AutomataLab v3.0.0

Download the installer for your platform from the Assets below. Desktop auto-updates are supported from v1.0.2 onward.

Added

• Turing Machines (TM). A new deterministic single-tape machine type completes the Chomsky hierarchy: transitions use the read → write, direction (L/R/S) format, the tape is two-way infinite, and acceptance is by halting in an accept state.
• Linear-Bounded Automata (LBA). A bounded Turing machine whose head is confined to the input region; a move past either end halts and rejects, with ⊢/⊣ boundary markers shown on the tape.
• Tape panel with a live head (▲), current-state label, instantaneous description (α q β), auto-scroll, and LBA boundary markers — available from the new Tape tab for TM/LBA machines.
• Reject states (TM/LBA). Mark a state as an explicit halt-and-reject state via right-click; reject states render with a red double-ring and are mutually exclusive with accept states.
• Configurable blank symbol and step limit in the toolbar for TM/LBA, with an infinite-loop guard that halts runaway computations as stuck and surfaces a toast pointing at the adjustable limit.
• Multi-tape Turing machines. Set the toolbar TAPES count (> 1) to give a TM several tapes in parallel: each transition reads/writes one symbol per tape (a → b, R | _ → c, L) and fires only when every tape matches. The input seeds tape 1 and the Tape panel shows one row per tape.
• Global transition (δ) table. A new δ side-panel tab lists every transition grouped by source state and lets you edit them as a table — add, delete, retarget, and relabel inline (FA / PDA / single-tape TM) — with a click-to-locate jump to the matching state or edge on the canvas.
• Data export. Export the transition table (CSV / LaTeX), the execution trace (CSV / JSON), the computation tree (JSON), and the machine definition (JSON) from a new Export dialog in the overflow menu.
• Batch / test-suite runner. A Batch… button runs many input strings at once with optional accept: / reject: expectations, showing a pass/fail table you can export as CSV.
• Declared stack/tape alphabets (Γ). Optional STACK (Γ) (PDA) and TAPE (Γ) (TM/LBA) toolbar fields drive non-blocking validation warnings — e.g. a pushed/written symbol outside Γ, or the blank symbol appearing in the input alphabet Σ.
• Canvas tool palette + visible connection dot. An explicit left-rail palette (select / add state / add transition / add text) makes editing modes discoverable, and states now show a connection dot on hover for drawing transitions.
• “Complete DFA” quick-fix. One click adds a trap state and the missing transitions to turn a partial DFA into a total one.

Changed

• The transition editor, edge labels, and canvas adapt to TM/LBA (a read → write, dir row with an L/R/S selector, edited through the modal like PDAs; multi-tape shows one cluster per tape).
• The Input bar seeds the initial tape for TM/LBA and hides its finite-automaton tape preview (the Tape panel is canonical).
• The Tape panel previews your input live while idle — it mirrors the input box as you type, with the head resting on the start state's first cell, instead of showing an empty placeholder.
• A faded grey arrow under the tape marks the head's last move (←/→), drawn under the cell the head came from, as a history cue for which way it just moved.
• Copy/paste now preserves TM tape moves (write/direction, multi-tape arrays) and reject states alongside the existing PDA stack operations.
• Help and fsm_format.md now document the TM/LBA JSON format, tape/blank/boundary semantics, multi-tape arrays, and include complete 0ⁿ1ⁿ (TM), aⁿbⁿcⁿ (LBA), and 2-tape aⁿbⁿ examples.
• Honest computation trellis. For NFA/ε-NFA the computation view is now labelled a trellis — paths reaching the same state at the same step are merged — and shows a ⇉ₙ chip for how many parents merged, instead of implying a true unmerged tree. NPDA keeps a real branching tree.
• More legible transition labels. Labels sit just off the curve with a thin leader line, and a merged edge shows a compact ×n count that expands on hover/selection.
• Accessibility pass. Visible keyboard focus rings, higher-contrast muted text, larger base fonts, and ARIA roles/labels for the side-panel tabs, canvas tool palette, context menus, and simulation controls.
• A partial DFA is now a warning, not an error. A missing transition rejects (as theory dictates) rather than blocking the run; the validator points you at the new Complete DFA fix.
• Toolbar tidied. Theme, help, update-check, and export moved into a ⋯ overflow menu; the speed control is now a compact read-out; the minimap auto-hides for small machines.

Performance & robustness

• Large inputs no longer freeze the UI. Finite-automaton engines (DFA/NFA/ε-NFA/DPDA) previously rebuilt the entire consumed/remaining input string on every step (and, for NFAs, for every active branch) — O(n²) over a run, with each step getting slower as the head advanced. The engines now surface a bounded window of the input around the head; per-step cost is constant. (Inputs short enough to fit the window are byte-identical.)
• Turing-machine tape is a bounded moving window. A head that travels far (or a huge seeded tape) no longer materialises — or renders — a cell array as wide as the whole tape on every step; the Tape panel shows a fixed span around the head and follows it.
• Bounded buffers. The execution-history log keeps a capped recent window (older steps are summarised as “N earlier steps hidden”), and the computation-tree node buffer is capped on very wide/long nondeterministic runs (the run keeps going; only the visualised tree stops growing).
• Multiple tabs are smoother. Switching tabs now resets the live simulation, so a run can no longer keep stepping a previous tab’s machine in the background, and a previous tab’s (possibly large) tape/tree/history no longer renders against the new one. Canvas node/edge syncing is O(n) instead of O(n²), and per-edge auto-routing is skipped on very large graphs to keep them responsive.
• Step Back is no longer O(n²). Retracing a step replays the engine silently and commits the result in a single store update instead of one update (and one tree rebuild) per replayed step.

Fixed

• A finished run no longer locks the editor. After a machine ran to a verdict, the Delete key, the “+ Add a state” button, the input field, and the hover connection dot could all appear dead (only right-click → Delete worked). Editing is now blocked only while a run is actively in progress; editing a finished run clears the stale result and returns to idle automatically.

Notes

• Fully backward compatible: existing .autolab.json files load unchanged; the new blankSymbol/stepLimit/tapeCount fields, the reads/writes/directions arrays, and the isReject flag are additive and optional. No existing DFA/NFA/ε-NFA/DPDA/NPDA behavior changes.

---

Full changelog: https://github.com/reeshavsinha/AutomataLab/blob/main/CHANGELOG.md