Files
marqdo-cli/spec/analysis-v2.md
T
cflmy 92bfbf2568 feat(rfc-cm): baseline P0-P3 compute machine, stdlib and resource slices
Implements the first vertical slices of RFC-CM-0001/0002 on top of a113d89:

- P1: opaque Memory regions (64 KiB pages, budgets, grow-only), byte
  read/write, Memory-index place assignment, wire/host-boundary rejection.
- CM-A04/A05: value-semantics List/Map/Text/Table primitives and the
  recoverable Result map convention.
- CM-A06: opaque Resource handles, resource.open/read/write/close, stdout
  write, granted project-file read and buffered write, strict fail-closed.
- CM-A07 Compat-A: explicit .mq.md format/utf8/io chain matching builtin
  print for the covered value types; builtin remains default.
- P2 fixtures: allocators, list/tree/graph/hash, integer codecs, UTF-8
  encode/decode, CSV/JSON subsets, table filter/sort/join, bytecode VM.
- ADRs 0022-0027, spec updates, evidence reports and 68 semantic goldens.

Gate: cargo fmt --check, clippy -D warnings, 407 tests passed.
2026-11-20 21:16:54 +08:00

182 lines
6.7 KiB
Markdown

# Marqdo Analysis v2
Status: **implemented by `marqdo check --emit <path>` (S2); ADR-0022 adds an
additive index target shape.** Supersedes
Analysis v1. v1 was never released; it is kept as history and is not written.
## 1. Invariants
`check --emit`:
* never executes the document, loads a plugin shared library, starts a foreign
kernel or imports a module implicitly;
* validates the loaded program first; on a compile error it writes no artifact
and removes an existing artifact at the target path;
* writes atomically through a temporary file and rename;
* contains no absolute machine paths, timestamps, PIDs or other environmental
data, so the same logical project emits byte-identical Analysis from
different checkout directories and is safe to compare across machines.
## 2. Top-level shape
```json
{
"schema": "marqdo.analysis",
"schema_version": 2,
"producer": {
"name": "marqdo",
"version": "<crate version>",
"markdown_profile": "marqdo.md.v1"
},
"project": {
"root_id": "root:quantum@0.1.0",
"identity_kind": "manifest",
"lock_digest": "sha256:<64 hex>"
},
"modules": [],
"diagnostics": []
}
```
`project.identity_kind` is:
| Kind | `root_id` | Meaning |
|---|---|---|
| `manifest` | `root:<project.name>@<project.version>` | Logical identity of a project with `marqdo.yaml`. Moving the checkout or a Git clone does not change it. |
| `workspace_snapshot` | `sha256:<64 hex>` | No manifest. Digest of sorted `(module_id, source_digest)` pairs; identifies content, not a persistent project. |
`project.lock_digest` is the SHA-256 of the exact `marqdo.lock` bytes prefixed
with `sha256:`, or `""` when there is no lock. It is not a logical identity.
`AnalysisDigest` is the lowercase hex SHA-256 of the exact UTF-8 bytes written
by `check --emit` (pretty JSON). This is what Trace v2 binds to
`source.analysis_digest`; an independent consumer can verify it by hashing the
file. Diagnostics and any configuration options that change the projection are
part of that digest.
Module ids are `<package instance>::<package-relative path>` (ADR-0008) and are
scoped by `project.root_id`. A module id alone is not a cross-project symbol.
## 3. Modules
The module object is unchanged from Analysis v1 except for `resources`:
```json
{
"id": "root:res@0.1.0::index.mq.md",
"package_instance_id": "root:res@0.1.0",
"relative_path": "index.mq.md",
"source_digest": "<64 hex of exact source bytes>",
"units": [],
"functions": [],
"imports": [],
"references": [],
"calls": [],
"resources": []
}
```
Modules are sorted by `id`; functions, units and calls keep document order and
the v1 semantic node shapes (`assign`, `expression`, `call_site`, `branch`,
`loop`, `return`; literals, variables, lists, records, unary/binary, member,
index, call). `imports`, `references` and `calls` remain static facts; a
