Methods chronology

[dy]

1. functional

• next iterator abstracting away walking
• Don't distract with TDD, that bloats up code
• Don't distract to illogical redundant functions

2. purely-functional

• functions correspond to parsing semantics
• str.search for faster navigation
• flattening input is easier to walk/process, but creates 2 conversion steps
• too many regexps, regexps are slow ! dispatching parsing result to varying output buffers ✔ smallest size

2.1 purely-functional-fast

− indexOf race vs regexp search − no caching :/

3. indexof

! normalize input - get rid of newlines ! indexOf and split are lighting-fast − split is slow, even slower with limit ↑ split is still faster than manual indexofs with slice − while, for manual loops are medium (10x slower than natives) − search, match are slower than split − 3×indexOf is ~20 times faster than regex (no matter precreated or not) − recursion is ~100 times slower than simple while loop :( − simple fn call is not that bad as recursion ! flash-forward by one char (faster and anyways you'd walk the same with htm's each-character-matching strategy) ! clusters-associated knowledge. ! merge i, j into a single numeric coordinate (comparable) ! put fields as private unicodes in a single string. ✔ no recursion − field placeholder

4. [htm boosted]()

• adds too much space

5. Iterative reactor

• too much space is dedicated to next in pure functional ✔
• pf can be de-recursioned, nexts replaced with reactor on pointers ✔
• possibly unicode flattening for easier traversal? ✔
• micro-htm walker introduced for parsing tag part (the most efficient) ✔

~ unicode and back conversion creates unnecessary slow down and bloated code, bringing it to 590kb. Ideal structure seems to be a meta-path, walked by the depth of modes (modes are indeed related by semantic depth).

6. Metamodes

• conditions are super-compressable, unlike indexOfs
• Modes are linked by semantic depth in single sequence, always: text, field, text, tag, space, name, space, value, space, name,..., text, ...
• Since we anyways have to maintain unicode/deunicode loops, it's more natural to keep two nested loops.
• there are some reliably best constructs, like if (char === '>') text()

8. Regex

• shorter code

[dy]

Clusters

<tag${'name'}.possibly${multi}#cluster
  param${'name'}cluster=${'value'}c${'l'}uster
  quoted="${'cluster'} ${'with'} ${null} > ${false}"
  route=/${'cluster'}/too
  ${'anonymous'}param
  ...${{ spread: 'cluster' }}>
  And ${'text'} clusters ${'too'} <${'with'}>nested</> tags
</>