The executor kept every commanded point on-screen via a per-strategy
BoundsPolicy of abort / clamp / reflect. Measured against the real
strategies, the other two earned nothing: abort truncated a sweep at the
first edge (line on a narrow screen ran only 90 of 250 steps), and clamp
could park the cursor against an edge (a monotonic ramp stalled 162 steps
in a row) -- both counter to the program's whole purpose of keeping the
cursor moving. reflect bounces off the edge and keeps going, and is
already what line/diagonal need in loop mode. arc's declared clamp was
provably dead code (it clamps its own endpoint, so no sample ever leaves
the screen).
Collapse to reflect-only:
- strategies.ts: remove the BoundsPolicy type and the `bounds` field from
the interface and all six strategies. Keep the local clamp() helper --
it's arc's endpoint geometry, not an on-screen policy; docstring says so.
- executor.ts: resolveTarget loses its policy parameter and its null
return and just reflects both axes; delete clampInt; SweepOutcome drops
"aborted"; ExecuteOptions drops `bounds`; remove the Out of bounds log.
- keeper.ts: loopOpts is now { restore: false, loop: true } -- the
reflect override added with loop mode is redundant.
- tests: drop the abort-outcome, clamp, and bounds-override tests; simplify
fixed() to take no policy; add a regression test that a monotonic ramp
past an edge never yields two identical points in a row (the guarantee
that motivated removing clamp).
Behavior is unchanged for every pattern at normal cursor positions
(verified: line's normal sweep is byte-identical). The only differences
are at a screen edge, where motion now bounces instead of stopping. No
config keys, flags, or pattern names changed.
Docs updated to match, including in-code comments, the README strategies
table (Bounds column removed) and verbose description, the sequence
diagram (resolveTarget signature + getPosition/width ordering + a loop-mode
note), and a CHANGELOG Changed entry.
Introduce a continuous "loop" setting so a triggered sweep keeps the
cursor moving until the user moves the mouse (or Ctrl+C), instead of
firing a single sweep.
- strategies.ts: add optional `loopPath` to MovementStrategy; give `line`
and `diagonal` infinite loop generators that pick a direction once and
ramp forever (4px/step). Their finite `path` and declared `bounds` are
unchanged, so single-sweep behavior is identical.
- executor.ts: add ExecuteOptions { restore?, bounds?, loop? }. Omitting
options reproduces the original single-sweep contract exactly.
- keeper.ts: in loop mode, run an infinite loopPath once (stopped only by
interruption) or chain a finite path cycle after cycle; force `reflect`
bounds for every pattern and suppress the between-cycle restore, so
line/diagonal bounce edge-to-edge instead of stopping at the first edge.
- config plumbing: new boolean `loop` through config.default.json,
config.ts, configFile.ts, cli.ts (-l/--loop), and move.ts, mirroring
the existing `verbose` precedence.
- docs: README loop-mode section + usage/validation updates; CHANGELOG
Unreleased entry.
- tests: loopPath generators, executor options (bounds override, loop
selection, restore suppression), config/configFile loop plumbing, and
keeper-level loop behavior (ramps far vs. bounded single-sweep, chained
cycles). 79 pass.
install.sh now installs the newest published tag by default instead of
tracking the master branch, so 'curl ... | sh' installs a real release and
reports its version. The tag is resolved from the Gitea tags API, falling
back to master if the lookup fails. MOVE_VERSION still pins an explicit ref.
Interactive installer: prompt before replacing an existing install or
overwriting an existing config, reading answers from /dev/tty so it works
under 'curl ... | sh'. Falls back to the prior non-interactive contract
when no terminal is available. MOVE_FORCE=1 skips all prompts; new
MOVE_RESEED_CONFIG=1 reseeds the config unattended (old file kept as .bak).
Fix: install.sh no longer wipes a working install before downloading. The
tree is built in a staging dir and swapped into place only once complete,
so a failed download/extract/build leaves the existing install intact.
Mermaid treats ';' as a statement separator, so the label
'lastPos (== start; re-sync)' terminated mid-line and the parser then
expected an arrow (the 'Parse error on line 59' the preview showed).
