tom_dcli_exec 1.9.0

Base D4rt CLI with dcli bridges using tom_d4rt_exec (analyzer-free). Extensible foundation for D4rt tools.

1.9.0

Changed — the VS Code scripting API's todos are the shared todo model

Resolves tom_vscode_scripting_api 2.0.0 (>=2.0.0), whose todo API speaks TomTodo from tom_todo_model, and tom_vscode_scripting_api_bridges is regenerated for it. A script that used TodoItem, TodoStatus, TodoPriority, TodoScope or TodoReference now uses TomTodo, TomTodoStatus, TomTodoPriority, TomTodoScope and TomTodoReference; reads answer TodoEntry (.todo, .sourceFile, .level) and lists TodoListResult (.todos, .entries, .invalid). The scripting API's CHANGELOG has the details.

1.8.5

Changed — resolves the web int fix release (dfin9)

tom_d4rt_exec ^1.51.0, tom_d4rt_ast ^0.210.0, tom_ast_generator ^0.1.17. No code change.

1.8.4

Changed — resolves the correctness release (dfin7)

tom_d4rt_exec ^1.49.0, tom_d4rt_ast ^0.209.0, tom_ast_generator ^0.1.16. No code change; the bridges are fresh against the new release and the suite passes, so no REPL script relied on a name only an import of an import declares (dfin6).

1.8.3

Changed — resolves the security release; bridges regenerated (dfin4)

tom_d4rt_exec ^1.46.0, tom_d4rt_ast ^0.207.0, tom_ast_generator ^0.1.15. cli_api_bridges carries the interpreter's new public surface (loadedSourceModuleCount), and tom_vscode_scripting_api_bridges its 1.1.3 API. Scripts run through the REPL keep every permission they had: the REPL grants FilesystemPermission.any and ProcessRunPermission.any, so neither the script-runner import check nor the execute rule narrows what a REPL script can do.

1.8.2

Changed — resolves the interpreter releases sci1/sci2 published (sci3)

tom_d4rt_exec ^1.43.0, tom_d4rt_ast ^0.206.0, tom_ast_generator ^0.1.14. No code change; the bridges are unchanged.

1.8.1

Changed — bridges regenerated with tom_d4rt_generator 1.51.0 (sch3)

dev tom_d4rt_generator ^1.51.0. 1.51.0 formats its output at this package's language version, so every generated *.b.dart is now dart format-clean as written. The generator's other 1.51.0 changes are in its own CHANGELOG.

bridges_fresh_test's BRIDGE-FRESH-02 now matches D4\s*.extractBridgedArg: the formatter may break the call across lines, and the old literal pattern stopped finding it.

1.8.0

Changed — resolves the published tom_d4rt_exec 1.39.0 / tom_d4rt_ast 0.195.0 (scf34)

tom_d4rt_exec ^1.39.0, tom_d4rt_ast ^0.195.0, tom_ast_generator ^0.1.13, dev tom_d4rt_generator ^1.50.0. cli_api_bridges.b.dart is regenerated: it bridges exec's own API, where registerFunctionTypedef gained requiredPositional / maxPositional and functionTypedefs records them (exec 1.37.0, scf20).

1.7.0

Changed — resolves the current interpreter; bridges regenerated (sce212)

Resolves tom_d4rt_exec 1.34.0 / tom_d4rt_ast 0.177.0 (the first interpreter releases since 2026-09-11) and tom_d4rt_generator 1.44.0. bridges_fresh_test went red on the upgrade, as it should: the regenerated *.b.dart now bridge the interpreter API that landed in between (for example registerFunctionTypedef's arity parameters, reuseAcrossRuns, dispose). Regenerated with d4rtgen and committed with the upgrade.

1.6.0

Fixed — three standing test failures, and the suite no longer contradicts itself (sce132)

The suite had been 421 passing / 3 failing long enough that its exit code had stopped meaning anything, which is how the next real regression gets missed.

TWO WERE THE SAME STALE ASSERTION, one layer apart. tom_d4rt_exec 1.16.0 made execute() REJECT source that does not parse, where before it ran whatever fragment the parser salvaged. A test asserting the old leniency was never updated — and it sat eight lines above one asserting the new strictness:

should exit 0 on unclosed string (parser is lenient)   <- red since 1.16.0
should exit 2 on syntax error (incomplete expression)  <- green

One said the parser was lenient, the next said it was strict, and a reader of either could not tell which was the contract. test/stdin/test_stdin.sh carried the same assertion in shell, which is the second failure — CONFIRMED rather than assumed, by running the fixture and finding 1 of its 23 checks failing on that one case.

Both now state ONE rule — source that does not parse exits 2 — and are named for it. Measured against a freshly compiled dclie: an unclosed string and an incomplete expression both report SourceCodeException: Fatal parsing errors and exit 2, the same as a runtime error.

