Skip to content

stream.md

Latest commit

 

History

History
626 lines (515 loc) · 24.8 KB

stream.md

File metadata and controls

626 lines (515 loc) · 24.8 KB

WIP

Working on validating v1.0 functionality.

TODO

v1.0.x

  • flatten turmite/lang/... into turmite/...
  • RuleConstants shifts are awkward ; see turmite/test.js for a starting point
  • debug initial hex scale bug ; also mobile

v1.1 refactoring

  • refactor glslshader.js and glprogram.js
    • move linking to program; program becomes new GLProgram(gl, shader1, shader2, ...)
    • provide uniform / attrib introspection from shader; shader become new GLSLShader({name, type, source, uniformNames, attribNames})
      • with name extraction done at build time by glsl-loader.js using the parser we added for minification
  • hextiletree.js refactor to linear(-ish) form
    • either on a Map<{key, mask}, tile> or on Array<tile> sorted by Array<{key, mask}>
    • probably a good time to move to maker form
    • once there's only need for a single pool (no more nodes), flatten it into hextile.js, dropping pool.js
  • revamp coord.js
    • push conversions out to callers
    • maybe drop the type field?
    • can we eliminate CubePoint (it only has a couple of users)?
  • switch classes out for maker pattern throughout; initial uplift pass above may do so when scope is small enough to be easily done in passing
  • decompose hexant.js
    • inline much of the specific integration/controller logic into public/index.html
    • factor out a stats component for fps, sps, timing, etc
    • factor out a keymap component, which can later be a basis for a help overlay

v1.2 systemitization

  • refactor World.ents to be properly system oriented, rather than have the world model itself participate in per-ent concerns; rectify how views relate to all that jazz
  • separate tile rendering module
  • separate turmite sim module

v1.3 language

  • use constants.Turn to type extracted turn results out of a Rules table, rather than using raw number types
  • switch away from nearley to a custom recursive descent parser
  • make turmite.ruleHelp interactive
  • feature to specify a fixed state rule
    • e.g. to "lift" an ant
    • e.g. to precede a simplified rule of the form when-c => then-c, then-turn
    • idea: state = 42: ant(...) ; state = 42: c => c + 1, turns(R L)
      • coidea: 
        color = 2*c: s => s + 1, L
color = 2*c+1: s => s + 1, R
color = 16*c+1: _ => F
  • ternary expressions?
    • what would that even mean in an LHS when match?
      • .... probably can work actually, but will need a revamp of the solver/matcher
    • most useful tho for the RHS... but what can it achieve that isn't possible with multiple rules?

v1.4 async

  • asyncify the gap between view and sim
    • view can somehow subscribe to updates every N steps
    • could also pair will with a viewport box, rather than always-global views
  • move sim into a web worker

ui

  • a low level immediate mode context over DOM elements and events
    • key processing
    • text
    • button
    • input.text
    • textarea
  • a high level immediate mode context over request commands and response events
    • {prompt: string, help?: string} ... {response: string}
    • {status: string[]}
  • should be feasible to write other low level layers eventually, e.g. one that would process an event stream and draw into a GL context ( or generate an intermediate buffer of GL commands... )
  • NOTE: the src/prompt.js module is experimenting towards this

server

  • port turmite sim to Go... shard and scale over goroutines
  • maybe allow shard(s) to run locally in a background web worker

META: see also / subsume the old TODO.md file

META: grep for code TODOs and cull/triage them into this stream document

2023-01-16

TODO

  • make ant(...) usable as a rule RHS to fix its whenState / thenState terms

WIP

  • vercel build description isn't picking up the tag since v0.11.1

Done

2023-01-15

  • simplified world/view surface down to just reset() and redraw()
    • world now tracks visited bounding box, which view syncs to at top of redraw, eliminating a potentially expensive and redundant call from world.stepn(n) inner loop
    • simplified various aspects of bounding box logic around OddQBox expansion, which now works correctly when starting from an empty box
    • simplified view color scheme palette build, to happen once at top of redraw, and simply sync to the number of colors / entities; no more entity add/remove/update/delete interface
  • this ended up fixing part of the initial view scale problem, since the viewport had previously been starting out empty, rather than containing the first entity-occupied cell

2023-01-13

  • expanded test coverage to include all sources
  • expand type checking to tests and index.jx
  • picked back up development
    • ran npm audit fix
    • ran npm update
      • fixed new typescript complaint

