially more readable.
This also renames `lsetData` to `setSources`, which should also help
readability.
* Validate Bottle Cap HP types
Fixes an issue reported here:
https://github.com/Zarel/Pokemon-Showdown/issues/5742#issuecomment-533850288
* Fix several move validation issues
Fixes#5742
We have a new MoveSource type: R for Restricted. R moves work like
level-up/tutor/TM moves, except you're limited to one R move.
- Shedinja move stolen from Ninjask in Gen 3-4 are now R moves instead
of event moves. This allows them to coexist with Nincada egg moves.
- Necrozma-DW/DM now inherit moves/events from Necrozma (like Rotom,
but with event validation). This allows them to be shiny.
- Pokemon can now get egg moves from their own evolutions. This fixes
some Tyrogue, Charmander, and Treecko sets mentioned in #5742
- Some more C moves were added, fixing some Hitmontop and Chatot sets
mentioned in #5742
* Improve ability/move compatibility validator
`import =` and `export =` are really only intended for backwards
compatibility with CommonJS. While I really don't like the new module
system TC39 has designed for us, it's what we should be using, and
consistency is important.
- The big change here is that player.userid can now be empty. You can
now fit state into RoomGamePlayer subclasses even when there are no
users associated with them.
- `game.players` has been introduced as the new canonical list of
players, including userless players. The old `game.players` has been
renamed `game.playerTable`, for clarity.
- `game.addPlayer` now returns the added player
All existing RoomGames have been updated for the new API, and
RoomBattle is now officially a RoomGame subclass.
Tournaments was also massively refactored to be properly updated for
the old API, since that never happened, and should now be a lot more
readable.
With `-i` in tests/mocha.opts, `npx mocha -g Foo` runs everything
*except* the tests you actually wanted to run (the `-g` overrides
the default `-g`, but unless you specify `--no-invert`, nothing will
override the default `-i`).
Having `test` not just be `mocha` is regrettable, but breaking `npx
mocha -g` or requiring users who want to filter remember to also
pass `--no-invert` is not a good tradeoff.
- Refactor dev-tools/harness.js and dev-tools/smokes.js to separate
out the script/CLI code and implementation code into separate
files.
- Rename 'smoke' to 'exhaustive' ('multi' mode can also be used
for "smoke testing") to better describe its behavior.
- Rewrite the runners in Typescript for type safety.
- Refactor common build utilities into dev-tools/build.js and
introduce the notion of a 'full' build analogous to 'full' tests.
See #5415 for context - having these methods is unsafe and leads
to crashes because not all ActiveMoves are created through the
Data.ActiveMove constructor. Instead of Pokemon, these could
alternatively be static methods on ActiveMove (or the ActiveMove
class could be completely abolished and reverted back to an
interface), but #5415 will deal with ActiveMove long term, this just
fixes the crashes.
Rollforward of 7a20245 which retains the `hasOwnProperty` checks.
Also changes the method to call `toId` earlier and use the id
as the key to the cache to ensure 'Stealth Rock' and 'stealthrock'
return the same. NOTE: 'move: Stealth Rock' and 'Stealth Rock' will
still continue to return different results.
Previously, using Leppa Berry by itself caused Endless Battle Clause
to consider Pokemon to be stale. However, it is reasonable to use
Leppa Berry without an intent to cause an endless battle - for
instance to increase the PP of a move with a low PP.
Also move mods/ to data/mods/
This makes PS more monorepo-like. The intent is to further separate
the sim and the server code, but without fully committing to splitting
the repository itself.
We now support `./pokemon-showdown start` in addition to
`./pokemon-showdown`. I'm not clear which I want to be the default
yet.
Useful for local testing, ignores matchmaking conditions like "can't
match same player twice in a row" and "can't match same IP", to make
it easier to test ladder stuff.
So the Pursuit unit test is broken in Node 0.11, because PS relies on
the sort function deterministically calling the comparison function,
and due to a recent V8 update, the call order changed.
This was a good thing, anyway, since PS didn't handle 3+-way speed ties
correctly, anyway, and it reminded me to go and fix that. Which I did.
This is as cartridge-accurate as possible, given the description given
to me by V4 in #1157.
Fixes#1157
Back when this code was introduced, the TLDs were relatively short,
with only two outliers: .museum and .travel.