THE THIRD WAS STALE BRIDGES, and the first reading of it was wrong. BRIDGE-FRESH-01 reported dcli_bridges.b.dart differing from a fresh generation. The diff looked like a REGRESSION rather than staleness — a re-export edge dropped, and Env.scopeKey degrading from D4.extractBridgedArg<ScopeKey<Env>> to value as dynamic with its signature string becoming InvalidType get scopeKey — so the obvious move was to refuse to commit it.

Three measurements said otherwise. The generator that WROTE the committed file (1.26.2), the one this package resolves (1.28.0) and the working tree's (1.42.0) all produce byte-identical output today, and check_package_configs.py reports the package's resolution healthy with scope 5.1.0 present. So the output is a function of the current resolution rather than of the generator: the committed file records a resolution that no longer exists, and regenerating makes it describe reality.

All eight *.b.dart files are now written by 1.28.0 — the generator this package's own resolution names — so its within-package mixed writers (1.26.2 x9) are gone rather than replaced by a mix of two.

NOTED, NOT FIXED HERE: InvalidType get scopeKey is a meaningless signature string, and the setter has lost its type check. It is GEN-079 declining to wrap a type no barrel re-exports, it is one member, and three other committed bridge files in this repo already carry it — so this regeneration brings the package into line with them rather than introducing it.

AND THE HARNESS COULD REPORT A FALSE GREEN. The two test files that drive the compiled dclie each carried their own copy of "rebuild it if stale", and the copies had DRIFTED: dcli_example rebuilt when anything under bin/ or lib/ changed, while stdin rebuilt only when bin/dclie.dart itself did — so the stdin suite could pass against a binary built before the change it was meant to test, which is worse than the red it was showing.

Making them agree then exposed what the drift had hidden. dart test runs files concurrently, and with both copies willing to rebuild, one suite recompiled bin/dclie in place while another was executing it: the shell fixture hung until the ten-minute test timeout. The narrow trigger had been concealing a real race by almost never firing.

So there is now ONE copy (test/support/dclie_binary.dart), it treats pubspec.lock as an input too — an upgraded dependency changes the binary without touching a file in this package — and it installs ATOMICALLY, compiling to .dart_tool/ and renaming into place. Rename leaves a running process on the old inode, so a concurrent rebuild can neither corrupt a binary another test is running nor race a second rebuilder. Verified by forcing the exact condition that hung: with every input newer than the binary, the suite rebuilds mid-run and exits 0.

Suite: 424 passing, 0 failing.

1.5.0

Removed — eight build_runner-era bridge files nothing imports or regenerates

lib/src/d4rt_library_bridges/ held bridges_trigger.b.dart and seven package_*_bridges.b.dart from the build_runner builder, dated 2026-02-26, ~740 KB. This package builds its bridges with the d4rtgen CLI, which never touches that folder, so the files could only rot or be hand-edited in the belief a regeneration would keep the edit.

Confirmed before deleting: no reference to the folder anywhere outside it, and d4rtgen --dry-run reports no run writes any of the eight.

tom_d4rt_generator 1.28.0's orphan check is what makes this visible, and test/bridges_fresh_test.dart now asserts nothing is orphaned.

lib/src/bridges/dcli_bridges.b.dart is deliberately NOT regenerated, for the reason recorded in tom_d4rt_dcli 1.6.0: a fresh generation degrades SettingsYaml to dynamic / InvalidType, and committing it would ship that regression to silence the gate reporting it.

1.4.0

Changed — realigned with the dcli 10 line

dcli, dcli_core and dcli_terminal move from ^8.4.2 to ^10.0.0, and dart_console to ^5.0.0 (required by dcli_terminal 10). win32 resolves 6.4.0 in consequence. Matches tom_d4rt_dcli 1.5.0, this package's analyzer-based twin.

The move was blocked, not deferred: tom_build_base pinned dcli: ^8.4.2 and sits transitively under this package, so version solving rejected dcli 10 regardless of what this pubspec declared. tom_build_base 2.11.0 widens that constraint to admit both majors.

Changed — bridges regenerated against the dcli 10 API

lib/src/bridges/*.b.dart are regenerated. dcli 9 removed members deprecated through the 8.x line, so the committed bridges named methods the package no longer has.

Removed — the FileSync.tempFile() bridge test

Removed in dcli 9 in favour of the top-level createTempFilename(), which already has two tests in the Temp Files group.

Suite 420/0/2, against a 421/0/2 baseline — the difference is exactly the deleted test, and both failures are the pre-existing stdin_preprocessing_test.dart pair.

1.3.0

Fixed — the --version banner reports the real version (sce9)

lib/src/version.versioner.dart said 1.2.2 while the pubspec said 1.3.0, so dclie --version named a release that is not what was running. The stamp is regenerated and version_stamp_test passes; published so the correction reaches the package people install rather than only this tree.

