Skip to content

Presentation and state

A presentation setting changes only pixels. It never changes the game’s state, the random numbers, the input log, a save, the script trace or the sounds the game asks for (plan 42, decision 312).

The owner’s words for it: desktop and DS “may look different, never play differently” (plan 16; decision 310: “im fine with different graphical detail settings for desktop and ds”). A player on a powerful desktop with every effect on and a player in melonDS play the same game. A recorded session plays the same on both.

Presentation: may differ Game: may not differ
Particle counts, draw distance, frame rate of the 3D field Where the player is, what is in view of the logic, which people and scripts are active
Resolution, anti-aliasing, upscalers, bloom, shadows, water The random numbers drawn, the battle, the rules
The screen’s layout: two screens or one wide canvas Maps, scripts, text, data
Colour-vision filters, less motion, less flashing Saves, the input log

The line is drawn by one question: does it change what happens when the game reads input? Battle animation speed, how fast health bars drain and auto-run all change how many ticks pass before the game reads the next button press, so they change the game. They are therefore game options, saved with the game and identical on every platform, never presentation rows (decision 332). The desktop’s speed-up runs the same ticks faster; it does not skip any.

Every presentation row declares its DS class:

Class Meaning Example
native The DS draws the same idea with its own hardware Anti-aliasing, fog, edge marking
approx A cheaper version that fits the DS Reflections of characters only
off Not drawn on DS; the DS’s options screen hides the row Bloom, resolution, shadow maps

Rows are named by what they do (bloom, reflections, particles), never by a game or a renderer’s internals. A game may set the preset a new player starts at, and cap what a player may choose, in [graphics]:

[graphics]
default = "classic" # the DS's picture
cap = { bloom = "low", shadows = "map" } # the most a player may choose

A creator’s caps keep a game’s art direction. Accessibility rows can never be capped or forbidden (decision 315). Presets are classic (the DS’s picture), high and ultra; auto picks one for the machine.

Presentation settings are in no save and no pack. On desktop they are lines of the user’s configuration file; on DS, a small device file beside the save slots. A save written at ultra is byte for byte the save written at classic.

The game never asks which platform it runs on. A device file with an auto line is a desktop; one without is a fixed device, which starts at the DS’s values (decision 327).

The settings live in their own crate, pl-present. It is a dependency of exactly the crates that draw: pl-game’s view builders, the desktop and DS backends, and the planned GPU renderer. The logic crates (pl-core, pl-data, pl-battle, pl-script, pl-world, pl-ui, pl-systems) never depend on it.

A test reads every Cargo.toml of the workspace and fails if any of those seven crates reaches pl-present, directly or through another crate. A setting then has no way into the game’s state: the code that decides the game cannot even name it.

Inside pl-game, which does both, the settings are passed only to the functions that build a frame, never to the ones that update the game. That is a convention, and a convention needs a proof.

The renderer receives the finished frame and has no path back into the game (see One description of a frame). Weather particles use a hash of their own index, never the game’s random number generator, so their count can change freely.

One recorded session is replayed headless at two settings, and everything the game decided is compared tick by tick:

Compared What it covers
The world’s digest Positions, facing, movement, flags, variables, the people and their scripts, the generator, the clock, every player in co-op
The scripts’ digest Their flags and variables, the machine, the game’s generators
The battle’s digest The whole battle state
The log The script trace
The sounds Every sound and piece of music the game asked for
Every save Byte for byte
The input log, recorded again Byte for byte

The sessions: the demo game’s opening into the rain, across two areas and the stairs between floors, saved from the menu; and the opening through its first battle. They are played at:

  • classic, the DS’s picture;
  • ultra, with the view rows at their maximum: particles ×4, draw distance ×3, a 21:9 screen, the display’s frame rate;
  • a fixed device with the DS’s own rows changed (60 frames a second, reflections off, an afterimage);
  • the accessibility rows on: a colour-vision filter, less motion, less flashing.

All must be equal. Any difference fails the test and names the first tick and the digest that differ. The proof is itself tested: a setting deliberately read where the game updates (a planted leak) must make it fail.

On DS, the present session of cargo xtask test-nds plays the same replay with the DS’s rows changed, on desktop and in melonDS, and its digests must equal the classic run’s.

  • Fairness and trust. No graphics setting can make a battle easier, a roll luckier or a walk faster.
  • One game to test. Every test, replay and comparison with the real games runs at classic; the result holds at every other setting because the proof says so.
  • Co-op and link play. Players at different settings, on different platforms, stay in lockstep, because only inputs cross the network and the same inputs give the same game.
  • Freedom to add effects. Ray tracing on desktop costs the DS nothing and risks nothing; the proof catches a mistake the day it is made.