Since then, a generic TLD system was introduced which opened
registration of top level domains. The number of TLDs registered this
way exceeds 1 200, with the longest TLDs having 18 characters.
Parsing longer TLDs for emails shouldn't cause practical issues, as
e-mail syntax is unusual enough. In fact, in the older versions, this
code was buggy as it linked example@example.exa in
example@example.example, not noticing that there should be a word
bounary.
A lot of complexity in a chat formatter link regular expression is to
support links with parentheses in them. This commit introduces a test of
this to avoid accidentally breaking this functionality.
PS is inconsistent about which parts do or don't support partial
choices.
This change makes it so battle.choose no longer supports partial
choices, while side.choose still does.
This fixes the bug where making a partial choice caused a battle
freeze.
So if a Prankster-mon chooses in two successive turns an attacking move and an Status move,
and this Pokémon is encored into the attacking move before the Status move proceeds,
the attacking move will be used and it won't affect a Dark-type target.
Often, you just need a random item in an array. Throughout Pokemon
Showdown's code, there are many instances of the following pattern:
let randomThing = things[this.random(things.length)];
Make this code easier to read by factoring the indexing into the
PRNG#sample function:
let randomThing = this.sample(things);
Run the following sed script to refactor lots of code to use sample:
s/\([a-zA-Z0-9.]\{1,\}\)\[this\.random(\1\.length)\]/this.sample(\1)/
This commit should not change behaviour. In particular, PRNG#next is
called the same number of times with the same number of parameter as
before this commit, and PRNG#next's results are interpreted in the same
way as before this commit.
Often, you just need a random boolean. Throughout Pokemon Showdown's
code, there are many creative ways of requesting random booleans. For
example:
if (this.random(10) < 3) {
if (this.isWeather(['sunnyday', 'desolateland']) || this.random(2) === 0) {
let shiny = !this.random(1024);
if (uberCount > 1 && this.random(5) >= 1) continue;
if (!this.random(3)) ability = ability1.name;
} else if ((ability === 'Iron Barbs' || ability === 'Rough Skin') && this.random(2)) {
if (typeof secondary.chance === 'undefined' || this.random(256) <= effectChance) {
if (accuracy !== true && this.random(256) > accuracy) {
Enable these methods to converge by introducing the PRNG#randomChance
function. It accepts a probability and returns true with that
probability.
Run the following sed script to refactor many common patterns to use
randomChance:
s/this\.random(\([0-9]\{1,\}\)) >= \([0-9]\{1,\}\)/!this.randomChance(\2, \1)/g
s/this\.random(\([0-9]\{1,\}\)) < \([0-9]\{1,\}\)/this.randomChance(\2, \1)/g
s/this\.random(\([0-9]\{1,\}\)) === 0/this.randomChance(1, \1)/g
s/!this\.random(\([0-9]\{1,\}\))/this.randomChance(1, \1)/g
The sed script takes advantage of the following properties:
random(x) < y is equivalent to randomChance(y, x)
random(x) >= y is equivalent to !(random(x) < y), i.e. !randomChance(y, x)
random(x) === 0 is equivalent to random(x) < 1, i.e. randomChance(1, x)
!random(x) is equivalent to random(x) === 0, i.e. randomChance(1, x)
This commit should not change behaviour. In particular, PRNG#next is
called the same number of times with the same number of parameter as
before this commit, and PRNG#next's results are interpreted in the same
way as before this commit.
Battle#getRelevantEffectsInner performs a lookup for the base species of
every Pokemon with ModdedDex#getEffect, then invokes callbacks. The
lookup is expensive, and callbacks very rare. In fact, there are only
ever two callbacks: one for Arceus (SwitchIn) and one for Silvally
(SwitchIn).
Instead of an expensive ModdedDex#getEffect lookup for the callbacks,
put the callbacks directly on the Pokemon's Template object.
On my machine, this commit speeds up Pokemon Showdown's tests by 20%.
Methodology: With and without this commit, I ran mocha four times with
zsh' 'time' builtin, dropped the first result, and averaged the wall
times:
mocha times before this commit:
18.20s user 0.33s system 118% cpu 15.704 total
17.91s user 0.34s system 118% cpu 15.454 total
18.11s user 0.33s system 118% cpu 15.558 total
mocha times after this commit:
15.58s user 0.33s system 122% cpu 13.028 total
15.32s user 0.33s system 121% cpu 12.890 total
15.56s user 0.32s system 121% cpu 13.068 total
Hardware:
Mid 2012 MacBook Pro
2.6 GHz Intel Core i7
Software:
Node v9.0.0
macOS 10.10.5
I originally thought this would have to be hardcoded, but actually this
can be coded slightly less hardly than expected!