2022-06-27

  • simplified RAF usage based motivated by typescript global error
  • adapted build script to not fully nuke the build directory, allowing usage of git-worktree to track the gh-pages branch
  • changed keymap to watch click events than mouseup events to easily fix mobile touch event, without needing to care about them directly yet

2022-06-26

  • shipped as v0.11.0

2022-05-27

  • further generalized prompt so that it doesn't require a text box
  • generalized keymaps into actions
    • improving representation of shifted keys and space keys
    • now supporting the escape key to cancel

2022-05-26

  • flattened turmite/lang/... into turmite/... now that there is only the nearley driven parser
  • refactored analyze.transform to support deleting nodes
  • moved turmite code actions into the analyze package

2022-05-25

  • subsumed comments and @directives into turmite language module
    • tried and failed to add comments/@directives to nearley grammar
    • ...then spent some time trying to write a new parser module form scratch, abandoned for now due to frustration with variadic tuple types
    • ...then circled back to the nearley grammar, and was able to add comments and @directives on a second try
  • eliminated legacy walk.dfs routines:
    • dfsPre and collect have already been replace by analyze.transform()
    • flattened remainder into toSpecString and simplified into a recursive generator, no more string building stack ops

2022-05-24

  • reified compile scope management
    • now there's a nestable makeScope() => Scope set-ish of symbols
    • used for block construction
  • added strict "all symbols must be defined" check to value compilation
  • flattened spec node, allowing assigns to interleave rules, and setting up for directives / comments next
    • such interleaving enabled by scope rework, since assigns can happen anywhere more robustly now

2022-05-23

  • refactored language analysis module
    • added suite of grammar node maker functions
    • refactored analysis around generic node tree transformation that does not mutate the input tree
    • evicted all application code into the compile module
      • turn counting and symbol collision check now happen where they did before, but have no potential to mutate the parsed input tree
      • tmp hoisting has been pushed down into a refactored compileRule unit that makes rule scope manipulation more coherent
      • commented out block about bare turns() auto-color-indexing also left for future resurrection
  • refactored language compile module
    • reified block() handling
    • made wrap() and amend() more ergonomic
    • rule compilation more coherent, easier to re-use for ant(...)
  • implemented ant(...) alternate rule form
    • compiles thru all the same rule paths as normal {when, then} form by way of a new analyze.antRule() transform

2022-05-22

  • added explicitly null then rules
    • this allows rules like 0, c % 32 => 1, _, _ that only change the state field, pass on the color and turn fields
  • let prompt {help} be an iterable string body, naturalizing functions like turmiteRuleHelp
  • ditched all WIP work towards a general interactive /command and help system in prompt, will experiment towards that again someday, currently planned for the v1.3 language phase
  • turmite/lang
    • now compiles directly to a Builder function
    • erased the _result temporary, inlining then-comments withing one multi-line bitwise or-shift construct
  • improved turmite test tool
    • added command to run a compiled module
    • fixed rules dump, and elide runs of 0-result rules
  • fixed mixed up World mask constants... wasn't detected before because of the update loop hardcodes...
  • fixed Turmite.TestSpec mask and shift constants
  • fixed symbolic solver to actually work for non-trivial matches like 16 * c - 1
  • added a bi-modal ant text example

2022-05-21

  • finished out @numColors directive

2022-05-20

  • finished out color halver test case and its collatz progeny
    • added verbose value and code logging options to isTurmite() test utility
  • started working on comment / directive syntax for specifying numColors broader than required for a turmite's turn sequence

2022-05-19

  • broke up turmite help routines ahead of making them more elaborate / interactive
  • fixed scripts/ed.js, wasn't including a final newline before, which led to eventual churn once re-added by $editor
  • switched rule then mode to = by default, which is less surprising than |=
  • refactored test harness ahead of adding more involved turmite examples
  • improved turmite rule compilation while trying to add a new "color halver" test, many small bug fixes, and the generated code is now much more readable / natural, including more spec-string fragment comments

2022-05-18

  • further work on prompt ux
    • made prompt semantics be a clean start each round, no carry over state
    • added explicit {title} header support
    • added {command} buttons
    • concatenate all {help} outputs

2022-05-17

  • refactored prompt module over a new Interactor<T> type that is a generator of outputs given inputs; this allows basic responsive help now, and will allow more structural interaction later like action commands
  • improved prompt legibility
    • don't let it take more than 50% of viewport height, make long errors scroll
    • provider clearer ant parse errors that don't leak turmite grammar rules
  • added conversion button for ant => turmites