Removed the semicolon and, defensively, replaced 'Iterable<Point>' with
plain text so the angle brackets aren't interpreted as an HTML tag in the
rendered label.
Audited comments project-wide against the current implementation:
- scripts/install.sh, scripts/uninstall.sh: header curl URLs pointed at a
root-level install.sh/uninstall.sh, but the files live under scripts/ —
the documented command 404'd. Corrected to the scripts/ path (matching
the README and the actual file location).
- src/move.ts: header said it ties "four logic modules" and omitted
editor.ts; the --edit terminal action was also missing from the order of
operations. Both corrected.
- src/config.ts: numeric Config fields are all milliseconds now; dropped the
stale "pixels" unit left over from stepCount/stepSize.
Comments-only (plus two script header lines); tsc clean, 64 tests pass.
The stepCount and stepSize knobs were two controls for one quantity users
actually care about (reach), and the number of steps is an implementation
detail nobody meaningfully tunes. Each pattern has a natural size and
resolution — a jitter is inherently small, an arc a broad curve — so those
now live as constants in each strategy rather than as global config.
- strategies.ts: each pattern defines its own step count and size; MoveContext
drops `config` down to pure geometry (start/width/height/rng), and the
module no longer imports Config at all (dissolving the type-only-import
cycle workaround). line stays byte-for-byte: 250 one-pixel steps.
- executor.ts: executePath takes `config` for pacing (stepDelay); the path
itself needs nothing from it.
- config.ts / cli.ts / move.ts / config.default.json: drop stepCount and
stepSize from the type, seed, validation, resolver, CLI flags (-n, -s),
and help. stepDelay stays as the one pacing lever.
- configFile.ts: tolerate the removed keys instead of rejecting them — every
pre-1.3.0 install seeded stepCount, so a hard "unknown key" failure on
upgrade is avoided. They're ignored with a one-line stderr notice; genuine
unknown keys still error.
The -n/--step-count CLI flag (shipped since 1.0.0) is now an unknown option;
config files degrade gracefully, command lines don't. Stays in the unpushed
1.3.0 release. 64 tests pass.
Turn the hardcoded straight-line sweep into a strategy system behind three
seams so new patterns are easy to add and, for the first time, testable
without nut.js or a real screen:
- src/device.ts: injectable Device seam over nut.js (autoDelayMs lives
here now); the only module that touches the native lib.
- src/strategies.ts: pure per-pattern path generators + registry + lenient
name resolution. Ships line, diagonal, jitter, walk,
arc, figureEight.
- src/executor.ts: single executePath driver owning bounds policy
(abort/clamp/reflect), pacing, interrupt detection, and
restore-on-clean.
keeper.ts's simulateActivity now selects a strategy and delegates to the
executor; the default `line` pattern is byte-for-byte the previous behavior.
New config surface, layered CLI > file > default with strict validation:
- -p/--pattern <name> movement strategy (names matched case/-/_-insensitive)
- -s/--step-size <px> pixels per step; stepCount is now a step *count*
Robustness for the new edge-seeking patterns: interrupt detection compares
against the last commanded (rounded) point with a 2px tolerance, and
clamp/reflect stay a couple pixels off the screen edge, so sub-pixel cursor
placement on scaled/multi-monitor displays isn't misread as user activity.
jitter's radius scales with sweep length so it moves at the default stepSize.
Tests: new suites for strategies, the executor (all bounds policies,
rounding, interrupt, tolerance, pacing), and the keeper loop; config and
configFile suites extended for pattern/stepSize. editor.test.ts moved to
tests/ for consistency. 64 pass.
Notable changes since v1.1.1:
- New -e/--edit flag opens the active config file in $EDITOR.
- Lazy import of keeper.ts so --help and --version skip the nut.js
native load on cold start.
- Various audit cleanups: failRuntime() mirror, named Logger interface,
errors.ts module, autoDelayMs moved out of module-load side effects,
defaultConfigPath guards against unset HOME, noUncheckedIndexedAccess
enabled in tsconfig.
- Added Bun test suite (34 tests across resolveConfig, loadConfigFile,
editConfig).