Getting a Blissey with Present + Heal Bell in Gen 2 works like this:
- Teach Smeargle Present + Heal Bell
- Breed Present + Heal Bell into Snubbull
- Chainbreed Present + Heal Bell into Blissey
The main issue is that checking chainbreeding is very hard, so PS
mostly just takes the stance of "chainbreeding multiple moves is
probably impossible; hardcode exceptions".
BUT! BUT!!!!
Instead of hardcoding this exact move combination, we can actually
just hardcode the fact "the first step of chainbreeding is always legal
if the first father is Smeargle". Which I did and it works!
A battle's inputLog is now stored separately from the output log. It's
not an exact log of inputs, but rather just a collection of the inputs
that resulted in the battle: a default choice expands to the choice
that was actually used, and the starting seed is logged whether or not
it was explicitly passed into the battle stream.
Fixes#4348Fixes#3201
This contains a lot of minor refactors, but the main thing that's going
on here is that battle stream writes have been streamlined to be a lot
easier for others to use.
We even support:
./pokemon-showdown simulate-battle
which provides a stdio interface for anyone using any programming
language to simulate a battle.
`Sim.construct` no longer exists. Battles are now constructed directly
with `new Battle()`. Parameters other than formatid are now passed as
`options`.
As far as I can tell, `curly, multi-line, consistent` does everything I
want; there's no reason to keep around a validate-conditionals rule.
Which is probably good, since eslint is deprecating the API for this,
anyway. The nice thing about not relying on deprecated APIs is that now
you can lint PS with `eslint` rather than needing to memorize
command-line switches.
Process Manager is now lib/process-manager.js
It's been entirely rewritten to reflect what I think a process manager
API should look like.
In particular, there are now two Process Managers, QueryProcessManager
and StreamProcessManager.
Pass QueryProcessManager a pure-ish query function (sync or async) that
takes a JSON value and returns a JSON value, and PM.query() will
execute that function in a subprocess, and return a Promise for its
return value.
StreamProcessManager is the same idea: Pass it a function to create an
ObjectReadWriteStream, and PM.createStream() will create a stream in a
subprocess and return a stream connected to it.
Now, seeds are passed as arrays, rather than needing to pass an entire
PRNG object. In addition, they're now passed in the options object,
instead of as a separate argument.
This is done mostly so the Miracle Eye can be rewritten with a custom
seed, which requires fewer turns and should overall be faster. Which
was in turn done because a Miracle Eye timed out on Travis CI earlier.
Overall, the speed increase is pretty negligible, so this is mostly
just about improving the test API.
Rooms now have their logging abstracted into their own file, which also
allows the rest of rooms.js to be simplified by a decent margin.
This is in preparation for using Redis as a backing for scrollback log
storage, which will give us a lot more RAM to work with.
My newest newest plan is actually to locally cache room scrollback, and
only read battle logs from Redis. This will make chatroom-joining
faster.
Ladder is now a subclass of Matchmaker, and all the APIs previously
attached to Matchmaker are now directly available in Ladder.
In addition, all functions that take a formatid are now of the form
`Ladders(formatid).function(other arguments)`
TODO: Reverse polarity; it makes more sense for Matchmaker to be
a subclass of Ladder. Fortunately, this wouldn't involve API changes.
pokemon.moveset is now pokemon.moveSlots, which is at least slightly
clearer about what it's doing (tracking move state, mainly PP).
Mostly, this gives a consistent naming scheme for `move` (a Move
object) vs `moveSlot` (a MoveSlot object).
This also refactors a lot of existing `moveSlot` accesses to be modern,
including using `for...of`.
New functions:
battle.makeChoices([side 1], [side 2]);
Intended to be a replacement for the previous .choose/.commitDecisions
API.
If we can get all the test code on it, we can maybe finally actually
deprecate LEGACY_API_DO_NOT_USE.
It's now used in Healing Wish, where I think it makes a very readable
test.
The client now interprets `rel="noopener"` to mean "check links for
interstice whitelist", and bypasses the interstice if it's not there.