2022-05-16

  • uplifted all tests and got them passing
    • including tests/hextiletree.js which has been broken since commit 7885fe5 removed OddQHexTile.prototype.grow 6 years ago
  • added coverage testing, got ci script passing
  • wrote a basic turmite test module
    • test ant parse/builder
    • test turmite parse/builder vs equivalent ant
    • TODO: more turmite test coverage, e.g. a multi-state turmite
    • caught and fixed many bugs
  • added easy access to turmite string => ...lines compilation, useful today for testing/debugging, maybe usable tomorrow for ahead-of-time compilation for something like a canned/fixed demo
  • improved prompt usability when dealing with large parse / compile errors

2022-05-14

  • finished off minified build
    • now using full html+css+js minification
    • code generator now drops .min.js alternate forms of .glsl shaders, which then get used by the .min.js build bundle

2022-05-13

  • added asset inlining to index.html build
  • added importmap.json auto generation from package.json
  • added index.html auto re-generation from importmap.json
    • added generation chain dependency support
  • minor fixes in generate
    • rollup command were broken
    • scanLines was wrong at scale, if not yet
    • improved command spawning to be closer to clikit.system

2022-05-12

  • switched back to sparse development
    • flattened public/ into toplevel
    • added generate.js driven vendoring of common.js modules
    • using emerging browser https://wicg.github.io/import-maps support
    • this means that hexant development may only be possible in Chrome currently
  • spent the rest of the day "researching" html inliners
    • spent too much time toying with a token-stack based monad...
  • started over and wrote a pragmatic scripts/build.js that does line based editing configured by build.config.js

2022-05-11

  • dropped pool.js module ahead of schedule, since its surrounding use of static blocks conflicts with terser v5.13.1 not yet supporting that syntax
  • added terser-minifed rollup bundle
  • a bit more work on scripts/generate.js
    • handle per-file build errors better
    • rename all shaders to .vert.glsl and .frag.glsl, helps editors
    • improved tmp file renaming and cleanup
  • dropped explicit 'use strict' from all uplifted .js files
  • split index.js out again, with sourcemaps, hardcoded to use the minified bundle for now

2022-05-10

  • finished of initial scope of scripts/generate.js
    • config structured in a humane way
    • ci check command wired up
  • got rollup + http-server based dev server working

2022-05-09

  • got tsc-driven lint passing
    • ignored shader imports for now, plan to revamp gl src / shader / program handling anyhow
    • ignored the nearley dep
      • tried @types/nearley but it got hung up on all of my destructuring post processors like ([tick, tock]) => {type: "meme", tick, tock} due to post procs getting passed an any[] which may have fewer than N elements...
      • ... but if it did, that'd be a parser library bug, since the grammar production rule guarantees N elements, so nearley's type declarations are insufficient
      • ... this will all be better once we replace nearley with a recursive decent parser that we own
  • dropped unused leftover dependencies: global, domready, and rezult
  • fully ripped out eslint, and integrated tsc as the lint script
  • got vercel deployment working
    • fixed glsl-loader to work under Node 14.x LTS... which is apparently the latest that vercel supports...
    • had to add npm run grammar && to build script, since postinstall hook was not running for some reason
  • started ripping out snowpack, since it's looking like abandonware
    • added go generate inspired approach: offline code generation, with committed artifacts:
      • driver is genkit.js as ran by npm run generate ; may move into scripts/generate.js
      • configured by gen-config.js
      • ported glsl-loader.{cjs => js}
      • runs nearleyc for now

2022-05-08

  • uplifted turmite/test.js, which is actually more of a CLI testing vehicle than an automated test
    • also the only user of the hexer dependency; eliminated usage, dropped dependency
    • much refactored around async line generators
  • set package.json type to modules to that turmite/test.js can be ESM without changing extension
  • bumped tsconfig module to es2022 so that turmite/test.js can use toplevel await
  • added rezult.toPromise(Result) => Promise

2022-05-07

  • finished the git rebake from yesterday
    • ripped out noUncheckedIndexedAccess and noPropertyAccessFromIndexSignature from tsconfig, turned out to be more trouble than worth
  • many minor changes all around ramifying from more type checking, some notes:
    • added rezult.catchErr() and rezult.bind() conveniences
    • hash binds now must provide a default and listener
    • factored domkit.mustQuery() out of prompt.js and hexant.js
  • fixed view tombstone pruning bug
  • added fullscreen toggle on Enter

