[DamianReeves]

Summary

This PR implements the source-tree slice of the OpenSpec change run-core-wasm-guests-from-any-language
(tasks 1.3, 2.1, 2.2, 2.4 and 2.5; beads morphir-ukvn.5.64.37.1 and morphir-ukvn.5.64.37.2).

• --language (SEL-3). When the guest claims one language, the CLI uses it. When it claims several,
  --language must name one: without it the command exits 2 with extension.input_invalid, and the message names
  the claimed languages. An unclaimed language exits 2 with extension.capability_unavailable. The walk reads only
  the file extensions of the selected language. Both refusals start no operation session.
• URIs from the source base (SRC-1). Each document URI is the file's path relative to --source-base, or,
  without it, to the directory that the user names (option B, decided by Damian Reeves on 2026-11-01). Thus
  compile src sends app/money.gleam from any current directory, as main does. With --source-base /work/shop,
  compile /work/shop/src/app sends src/app/money.gleam. An input outside the source base exits 2 with
  extension.input_invalid, and the message names the base. A single file without --source-base keeps the
  earlier rule, so MoonBit requests do not change. The request has no sources.root.
• Percent-encoded segments (SRC-3). Unreserved characters stay; every other UTF-8 byte is %XX.
  my module.gleam becomes my%20module.gleam. Names with : or \ are now encoded, not refused.
• Order and refusals (SRC-2, FIT-1). Documents are sorted by the whole URI in byte order (a-c.gleam before
  a/b.gleam). A repeated URI is refused. A new 1 MiB bound applies to one file. Every limit message names the
  limit and its value; a link or special file is named by its path relative to the base; a file that is not UTF-8
  is named by its URI. All of these keep extension.input_invalid, exit 2. The 4 MiB request check stays last.
• Determinism (2.4). Two compiles of the same tree send byte-identical compileModel requests; other trees
  with the same package and options keep the same contextDigest.
• Docs (2.5). apps/morphir/docs/extension.md step 4 and a new section "Source trees and document URIs";
  the ignore-file docs move to task 2.6, which waits for Ignore-file library and moonx tool: Git-compatible, custom ignore files, explain, layer and containment options moonrockz/moonx-tools#16.
• A one-line fix clears the only MoonBit warning in apps/morphir on wasm and wasm-gc (unused stop_requested).

Without --source-base, a directory compile sends the same URIs as main, so this PR has no breaking change for
directory input. The first version made the current directory the default base (option A); the extension
qualification suite of #298 showed that it broke the published Elm 0.5.0-preview.2 and Python 0.4.2 extensions,
so Damian Reeves chose option B. A Cargo-style source root (from the project manifest and the extension's
workspace discovery) is a separate follow-up.

Out of scope: 2.6 (ignore files), the group 1 conversion tasks (1.6 to 1.13), groups 3 and 4.

Check with the released Gleam guest

Local pin (contract 0.2.0, wasm / morphir-bytes-v1, descriptor) over extension/gleam/v0.4.0
(morphir-gleam-0.4.0.wasm SHA-256 12349dac…205e, release.json SHA-256 be337f8f…93dc), on macOS arm64.
Tree: src/shared/numbers.gleam, src/app/money.gleam (imports shared/numbers.{type Amount}),
src/app/orders.gleam (imports app/money.{type Money}).

Run	Result
native extension compile src from the project, no --model-format	exit 0, JSON tuple agreed, modules app/money, app/orders, shared/numbers, cross-module reference acme/shop:shared/numbers#amount
native, compile . from the project, and compile <abs>/src from another directory	exit 0, model.json byte-identical to the run above (option B, MoonBit 0.10.18)
native, compile <abs>/src --source-base <abs> from another directory	exit 0, model.json byte-identical to the run above
native, compile src/shared	exit 0, module shared/numbers
native, --source-base /tmp/elsewhere	exit 2, extension.input_invalid naming the base
native, --model-format ion	exit 2, extension.tuple_unavailable
native, extension generate model/model-envelope.json	exit 0, app/money.gleam, app/orders.gleam, shared/numbers.gleam
npm CLI, same compile and generate	exit 0; model.json byte-identical to the native one
file src/app/my notes.gleam	URI src/app/my%20notes.gleam; the guest decodes it once to my notes and refuses it as a module name (INVALID_MODULE_URI)