Changed — eight stale bridge files regenerated at generator 1.26.2 (sce1)

This package was recorded FRESH in the 2026-09-11 bridge survey and was not. Its own bridges_fresh_test.dart reported eight stale files — every bridge module, the relaxers, the dartscript and the test runner — two of them outside lib/src/bridges, where a regeneration scoped by directory would have missed them.

All eight are now what the generator produces, and the freshness gate passes for the first time.

1.2.2

Changed — formatted the tree once (scd82)

The analyzer-free twin of tom_d4rt_dcli, formatted in the same pass and for the same reason: the two are the same REPL on the two interpreter lines, kept in step by diffing, so unformatted layout was costing the check that keeps them equal rather than merely looking untidy. 57 of 84 files under lib and test were affected.

Layout plus two brace pairs, proven per file rather than asserted. git diff -w cannot establish inertness, because the tall style splits lines and a whitespace-insensitive diff still counts a moved boundary as a change. What was checked is the token stream, twice: whitespace stripped, then whitespace and commas stripped. 56 of the 57 files are identical to HEAD under that normalisation.

The exception is lib/src/cli/vscode_integration.dart, which gained two brace pairs — re-wrapping a long braceless if splits it across lines, which is what makes curly_braces_in_flow_control_structures fire. They were added by dart fix --code=curly_braces_in_flow_control_structures, not by hand, and every inserted chunk in that file was checked to be exactly { or }.

Worth noting for the mirror: vscode_integration.dart needed the same two braces in both twins, at the same two sites. That is the twins being twins.

dart analyze is clean, and the formatter is now idempotent here.

1.2.1

Fixed — barrel re-exports were registered under an unmatched key (GEN-125)

bridgeReExports() recorded its source as lib/tom_d4rt_cli_api.dart instead of package:tom_dcli_exec/tom_d4rt_cli_api.dart, because the bridge generator failed to root a relative source path before mapping it to a package URI. registerLibraryReExport matches that key against a script's import, which is always a package: URI — so every symbol reached only through a re-export of the CLI API barrel was invisible to interpreted scripts.

Fixed upstream in tom_d4rt_generator 1.15.3 and picked up here by regeneration; no hand edits to any *.b.dart.

1.2.0

Fixed — .start-execute blocks of bare statements failed to parse (tccc5)

D4rt.execute(source:) parses its argument as a compilation unit and calls the top-level main. A .start-execute block was handed straight to it, so a block of bare statements

.start-execute
var testVar = 100;
verify(testVar == 100, '.start-execute variable works');
.end

parsed as a top-level variable followed by a function declaration with no body and failed with Expected to find ')' / A function body must be provided.

The three-case rule that fixes this already existed — but only on the stdin path. It is now the top-level prepareProgramSource, applied at every entry point that means "run this as a fresh program": stdin, the exec command, .start-execute, and CliController.execute.

• imports + main → run unchanged
• main without imports → the bridge import block is prefixed
• bare statements → prefixed and wrapped in a generated main

The generated wrapper is async when the statements await, matching what .start-script already did; await inside a string or comment does not count. .start-execute also awaits a Future result now instead of printing it.

Requires tom_d4rt_exec >= 1.13.0, without which a run whose assertions all pass is still failed by phantom extension-registration errors. Routing .start-execute through the bridge import block — which it needs, because a bridged library's top-level functions (verify, verifyEquals, …) only become visible through its import — is what made those errors reachable from this path.

1.1.3

• Housekeeping: test artifacts now live in a gitignored testlog/ folder; doc/ no longer ships machine-generated baselines or last_testrun.json. No code changes.

1.1.2

Bug Fixes

• GEN-070 follow-up: Find class now properly bridged via generator fix (multi-chain barrel re-export)
• Removed dcli_missing_bridges.dart supplementary bridge (no longer needed)
• Removed lastModified/setLastModifed tests (not exported from dcli barrel)
• Replaced deprecated symlink() tests with createSymLink() tests

Tests

• All 389 tests pass, 0 failures, 0 skips

1.1.1

Bug Fixes

• DCLI-GEN-001: Added supplementary bridge for missing global functions (lastModified, setLastModifed, symlink)
• DCLI-GEN-002: Added Find class bridge with static getters (file, directory, link)
• DCLI-VSCODE-001: Fixed VS Code bridge import path and test constructor arguments
• DCLI-LOCK-001: Updated tests for deprecated NamedLock.withLock (dcli 8.4.2), added withLockAsync tests
• DCLI-API-001: Fixed expandDefine test prefix ($ → @)
• Symlink bridge uses createSymLink internally (avoids deprecated symlink() warning)

Tests

• All 391 tests pass, 0 failures, 0 skips

1.1.0

• Full DCli scripting support now
• Updated tom_d4rt dependency to ^1.8.1
• Regenerated bridges with latest generator (multi-barrel registration, extension filtering)

1.0.0

• Initial version.