- scripts/dev-setup.sh made POSIX-portable.
New module src/editor.ts handles the flag end-to-end:
- editorCommand(editor, path) returns the sh -c argv that lets the
shell tokenize multi-word $EDITOR values like 'code --wait'.
Extracted so editor.test.ts can verify construction without
actually launching an editor.
- editConfig(path) checks $EDITOR is set, checks the target file
exists, spawns 'sh -c <editor> "$@" -- <path>' with stdio
inherited, and exits with the editor's status code.
Refuses (CliError -> exit 2) when:
- $EDITOR is unset or empty.
- The target config file doesn't exist. (Same recovery hint as
elsewhere: 'run move once or reinstall'.)
src/cli.ts adds the flag to the parser and printHelp(). src/move.ts
dispatches it after --version and before config-load. Resolution
mirrors the loader: --config <path> wins, else defaultConfigPath().
8 new tests in src/editor.test.ts cover the pure helper and both
refusal paths; the spawn success path is verified via manual
'EDITOR=true move -e' (would otherwise kill the test process).
README Usage block, Configuration section (new Editing subsection),
and Files table all updated to match.
keeper.ts statically imports @nut-tree-fork/nut-js, which dlopens a
sizeable native .node binary. That load dominated cold-start: ~1.2s
for 'move --version' immediately after install, vs ~125ms warm.
For a flag that just prints a string, that's all overhead.
Change: dynamic 'await import("./keeper.ts")' placed after the
--help and --version short-circuits. The dynamic import is wrapped
in try/catch so import-time failures (e.g., missing native binary,
unsupported architecture) route through the same failRuntime() path
as anything thrown by the loop.
Expected cold start for --help and --version drops to ~50-150ms
(Bun + reading a handful of small files, no native module load).
move with no flags pays the same load cost as today.
tsc --noEmit and the test suite remain clean.
The CLI reads its version string from package.json at runtime, so
'move --version' was printing 1.0.0 even after the v1.1.0 tag landed.
Bumping the manifest here so v1.1.1 (carrying the tests/ move and the
TS LSP fix) ships with matching output:
$ move --version
move 1.1.1
Discipline going forward: bump package.json BEFORE creating a tag.
bun:test types live in 'bun-types', which is pulled in transitively by
@types/bun via a /// <reference types="bun-types" /> directive. bunx tsc
finds this via auto-discovery, but VS Code's TS Language Server doesn't
always honor auto-discovered @types/* packages under moduleResolution:
bundler, so it shows a phantom 'Cannot find module bun:test' error in
the editor.
Adding 'types': ['bun'] to compilerOptions makes the inclusion explicit
and matches what 'bun init' generates for new projects. tsc still
passes; the bun test suite still passes; the editor LSP now resolves
bun:test correctly.
src/ now contains only source code; tests live alongside it under
tests/. Bun's test runner discovers '**/*.test.ts' so no test-runner
config change is needed.
Changes:
- tests/config.test.ts (was src/config.test.ts)
- tests/configFile.test.ts (was src/configFile.test.ts)
- Imports updated: ./config.ts -> ../src/config.ts (likewise for
configFile.ts and errors.ts).
- tsconfig.json include adds 'tests/**/*.ts' so tsc type-checks
the test files too.
All 25 tests still pass; tsc clean.
Tighter type-checking: array indexing and dynamic property access now
return T | undefined rather than T. Catches bugs where code assumes a
key exists without checking.
No code changes were needed — the existing modules already either:
- cast through Record<string, unknown> (configFile.ts), where access
is already 'unknown';
- cast values from parseArgs to '... | undefined' (cli.ts), already
nullable; or
- use static, statically-known fields (Config, ConfigOverrides,
DEFAULT_CONFIG).
tsc --noEmit remains clean. The full test suite still passes.
The contributor bootstrap script was bash-only ('#!/usr/bin/env bash',
'set -euo pipefail', '[[ ... == ... ]]') while install.sh and
uninstall.sh are POSIX sh. Consistency lines them up:
- shebang -> '#!/usr/bin/env sh'
- 'set -euo pipefail' -> 'set -eu' (pipefail isn't POSIX; not needed
here either, no risky pipelines in this script)
- '[[ ... == ... ]]' -> '[ ... = ... ]'
Also added a 'bun run test' hint to the post-install workflow note now
that there's a test suite to run.