The last row is evidence for the open point "Percent-encoding (SRC-3)" of design.html: the guest decodes encoded
segments. Gleam 0.4.0 itself does not resolve imported or qualified values (GLEAM_UNSUPPORTED_VALUE), so the
tree imports types only; that is a guest limit, not a host one.

Test plan

• moon test in apps/morphir on native, js, wasm and wasm-gc
• moon check in apps/morphir on all four targets: no warnings in the module
• mise run lint, mise run lint:moonbit, mise run validate
• openspec validate run-core-wasm-guests-from-any-language --strict
• Released morphir-gleam 0.4.0 module through the native and npm CLI (table above)
• mise run test:wasm-workers -- --target native: 3791 of 3792. The failure is the timing test
  a launch waits while another process briefly holds the copy open for writing
  (pkgs/morphir-host/worker/process/native/unix_wbtest.mbt:124), a package that this PR does not change; it
  passes 3 of 3 when run alone
• bun run --cwd apps/morphir test native: the process-worker acceptance and the Extism acceptance (Python
  0.4.2 compiles src) pass. Each of two runs stopped on a native SIGSEGV in another
  command (extension describe in run 1, morphir generate in test-generation.ts in run 2), which matches the
  open bead morphir-fr6z (flaky native SIGSEGV); the same describe passes 3 of 3 when run by hand

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-31T15:45:33.192728Z	398e86a	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[DamianReeves]

Rebased onto main at 109ee20 (MoonBit 0.10.18+6c83a7f62, moonbitlang/async 0.22.7, #286 and #290). New head: 12163ba.

Rebase. There were no git conflicts. But the extension qualification of #298 showed a behavior conflict with the first default source base (the current directory, option A). compile generated/src sent generated/src/Acme/Rules/Main.elm, so Elm 0.5.0-preview.2 refused the recompile (elm.module_name). compile src named the Python 0.4.2 module src.acme.rules, so the consumer import and the recompile failed. Damian Reeves chose option B on 2026-11-01: without --source-base, a directory input is its own base (the URIs of main), and --source-base overrides it. Commit 12163ba makes this change and updates the spec "Relative document URIs", design.html (SRC-1), tasks.md, proposal.md, the CLI help and docs/extension.md. The qualification driver needs no change.

Checks on 0.10.18 (macOS arm64):

Check	Result
moon test in apps/morphir	native 3792/3792, js 3883/3883, wasm 3476/3476, wasm-gc 3430/3430
moon check, 4 targets	0 warnings in the module
mise run lint, lint:moonbit, validate; openspec validate --all --strict	pass; 5 of 5
mise run test:cli	pass: worker host and Extism acceptance on native, js and npm
mise run test:extension-qualification	all 5 suites pass: elm 16 + 2 known failures, python 32, moonbit-native 12 + 1 known, moonbit-portable 11 + 1 known, conformance 11
pre-push gates	pass
Gleam 0.4.0 module	compile src, compile . and compile <abs>/src from another directory give a byte-identical model.json (modules app/money, app/orders, shared/numbers), equal to the 0.10.17 run; generate gives the same 3 files

Under option A (before 12163ba), one native run had a failure in the known timing test unix_wbtest.mbt:124; it passed 39/39 when run again. No native SIGSEGV (morphir-fr6z) occurred. No test asserts on the new 0.10.18 stderr output, so no test changed for the toolchain. Locally, test:cli needs moon build pkgs/morphir-host/worker/thread/entry --target js --release first, as CI does.