[mparrett]

Fixes #1096. Replaces #1097 with the branch hosted directly in nooga/let-go.

lg can load a namespace's file but cannot call its -main from the command line. Programs need a launcher or manual -e call, which also prints the return value and treats the first program argument as a script.

Adds lg [-source-paths …] -m ns [args…], with --main as an alias and ns/fn to select another callable var. The namespace is required, all following arguments are passed unchanged and bound to *command-line-args* before initialization or loading, and the entry's return value is discarded. Loading and invocation errors exit nonzero; missing or non-callable entries report the CLI target.

-m is terminal, so lg options precede it and flag-shaped arguments or a literal -- afterward belong to the program. An -e expression runs first and keeps its usual printed result; initialization failure prevents the entry call. -r retains the post-execution REPL behavior. Combining -m with -c, -b, or -w is a usage error.

Persistent storage uses the selected source filename, matching script mode even when namespace hyphens map to filename underscores. The CLI and loader share the same source-path search, preserving root, extension, and filename precedence. Selection does not execute source, so storage is installed before namespace initialization. Explicit -storage-id still wins; namespaces without a filesystem source retain the namespace-derived fallback.

Compiler paths: the entry uses the existing namespace loader and VM execution context. Source selection is shared by the default, bootstrap, and Go-lowered builds; compiler/IR semantics are unchanged. Regenerated manifest and sums accompany the resolver refactor.

Validation:

• Full make quality before/after runs (1,053 tracked files, fixed --since 2026-04-29): erosion 0.585 → 0.584, callables over CC 10 535 → 534.
• Quality-driven refactor: runMain CC 73 → 55 (upstream base: 58); parseRunFlags 13 → 10. Main-option validation, storage-source selection, and program execution are separate helpers; all new helpers are at CC 10 or below, and main-entry failure handling uses early returns.
• PR-delta report over pkg/cli, pkg/resolver, and the new main-flag tests: erosion 0.581 → 0.514 after the refactor (upstream baseline for that same scope: 0.533). Scoped aggregate debt 0.405 → 0.376; remaining findings include pre-existing complexity and history-derived terms, not a claim that the whole corpus is clean.
• CLI main/argument regressions and CLI/resolver unit tests pass.
• Storage regression cases fail before the fix and pass afterward: underscore .clj, nested .cljc, literal hyphens, competing extension/name candidates, load-time storage reads, isolation, and explicit IDs.
• Bootstrap CLI/resolver tests and main-entry regressions pass.
• make generate and make check-generated pass, including Go-lowered runtime compilation and native dispatch.
• Repository commit and pre-push checks run before publication; formatting, test locations, scoped comment review, and manifest freshness are checked.

Local lint limitation: the pinned linter cannot read local Go 1.27.3's export-data format (internal/goarch, version 5 versus supported version 4). Only that commit hook is skipped; CI remains the lint check.

[nnunley]

The nesting looks a little deep in the code. arethe erosion metrics running? Otherwise, looks like it's ok.

[mparrett]

@nnunley Thanks for the review. Ran the full erosion/quality report and refactored the nested CLI logic into helpers in 4370bfe. runMain complexity dropped from 73 to 55 (upstream: 58), the flag parser from 13 to 10, and full-project erosion from 0.585 to 0.584. Regression tests and all pre-push checks passed; the PR description has the measurements.

[mparrett]

cc: @dharmatech

[dharmatech]

cc: @dharmatech

Love it!
Thanks guys! 🫡

CC: @gavr123456789

[nooga]

Approving. Read the CLI diff: parseRunFlags makes -m terminal so flags and -- after it reach the program, validateMainMode rejects -c/-b/-w, and SourcePath is shared with the loader so storage id and namespace lookup agree. nnunley's nesting nit is addressed. As project owner I'm fine with the new -m/--main flag.