Two new test files exercise the layered config resolver and the JSON
config-file loader:
src/config.test.ts 25 cases:
- resolveConfig precedence (CLI > file > default) per field
- seconds-to-ms conversion at the resolver boundary
- verbose precedence across all four cells
- defaultConfigPath: XDG honored, HOME fallback, empty-as-unset,
throws CliError when both are missing
src/configFile.test.ts 11 cases:
- valid file -> overrides, with undefined for unspecified keys
- missing explicit path throws
- malformed JSON / non-object root throws with file path
- unknown key / wrong type / non-positive number all throw with
the offending key named
- default path missing -> null (silent default)
Runs via 'bun test' (or 'bun run test', now wired in package.json).
All 25 tests pass on the current codebase.
The two stderr-fatal paths used to be asymmetric: failUser() is a named
helper with 'move:' prefix and a 'Try --help' hint, while the runKeeper
catch was an inline 'console.error("Error:", err)' + process.exit(1).
failRuntime() now sits next to failUser(), so the entry-point reads as
'one of two well-named fatal paths.' It also prefers err.stack when
available, giving better diagnostics for nut.js / Accessibility-permission
failures than the previous formatter.
Same observable behavior on the success path; failure messages on the
runtime path are now more debuggable.
Replaces 'ReturnType<typeof makeLogger>' with a small named interface so
simulateActivity's signature reads as (config, log: Logger) instead of a
type-derivation chain. Also makes it cleaner to substitute a mock logger
if simulateActivity is ever unit-tested.
No behavior change.
The assignment used to run at import time, mutating the shared nut.js
singleton the second anything imported keeper.ts. Moved into the top
of runKeeper(), where the only caller actually needs it.
Same observable behavior — runKeeper is called exactly once per process
— but keeper.ts is now import-side-effect-free, which makes it
straightforward to import for tests or future tooling without touching
global mouse state.
CliError used to live in cli.ts and was imported by configFile.ts purely
to grab a one-line class — the dependency arrow said 'config-file loader
depends on the CLI parser' when really it just needed a shared error
type.
Moving CliError to its own errors.ts module flattens the graph:
errors.ts (no internal deps)
|
+-- cli.ts
+-- configFile.ts
+-- config.ts (will use it in the next commit)
cli.ts re-exports CliError for any consumer that prefers to keep
importing it from there; the canonical home is now errors.ts.
The defaults now live in a single file, scripts/config.default.json:
- src/config.ts imports it via 'with { type: "json" }' and derives
DEFAULT_CONFIG (with seconds->ms conversion at the boundary), so the
CLI help text always matches what the seeded user config contains.
- scripts/install.sh copies the same file to
$XDG_CONFIG_HOME/move/config.json only if no file is already there.
Existing configs (yours or from a previous install) are never
overwritten.
- scripts/uninstall.sh does not touch the user config file at all,
following the strict Unix convention. It does print a one-line
notice pointing at the path so the user can rm -rf the dir
themselves if they want a fully-clean removal.
- tsconfig.json gains resolveJsonModule:true so tsc accepts the JSON
import.
- A small runtime assertion in src/config.ts (assertSeedShape) fails
loudly if the seed file is missing keys or has wrong types.
- README Configuration section documents the seed behavior; Uninstall
section documents the leave-config-behind policy with the rm-it-
yourself command; Files table gains scripts/config.default.json.
Verbose was awkwardly a separate runKeeper argument with a separate
resolveVerbose helper, even though it's just another tunable on the
same precedence ladder as the numeric fields. This commit collapses it.
Changes:
- Config gains 'readonly verbose: boolean'. DEFAULT_CONFIG sets it to
false.
- resolveConfig now returns the full Config including verbose. Verbose
layers with the same 'first defined value wins' precedence as the
numeric fields.
- resolveVerbose is gone.