This is is the best approach for [[google links]] to bypass the
interstice, while not whitelisting all of Google because Google
redirects are not generally trustworthy.
This mainly adds support for parenthetical spoilers:
(parenthetical spoiler: these)
They'll be useful for closing any other autoclosing spans in the
future, but currently spoiler is the only one.
This new version supports significantly more client chat formatting
than before, and is more readable.
Also new: the `code` formatting now uses basically the same system as
Markdown, where you can use as many `` ` `` in a row as you want as
your delimiter.
I'm not really sold on "TeamValidatorAsync" as a class name or file
name, so I'm open to suggestions.
The long-term plan is to put TeamValidator in sim/ but this is in three
commits because I really want the right file to retain blame history.
BattleRoom is now GameRoom, a slightly more abstract room type capable
of holding any of a number of possible roomgames. Currently, battles
are the only such roomgame, but any future roomgame could possibly use
it.
Differences between ChatRoom and GameRoom:
- GameRooms start at the left, ChatRooms start at the right
- GameRooms have chat on the right, and a game area on the left, with a
view area at the top left and a controls area at the bottom left
- GameRooms retain all their logs, ChatRooms only retain the latest 100
lines
- GameRooms expire after 40 minutes of inactivity or 10 minutes of zero
online users (configurable in the future)
- GameRooms are guaranteed to have an associated game, while ChatRooms
may or may not
Random team generation scripts are no longer in scripts.js, but instead
in a new file random-teams.js.
The scripts are now also no longer run from inside battles, but in a
new team generator object. This makes it easier for external scripts
to generate random teams by running Dex.generateTeam(format).
- spawning and setting up child processes is now handled by
ProcessManager#constructor rather than needing to be done manually for
each module that uses ProcessManager
- direct references to ProcessManager in ProcessWrapper were removed to
prevent dead ProcessWrapper instances from being leaked when hotpatching
- process-manager.js and verifier.js now use Typescript
The new FS module is an abstraction layer over the built-in fs module.
The main reason it exists is because I need an abstraction layer I can
disable writing from. But that'll be in another commit.
Currently, mine is better because:
- paths are always relative to PS's base directory
- Promises (seriously wtf Node Core what are you thinking)
- PS-style API: FS("foo.txt").write("bar") for easier argument order
- mkdirp
This also increases the minimum supported Node version from v6.0 to
v7.7, because we now use async/await. Sorry for the inconvenience!
This also drops the mock-fs-require-fix dependency
mock-fs-require-fix was always kind of a huge hack. It's no longer
necessary, with an FS API that does everything it used to.
This removes a lot of other hacks from test/main.js, which is nice.
* Rename Gen6 formats to have [Gen 6] in name
* Add Gen 6 format aliases for bots
Those will be removed later when bots will be updated to know
about new format names.
- The before hook in test/main.js has no reason to be handled
asynchronously when nothing it handles is async
- test/simulator/index.js has no reason to exist when mocha.opts
already handles finding files to test
This is a surprisingly minor refactor considering how many files it
touches, but most of this is only renames.
In terms of file renames:
- `tools.js` is now `sim/dex.js`
- `battle-engine.js` is now `sim/index.js` and its three classes are
in `sim/battle.js`, `sim/side.js`, and `sim/pokemon.js`
- `prng.js` is now `sim/prng.js`
In terms of variable renames:
- `Tools` is now `Dex`
- `BattleEngine` is now `Sim`
- `BattleEngine.Battle` is now `Sim.Battle`
- `BattleEngine.BattleSide` is now `Sim.Side`
- `BattleEngine.BattlePokemon` is now `Sim.Pokemon`
Battle will soon be a Node.js Stream, which has an .on() function,
which Battle#on would conflict with.
PS events aren't Node events, so naming a PS event function .on()
was kind of misleading anyway.
Using internal module API to require the Trivia module is insane when it's
very simple to do without it. The trivia room is deleted after the tests
finish, and users are now destroyed properly.
* Matchmaking: move battle logging back to Rooms.global
For now, logging should be dealt by the global room until logging can be
abstracted away from it. This makes it simpler to refactor the logic in
Matchmaker#startBattle to be handled by Rooms.createBattle where it belongs.
* Matchmaking: use Matchmaker#cancelSearch format parametre
Optimizes cancelling searches if the format is known
PS's choice system has now been majorly rewritten!