2022-05-06

  • realized the need to refactor rezult objects to be pure data, dropping the class Result and exporting its toValue() and toCallback() methods directly as result functions
    • this started because error results really want to be Result<never> not Result<null>, and I couldn't find a good way to let the class-based version use T=never
    • however it was already a good idea, since no longer carrying methods allows result values to be spread extended like return {...res, extra: stuff} if desired, without loss of Result<T> covariance
  • started baking down git history into per-module uplift monoliths, eliminating a lot of micro commit churn, and validating against the new stricter tsconfig (re)starting point
    • some misgivings about typescript inability to prove narrower array types when using noUncheckedIndexedAccess have started to crop up; may unring that bell before shipping to main branch

2022-05-05

  • added better error handling to Hexant
    • lockout play after a caught error
    • log error with runtime stats and config dump
  • fixed hashbind
    • parse error handling so that it doesn't process a null value
    • undefined value/default/parse handling
  • ratcheted up tsconfig strictness and target
  • started retconing tsconfig strict es2022 rules and re-staging commits one module at a time, building an rc branch that can merge into main, rather than have dev continue to diverge so long term

2022-05-04

  • fixed hashbind default loading: it had been writing undefined strings back into the cache
    • since boolean value representation is "true if present, false otherwise" this was causing them to flip on by default
  • fixed rle-builder: was getting initial counts and final flush wrong
  • fixed removal of dirty tiles; was causing crash at higher step rates when dirty tiles get compacted
  • chased down an ephemeral rebase-induced bug by auditing the ViewGL typification commit line by line:
    • bug presented as hex cells rendering extremely "sparsely", and seeming to be aligned to square grid positions...
    • turned out to be a mere loss of offset++ in TileBufferer.flushTileBuffer
    • which caused each tile draw to merely (over)draw a single hex... revealing the emergent quad tree structure, which is centrally sparse, but dense toward the outer boundary... and dense around the origin
  • refactored prompt module around an async *interact() loop
  • simplified hash.set() return type, no more callbacks
  • switched to a $element idiom when passing bound DOM nodes

2022-05-03

  • reorganized the TODO section above around version milestones
  • fixed infinite recursion bug in HexTileTreeNode compaction
  • fixed the hashbind module
    • lots of minor refactoring bugs
    • make the user do call load, preferably after all bindings
    • improve initialization / round tripping
    • handling of default values when initial set parses to an error
  • fixed turmite/parse.js handling of error results

2022-05-02

  • debugging dev
    • fixed GLPalette.setColorsRGB data indexing
    • fixed world step logic: needed to add in cube space
  • started minimizing and erasing view object structure towards an eventual refactor to maker form
  • restructured ViewGL.redraw to clarify tile and entity phases
  • factored out setNumbers() et al utilities to clarify data array builds

2022-05-01

  • debugging dev
    • fixed NaN turn bug in newly refactored constants.turnDirs
    • fixed invocation error for performance.now bound to world.now

2022-04-30

  • uplifted turmite/index.js
    • broadened Turmite.reset
    • factored out constants.turnDirs(turn, dir)
    • factored out `world.udpateEnt()
    • finishes the compiled form refactoring started in the lang modules
  • uplifted hexant.js
    • fixed missing World.MaskResultColor constant
    • finished All The Refactoring threads from prior modules

2022-04-29

  • uplifted turmite/lang/parse.js
    • moved evaluation to compile module, making return a typed result object
      • fixed option default and return string type from yesterday's uplift
    • flattened and simplified parse.js module, setting it up well to grow a custom recursive descent parser
  • uplifted turmite/parse.js
    • around new Builder => Built types which will be merged with turmite/lang.Spec during parser unification
  • added a new from convenience to rle-builder.js
  • added explicit Rules table type with docs; documented RuleConstants

2022-04-28

  • uplifted turmite/lang/compile.js
    • completely refactored over iterable-strings paradigm
    • reworked nearly all compileFoo function boundaries
    • greatly clarified symbols in scope
    • tightened down the grammar node field types
    • refactored toSpecString into the iterable-strings paradigm
    • preserved integer bases from parsing to code gen time