- ParsedCliArgs.verbose becomes boolean | undefined: undefined when -V
was not passed, true when it was. parseCliArgs maps accordingly.
This lets the layered resolver treat verbose uniformly.
- runKeeper takes a single Config arg. The internal logger reads from
config.verbose at the top of runKeeper.
- move.ts no longer plumbs verbose separately; the single resolved
Config drives everything.
- README How-it-works and Files-table entries updated to match.
Behavior verified for all five verbose-precedence cases (no file/no
flag, no file/-V, file:true/no flag, file:false/-V, file:false/no
flag) and config-error scenarios remain intact.
The static usage example in README.md was added in milestone 1 and
hand-maintained. When -C/--config landed in the JSON-config-file
commit, printHelp() was updated but this block wasn't. This commit
brings them back in sync and adds the precedence line.
End users can now set defaults in a config file at:
${XDG_CONFIG_HOME:-$HOME/.config}/move/config.json
CLI flags continue to win when both are set:
CLI flags > config file > built-in defaults
Schema mirrors the CLI flag names and units. Loader is strict: unknown
keys, wrong types, and non-positive numerics are rejected with a clear
message naming the file and key, and the process exits 2.
Changes:
- New src/configFile.ts: existence-aware loader + strict schema
validation. Throws CliError; the entry point converts those to exit 2.
- src/config.ts: ConfigOverrides gains 'verbose'; resolveConfig takes
both file and CLI override layers; new defaultConfigPath() honors
XDG_CONFIG_HOME; new resolveVerbose() layers verbose with the
presence-only-CLI semantics documented.
- src/cli.ts: -C/--config <path> flag. printHelp prints the default
config path and the precedence rule.
- src/move.ts: loads the config file (default XDG path or --config),
passes both override layers into resolveConfig, resolves verbose
separately, exits 2 on any validation failure.
- README: new Configuration section with path, precedence, example,
validation rules, and the verbose CLI-can't-turn-off limitation.
Files table gains src/configFile.ts.
- install.sh, uninstall.sh, dev-setup.sh now live under scripts/.
- dev-setup.sh's 'cd $(dirname $0)' updated to '.../..' so it lands
at the repo root regardless of CWD.
- README install/uninstall curl URLs now reference scripts/<name>.sh.
- README contributor section and Files table updated to match.
Behavior is unchanged. The curl one-liner URL changes from
.../master/install.sh to .../master/scripts/install.sh; users following
the pre-v1.0.2 README will get a 404 from the old URL.
End-user installation collapses to a single command:
curl -fsSL https://gitea.cahlen.com/nokeo08/Move/raw/branch/master/install.sh | sh
Changes:
- install.sh rewritten as a POSIX sh, curl-pipeable end-user installer.
Honors XDG_DATA_HOME and XDG_BIN_HOME (de facto). Idempotent via a
version-marker file at $INSTALL_DIR/.installed-version; MOVE_FORCE=1
overrides. Hard-fails if Bun is missing (no auto-install). Includes
a safety guard refusing rm -rf on too-broad install dirs.
- uninstall.sh added; same curl-pipe pattern, shared XDG path
resolution, leaves Bun alone.
- dev-setup.sh added (= the previous install.sh content, retitled for
contributors and pointing at bun run start / bun link / install.sh).
- README updated: curl one-liner Install section, XDG behavior tables,
new Uninstall section, new 'For contributors' section, Files table
rows for all three scripts.
The plan is being implemented in the following commit. The working-notes
document has served its purpose and is removed to keep master focused
on shipping artifacts.
- src/move.ts entry point with CLI parsing, --help, --version
- src/cli.ts: parseCliArgs, printHelp, ParsedCliArgs, CliError, VERSION
- src/config.ts: Config type, DEFAULT_CONFIG, resolveConfig
- src/keeper.ts: synthetic-activity sweep + idle-watch loop
- package.json bin entry + shebang for 'bun link' global install
- install.sh: contributor bootstrap (will be repurposed; see
DISTRIBUTION-PLAN.md for the end-user installer design)
- DISTRIBUTION-PLAN.md captures the tabled end-user distribution work
nokeo08