# Semantic rewrite: migration guide and inventory Status: **migration in progress**. The target language contract is [RFC-LANG-0002](../architecture/RFC-LANG-0002-unified-semantics.md) and [`spec/mapping.md`](../../spec/mapping.md). The Rust frontend and bundled stdlib are **not yet migrated**, so executable tests are expected to be red until the implementation lands. See [`../STATUS.md`](../STATUS.md). This directory records the repository-wide migration required by the RFC: every active `.mq.md` source, example, fixture and documentation snippet must use the single new syntax. Old syntax is not an active language. ## Current state (2026-11-21) - **Text migration: complete.** Specs, active docs, examples, all non-stdlib test/fixture `.mq.md` files, the security-lane embedded fixtures and Rust-embedded `.mq.md` literals were rewritten. 1009 present tracked+new files are classified in the inventory; there are no `pending` text items. - **Implementation: pending.** 51 inventory rows remain `pending-implementation`: the 9 frontend/expression/IR files that implement the mapping and the 42 stdlib fileset entries (sources + manifest). - **Verification:** `cargo fmt --check`, `cargo check --workspace --tests`, `conformance_ast` (4/4) and `docs_consistency` (4/4) pass. The semantic, negative, property and CLI suites that execute Marqdo source are expected red against the pre-RFC frontend. Evidence: [`../evidence/2026-11-21-semantic-docs-migration.md`](../evidence/2026-11-21-semantic-docs-migration.md). - Old v1 compile-negative fixtures were deleted and replaced by new-semantic negatives; they are intentionally absent from the inventory rather than counted as reviewed. ## Migration rules for this repository 1. **One syntax.** Do not add `marqdo_semantics`, `--legacy`, `syntax: v1` or a second example. Old syntax is either migrated or marked historical. 2. **Exec H2 is explicit.** Write `## @exec 名称` for every callable function, even under an Exec H1. Heading names keep the original semantic name (`## @exec 阶乘`, not `## @exec @exec 阶乘`). 3. **Mode on H1.** Prefer an explicit `# @knowledge 名称` for document namespaces and `# @exec 名称` only when the whole namespace is executable. 4. **One payload, one statement.** Use a standalone inline code span by default (`` `x = 1` ``) because it is Markdown-unambiguous. Use `**...**` or `*...*` when the fixture is specifically testing carrier equivalence. 5. **Control flow.** `if ` … `[else]` … `end`; `while ` … `end`. A control block never crosses a heading. `return ` replaces the old italic return. 6. **Branch migration.** An ordered list with N arms is *first true arm wins*. Convert N arms to a nested if/else chain (an inner `if` inside `else`), not to N independent `if`s. A single arm becomes `if`. A bullet list is a sequence of independent `while` loops, one per item, in item order. 7. **Calls.** `> print(x)` was an old call; in the new language it is inert prose. Write `> **print(x)**`, `` `print(x)` `` or a plain paragraph payload. A block quote itself never calls. 8. **Signatures.** The first standalone code span `(n: Int)` becomes `params(n: Int)` as the first direct payload of the Exec H2. Keep parameter names, order and defaults. 9. **Knowledge.** Purely visual emphasis and explanatory lists belong in `@knowledge` sections. A `Strong` inside an Exec section is a payload and must be a valid statement. 10. **Preserve meaning.** Do not change function names, argument names/defaults, evaluation order, early-return behavior, values, stdout or expected error intent. For `.expected.json` value goldens (return value/args/stdout), keep the file unless the semantic name really changed. 11. **Expected diagnostics.** Old shape errors about lists/quotes/emphasis become valid inert Markdown, not errors. Replace them with new negatives for `sem.*` codes (see `spec/mapping.md` §8). Keep genuine runtime negatives (bounds, type, permission) and only migrate their source spelling. 12. **No silent golden regeneration.** Do not run `UPDATE_*_GOLDENS=1` to mask a semantic change. A golden update needs a stated new-semantic reason. 13. **Tests are text-only here.** Rust test strings and the `init` template text are migrated; the frontend, expression parser and stdlib `.mq.md` sources are not. Those implementation items are tracked as pending in the inventory. Do not change protocol schemas or security logic in this migration. 14. **Use the review aids, not them as proof.** `tools/dev-only/check-mq-syntax.py` catches unbalanced control blocks and unknown `sem.*` codes; `tools/dev-only/gen-migration-inventory.py` regenerates the inventory. Both are audit aids; the real gate is the migrated frontend plus the RFC §12 acceptance matrix. ## Conversion examples Old: ```markdown ## 分类 1. `n < 0` *-1* 2. `n == 0` *0* 3. `n > 0` *1* ``` New (first-true chain, explicit `end`): ```markdown ## @exec 分类 `if n < 0` `return -1` `else` `if n == 0` `return 0` `else` `return 1` `end` `end` ``` Old loop: ```markdown ## 求和 `(n: Int)` **i = 0** **sum = 0** - `i < n` **sum = sum + i** **i = i + 1** *sum* ``` New: ```markdown ## @exec 求和 `params(n: Int)` `i = 0` `sum = 0` `while i < n` `sum = sum + i` `i = i + 1` `end` `return sum` ``` ## Inventory `semantic-rewrite-inventory.tsv` is generated from the tracked tree with `python3 tools/dev-only/gen-migration-inventory.py` and records each file's category, action and status. Columns: ```text path category active action verification status ``` Categories: `spec`, `architecture`, `tutorial`, `example`, `stdlib`, `test`, `rust-test`, `historical`, `non-semantic`. Status: `migrated`, `reviewed`, `historical`, `pending-implementation`, `pending`. The final gate is **100% of tracked files classified** and all active `.mq.md` files migrated, not a grep with no matches.