visiter 0.20.0

Build and visualize orbit graphs for discrete iterations under guarded rules.

VisIter

See what a discrete iteration actually does — as a graph.

The simplest case

Integers 1–9. Case: divisible by 3 → divide by 3. Default (everything else) → add 2. Where does each value end up?

#!/usr/bin/env viter
(viter(range(1, 10))
 .case(lambda x: x % 3 == 0, lambda x: x // 3)
 .default(lambda x: x + 2)
 .render())

Save as descent.vit, run with viter descent.vit > out.svg. The #!/usr/bin/env viter shebang also lets you chmod +x descent.vit and execute the file directly.

.render() is the shortcut terminal — it builds the graph, converts to Graphviz, and writes SVG to stdout in one call.

Install

pip install visiter

Graphviz must be available on PATH (brew install graphviz / apt install graphviz).

Optional extras: pip install "visiter[native]" adds the native engine (prebuilt wheel, no toolchain needed), [storage] the columnar .vitgraph format, [validate] the JSON-Schema validator, [analytics] the NetworkX bridge. See Performance & storage below.

Going further

.render() is convenient for the common case. For anything more — cropping, custom colors, side-effects, filters — materialize the Graph explicitly via .build() and keep chaining:

#!/usr/bin/env viter
(viter(range(1, 10))
 .case(lambda x: x % 3 == 0, lambda x: x // 3)
 .default(lambda x: x + 2)
 .build()
 .to_dot(anchor=1, radius=8, direction="backward")
 .render())

Save intermediate results with .tap():

(viter(...).case(...).default(...).build()
 .tap(write(file="graph.json"))
 .to_dot()
 .render(file="out.svg"))

Use NetworkX for graph analysis via .filter():

import networkx as nx
(viter(...).case(...).default(...).build()
 .filter(NxFilter(nx.condensation))
 .to_dot()
 .render())

If-elif-else semantics (first matching case wins) via match=Match.FIRST:

(viter(range(1, 17), match=Match.FIRST)
 .case(lambda x: x % 2 == 0, lambda x: x // 2)
 .case(lambda x: x % 3 == 0, lambda x: x // 3)
 .default(lambda x: x * 5 + 7)
 .render())

Use the Python API directly (outside .vit files):

from visiter import viter

graph = (viter(range(1, 10))
         .case(lambda x: x % 3 == 0, lambda x: x // 3)
         .default(lambda x: x + 2)
         .build())
graph.to_dot().render(file="descent.svg")

Per-call edge labels via OpResult — for when the static label= is not enough and each step's edge should carry its own annotation:

from visiter import OpResult, viter

def odd_step(x):
    increased = 3 * x + 1
    div = (increased & -increased).bit_length() - 1
    return OpResult(increased >> div, label=f"3x+1, ÷2×{div}")

(viter([27])
 .case(lambda x: x % 2 == 0, lambda x: x // 2, label="÷2")
 .default(odd_step, label="3x+1")
 .render())

Performance & storage (optional)

Pure Python with JSON is the default and needs no toolchain. For large state spaces, three optional, additive features kick in — all producing the same graph, byte-for-byte:

• Native engine (engine="auto", the default once present). pip install "visiter[native]" pulls a prebuilt wheel that runs the BFS bookkeeping natively — for every build, bounded or not — while keeping your Python callbacks (or make native to build it locally). Falls back to pure Python when absent.

• Inline Rust callbacks (lang="rust") — for when the callbacks are the bottleneck. .case() takes Rust expression strings (value bound to the name you pass as the required bind=), compiled on the fly with rustc and run natively:

  (viter(10, lang="rust", bind="n")
   .case("n >= 1", "n - 1", label="take 1")
   .case("n >= 2", "n - 2", label="take 2")
   .render())
  

• Columnar storage (pip install "visiter[storage]") — .vitgraph files (graph.to_vitgraph(...) / Graph.from_vitgraph(...)) are ~10–25× smaller than JSON and far faster to load for large graphs.

Details in the manual.

Why VisIter?

Free, scriptable, Graphviz-native orbit-graph rendering for discrete iterations under guarded rules — with cutoff boundaries (bounds, depth limits, render crops) as a first-class visual primitive, not silent truncation.

Full honest comparison against NetworkX, NestGraph (Mathematica), Maude, LoLA, and continuous-dynamics tooling: docs/comparison.md.

Documentation

• docs/tutorial.md — gentle introduction: what problem the tool solves, smallest example, what each piece does, what the dashed arrows mean. Start here.
• docs/manual.md — reference: every parameter, every data field, the rendering model in full, design decisions.
• docs/comparison.md — how VisIter relates to other tools in the ecosystem, and when to pick something else.
• demos/ — runnable .vit examples organized by topic (basics/, rendering/, integration/, applications/, rust/).

License

MIT