2022-04-27

  • continued uplifting turmite/lang/compile.js... not done yet
  • noticed how the old grammar definition locks us into expressions over mixtures of turns and primitive numbers everywhere
    • made a parametric Expr<Literal> type that makes this explicit for now, until we can rework the grammar to prevent turns where the context doesn't call for them

2022-04-26

  • uplifted turmite/lang/analyze.js
    • gnarly logic under `analyze() each() switch-case 'then'
      • looks to be implictly indexing the then.turns name(s) by the first then.color symbol? wat?
  • started reformatting code as I go, starting with the current turmite/lang/... modules undef focus

2022-04-23

  • uplifted turmite/lang/walk.js, standing up a new gramma type tree alongside the nearley parser, for future use by its recursive descent replacement
  • uplifted turmite/lang/tostring.js
    • noted how it would be better without a "generic" DFS walker
    • hardened its cases over the new types, rather than allowing invalid grammar tree node data
  • subsumed turmite/lang/solve.js into .../compile.js, resolving their circular relationship, and deduplicating an array declaration

2022-04-22

  • uplifted turmite/rle-builder.js; greatly refactored it:
    • no more string accumulation, result is an array of term pairs
    • returned surface is now {consume([count, ]sym) => void, finish() => [[count, sym]]}, with sensible "prefixed count parsing" out of symbol strings
  • updated turmite grammar definition
    • subsumed turmite/lang/build.js module, just inline all those post-processing rules
    • fix build script
    • switch to ESM output

2022-04-21

  • uplifted world.js
    • that old entity "system" sure is 🥔🍐💀
  • uplifted view_gl.js
    • dropped obsolete `"'experimental-webgl"' rendering context branch, as typescript definitions don't acknowledge it
    • refactored a lot of drawing code to be more coherent
    • many small changes and fixes throughout
    • ramified usage fix into GLProgram wrt uniform and attr collections
    • ramified usage fix into HexTileTree wrt tiles collection
  • uplifted turmite/constants.js
    • switched to Map()s where it made sense

2022-04-20

  • uplifted pool.js
    • refactored to a makePool(cons) => {alloc, free} for use in static blocks
  • uplifted hextile.js
  • uplifted hextiletree.js
    • lots of subtle changes, fixes, and improvements on the way around weak points revealed by type checking
  • uplifted prompt.js
    • the initial gutennacht pass was insufficient, type checking revealed typos, and other deficiences
  • uplifted rangelist.js
  • uplifted sample.js
  • uplifted tileglbuffer.js
    • subsumed TileGLBuffer, LazyGLBuffer, and GLBuffer out of view_gl.js
  • simplifed old cell width/height calc logic, a holdover from the times when we used to compute 6 vertices per cell

2022-04-19

  • uplifted glpalette.js thru type checking
    • minor change to data array fill to use UInt8Array.fill and .set
    • changed constructor to options pattern, made srgb option an enum
  • uplifted glprogram.js thru type checking
    • eliminated shader load indirection, a problem for our future caller
    • switched attr and uniform collections to Map, dropped attrs array, just use attr.values()
  • uplifted glslshader.js thru type checking
  • fixed glsl-loader.js to generate ESMs with aliased imports, and to have better internal io semantics
  • added build-time minification to glsl-loader.js
  • fixed done module denominator... oops

2022-04-18

  • refactored Hexant world stepping logic and reified its animation loop into while-await form
  • uplifted rezult.js thru type checking
  • uplifted colorgen.js thru type checking
    • now only exports parse(), string conversion is now implicit, and fortunately nothing was using the gens export
  • surveyed remaining module status to sketch a better stream status picture
  • uplifted coord.js thru type checking
    • elminated callable-constructor oddity

2022-04-17

  • further refactored hashbind to take over all codec concerns (base64) from hexant.js, obviating need for Base64 shim dependency

2022-04-16

  • inlined page integrated javascript into its html file
  • added an uplifted copy of rezult, started using it in the hexant.js module
  • added a reworked copy of hashbind, and converted hexant.js to use it
  • fixed raf leftover from yesterday, and added a sprinkle of type hinting to assist

2022-04-15

  • opened back up for the first time in over 5 years, started sketching a plan
  • renamed master branch to main
  • started npm uplift
    • dropped old html-inline based build, as it's not been maintained and has critical transitive vulnerabilities under npm audit
  • switched to snowpack, including a port of my glsl loader
  • excised gutentag
    • classified Prompt and Hexant in passing
    • switching to simpler component wiring along the way