Markdown reference is not a call. `effects` remains `["unknown"]` (a
conservative summary, not a purity proof).
An `assign` node keeps the v1 string `target` for a local rebinding. Addressable
memory adds one additive target form (ADR-0022); consumers that only understand
local assignments can ignore an object target:
```json
{"kind":"assign",
"target":{"kind":"index",
"base":{"kind":"variable","name":"内存","span":{"start":23,"end":26}},
"index":{"kind":"literal","type":"int","value":"0","span":{"start":27,"end":28}},
"span":{"start":23,"end":29}},
"value":{"kind":"literal","type":"int","value":"7","span":{"start":32,"end":33}},
"span":{"start":18,"end":35}}
```
`target.kind` is `"index"`; `base` and `index` use the same expression node
shapes as every other expression. Analysis is static: it records the place but
never allocates a region or executes the write.
## 4. Resources
`resources` is an array, not a count. v2 projects exactly the two resource
kinds represented by `marqdo-ir::Resource`; it does not invent table, HTML or
link resources.
```json
{
"id": "root:res@0.1.0::index.mq.md#resource:formula-1",
"module_id": "root:res@0.1.0::index.mq.md",
"unit_id": "root:res@0.1.0::index.mq.md#calc",
"kind": "formula",
"span": {"start": 30, "end": 44},
"content_digest": "<64 hex>",
"language": null,
"display": true
}
```
| Field | Semantics |
|---|---|
| `id` | `<module_id>#resource:<kind>-<n>`; `n` is the 1-based ordinal among resources of that kind in the module, in document order |
| `module_id` | redundant with the enclosing module, present for flattened consumers |
| `unit_id` | innermost owning unit |
| `kind` | `formula` or `code`; future kinds are additive and unknown kinds must be ignored |
| `span` | Markdown node byte span, including math delimiters/fence, relative to that module's source |
| `content_digest` | SHA-256 of the payload: formula LaTeX bytes without delimiters; exact code fence source bytes |
| `language` | code fence info string (may be `""`); `null` for formulas |
| `display` | formula only: block vs inline math |
Resource ids are Analysis-local ordinals, not persistent symbols and not `ref`
anchors. Content identity is `(module_id, content_digest)`; location identity is
`(module_id, span)`. Inserting an earlier resource renumbers later ordinals, but
the content digest and span still let a consumer detect the change. Payloads
are not copied into Analysis: consumers read the module source (via
`relative_path`) and slice `span` or verify `content_digest`.
## 5. Diagnostics
Diagnostics no longer embed absolute physical paths:
```json
{
"module": "root:res@0.1.0::index.mq.md",
"relative_path": "index.mq.md",
"span": {"start": 10, "end": 20},
"line": 2,
"column": 3,
"severity": "warning",
"code": "markdown.inert_executable",
"message": "...",
"hints": []
}
```
`span` is the primary coordinate; `line`/`column` are presentation-derived from
the owning module. `span`, `line` or `column` may be `null` when no precise
location exists.
## 6. Compatibility
* Writers emit `schema_version: 2`. There is no v1 writer and no
`--emit-version` flag; v1 was experimental.
* Consumers must reject unknown `schema_version` values and unknown resource
kinds rather than silently accepting them.
* Changing `root_id`, resource fields, diagnostic fields or the digest
definition is a protocol break requiring a new version and an ADR.
* JSON Schema: `spec/schemas/analysis-v2.schema.json` (Draft 2020-12).
Schema validation is structural; cross-record checks (span inside source,
call targets, digest verification) remain semantic tests.
## 7. Tests
* `analysis_is_deterministic_and_has_stable_semantic_nodes`
* `analysis_v2_identity_is_location_independent`
* `analysis_v2_projects_addressable_formula_and_code_resources`
* `failed_check_removes_stale_analysis`
* `tests/consumers/analysis_consumer.py` (independent Python consumer)