Battle#parseChoice has been eliminated, and Battle#choose is now a
very lightweight wrapper around the BattleSide#choose* functions, which
now handle validation.
Partial decisions have been mostly removed. You can manually construct
decisions partially with the side.choose* functions, but there's no
other support for them. Partial undo has been removed completely.
Choice tracking has been renamed from side.choiceData to side.choice.
side.choices has been removed and is now autogenerated from side.choice
when needed.
side.choiceData.decisions has been renamed side.choice.actions. In the
future, "decision" is a deprecated term and should be called "action"
wherever it shows up.
side.choiceData.waiting and side.getDecisionsFinished() have been
merged into side.isChoiceDone().
Other values in side.choiceData have either been rendered unnecessary
or renamed to something clearer.
The "skip" and "pass" choices have been merged together. Passes can
still be filled in automatically (so you can just use `/move 1` in
doubles when you have only one Pokémon left).
This abstracts matchmaking logic from the global room away to its own
module, allowing the two to be decoupled from each other entirely with
some refactoring.
Related to #3361
So I've been working on a massive refactor to the choice system for,
like, over a week. It's still not done, but it's found some mistakes
in the tests that should be fixed.
* Sockets: fix Sockets.killWorker not disconnecting connections
* Sockets: fix unit tests
- Fix crash when constructing mock sockets in certain cases
- Properly prevent workers from writing to stdout
- Fix race conditions in workers-related tests that were causing false
positives
* Tests: mock workers now more closely imitate sockets' workers
This helps catch cases where messages are being sent in the wrong order
to the workers, e.g. messages sent to sockets that no longer exist.
This enables battles in tests to reset their RNG to what it originally
was when they were created. A number of tests do this already by
breaking encapsulating and modifying the prng variable directly. This
should fix that.
We also remove createBattleWithPRNG and min/maxSeed properties.
Firstly, the tests that were still using the maxSeed property were
setting it to Battle.seed which has no effect since the PRNG class does
not look at this property any more - so these were no-ops.
After removing this property from tests, maxRollSeed was never used, so
I removed it and renamed minRollSeed to DEFAULT_ROLL_SEED (since it's
constant).
The places that were still using minRollSeed also were no-ops or were
using createBattleWithPRNG. Since every single instance was actually
just using the same code, I removed createBattleWithPRNG and made
createBattle default to giving you a seed with DEFAULT_ROLL_SEED and
removed the minRollSeed property too.
This makes tests much simpler and reduces the usage surface of
TestTools; now, you must define your *own* seed if you're making a PRNG
for a test, or you use one that TestTools gives you; you may not use a
public property that TestTools gives you.
We also remove the code that removes a listener in the Battle Engine.
This was rendered obsolete by f6a7c4b (see more info [in the discussion
on github][disc]
[disc]:
https://github.com/Zarel/Pokemon-Showdown/pull/3272#discussion_r102414795
This removes the 'deterministic test' tools by preventing action at a
distance (namely, preventing the modification of the `init` method in
`Battle` during tests). This action at a distance is incredibly
confusing.
All this action at a distance did was discard any parameters that were
passed to `Battle` that weren't the first three (which was probably a
mistake by the original author) and also hard code `this.seed` and
`this.startingSeed` in `Battle`.
This functionality has now been moved to the `PRNG` class, so instead
users should pass a `PRNG` to `Battle` as the 5th constructor argument.
Users can also pass one as the third argument to `common.createBattle`
or use `common.createBattleWithPRNG` with the PRNG as the first
argument.
The PRNG is just an encapsulation of the pseudo-random algorithm in a
class. It is stateful, so make sure to take a `clone()` of the PRNG if
you want to re-use it.
Terrain seeds weren't activating correctly on the first Surge of two
Surge ability activations turn 1. This fixes it by allowing Update to
run on inactive Pokémon, allowing a Pokémon's item to activate onUpdate
before their own ability activation.
These tests are just sanity checks for sockets.js' exported methods.
Tests for bugs will come later when those for related users.js methods
have been written.
This isn't very easy to reproduce, so I can't properly test any fixes.
Trivia#broadcast writes to debug monitor rather than throwing if called after
the game instance destroys itself.
Previously, if a pokemon mega evolved and used Pursuit, and the foe
switched on the next turn, it would use Pursuit again on the next turn,
in addition to its usual decision.
This is a hack to fix it; a full fix is pending a rewrite of the
decision structure.
