@geml/geml 1.3.2 → 1.4.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/geml.js CHANGED
@@ -5,15 +5,18 @@
5
5
  // `meta` data block, ATX headings, lists and paragraphs, the attribute object
6
6
  // with §4 value typing, and a document-model JSON serialization.
7
7
  //
8
- // M2: inline parsing of flow blocks (§5 — emphasis/strong/strike, code, math,
9
- // media embeds, links, auto-references, footnotes) and build-time reference
10
- // validation (§8 — unique ids, resolvable internal/cross-document references).
11
- import { readFileSync, writeFileSync, realpathSync, statSync } from "node:fs";
8
+ // M2: inline parsing of unfenced blocks (§5 — emphasis/strong/strike, code,
9
+ // math, media embeds, links, auto-references, footnotes) and build-time
10
+ // reference validation (§8 — unique ids, resolvable internal/cross-document
11
+ // references).
12
+ import { readFileSync, writeFileSync, realpathSync, statSync, existsSync } from "node:fs";
12
13
  import { basename, dirname, isAbsolute, join, relative, resolve as resolvePath, sep } from "node:path";
13
14
  import { fileURLToPath } from "node:url";
14
15
  import { spawnSync } from "node:child_process";
15
16
  import { commit, restore, verify, listRevisions, resolveContent, firstChangedContent } from "./history.js";
16
17
  import { renderHtml } from "./render-html.js";
18
+ import { normalizeBlockId } from "./block-edit.js";
19
+ import { normalizeSource } from "./diagnostics.js";
17
20
  import { coerce, parseAttrs } from "./attrs.js";
18
21
  import { META_REF_SRC, parseInline } from "./inline.js";
19
22
  import { parseTable } from "./table.js";
@@ -25,6 +28,17 @@ export { mdToGeml } from "./from-md.js";
25
28
  export { renderHtml } from "./render-html.js";
26
29
  export { serialize } from "./serialize.js";
27
30
  export { gemlToMd } from "./to-md.js";
31
+ // A block id is any non-whitespace run (§4), so it may contain regex
32
+ // metacharacters. Every place that builds a RegExp from an id MUST run it
33
+ // through this first, or a crafted id (`#a(`, `#(x+x+)+y`) turns a labeled-close
34
+ // or reference match into an uncaught `SyntaxError` or a ReDoS on the main
35
+ // parse path (SEC: document-controlled RegExp injection).
36
+ function reLit(s) {
37
+ return s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
38
+ }
39
+ // Re-exported from ./diagnostics.js so that `Diagnostic` stays importable from
40
+ // the package root. The catalogue of codes lives there (spec Appendix A).
41
+ export { SEVERITY } from "./diagnostics.js";
28
42
  // Type registry: which body mode each typed block uses. Unknown types are a
29
43
  // warning and fall back to `raw` (forward compatibility, §3/§8).
30
44
  const REGISTRY = {
@@ -118,7 +132,7 @@ function interpolate(text, line, ctx) {
118
132
  if (ctx.meta.has(key))
119
133
  out += ctx.meta.get(key);
120
134
  else {
121
- ctx.diags.push({ severity: "error", message: `unknown metadata reference \`{{${key}}}\``, line });
135
+ ctx.diags.push({ severity: "error", code: "unknown-metadata-reference", message: `unknown metadata reference \`{{${key}}}\``, line });
122
136
  out += m[0];
123
137
  }
124
138
  i = META_REF.lastIndex;
@@ -133,7 +147,7 @@ function interpolate(text, line, ctx) {
133
147
  // Register a block id, flagging duplicates as errors (§4: ids unique per doc).
134
148
  function registerId(ctx, id, line) {
135
149
  if (ctx.ids.has(id)) {
136
- ctx.diags.push({ severity: "error", message: `duplicate id \`#${id}\` (first defined at line ${ctx.ids.get(id)})`, line });
150
+ ctx.diags.push({ severity: "error", code: "duplicate-id", message: `duplicate id \`#${id}\` (first defined at line ${ctx.ids.get(id)})`, line });
137
151
  }
138
152
  else {
139
153
  ctx.ids.set(id, line);
@@ -203,7 +217,7 @@ function parseList(lines, i, base, ctx) {
203
217
  // rather than building a model that overflows the renderer (DoS). One
204
218
  // diagnostic per over-deep list; content is preserved, just flattened.
205
219
  if (!tooDeep) {
206
- ctx.diags.push({ severity: "error", message: `list nesting too deep (max ${MAX_NESTING})`, line: base + i + 1 });
220
+ ctx.diags.push({ severity: "error", code: "list-nesting-too-deep", message: `list nesting too deep (max ${MAX_NESTING})`, line: base + i + 1 });
207
221
  tooDeep = true;
208
222
  }
209
223
  cur = top.list;
@@ -276,7 +290,7 @@ function scanBlocks(lines, base, ctx, depth = 0) {
276
290
  // of any length ≥ 3 followed by the block's id). The labeled close is a
277
291
  // *local* close: it can't be gotten wrong by miscounting `=`, so it is the
278
292
  // safe way to nest (§3).
279
- const labeled = attrs.id !== undefined ? new RegExp(`^={3,}[ \\t]+#${attrs.id}[ \\t]*$`) : null;
293
+ const labeled = attrs.id !== undefined ? new RegExp(`^={3,}[ \\t]+#${reLit(attrs.id)}[ \\t]*$`) : null;
280
294
  const body = [];
281
295
  let j = i + 1;
282
296
  let closed = false;
@@ -289,11 +303,11 @@ function scanBlocks(lines, base, ctx, depth = 0) {
289
303
  }
290
304
  if (!closed) {
291
305
  const how = attrs.id !== undefined ? `${"=".repeat(openLen)} or \`=== #${attrs.id}\`` : "=".repeat(openLen);
292
- diags.push({ severity: "error", message: `unterminated \`${type}\` block (no matching ${how})`, line: openLineNo });
306
+ diags.push({ severity: "error", code: "unterminated-block", message: `unterminated \`${type}\` block (no matching ${how})`, line: openLineNo });
293
307
  }
294
308
  let mode = REGISTRY[type];
295
309
  if (mode === undefined) {
296
- diags.push({ severity: "warning", message: `unknown block type \`${type}\`; body kept as raw`, line: openLineNo });
310
+ diags.push({ severity: "warning", code: "unknown-block-type", message: `unknown block type \`${type}\`; body kept as raw`, line: openLineNo });
297
311
  mode = "raw";
298
312
  }
299
313
  const block = {
@@ -317,7 +331,7 @@ function scanBlocks(lines, base, ctx, depth = 0) {
317
331
  // Refuse to recurse past the cap: emit a diagnostic and keep the body
318
332
  // as raw so the parser returns cleanly instead of overflowing the
319
333
  // call stack on a pathologically nested document (DoS).
320
- diags.push({ severity: "error", message: `block nesting too deep (max ${MAX_NESTING}); body kept as raw`, line: openLineNo });
334
+ diags.push({ severity: "error", code: "block-nesting-too-deep", message: `block nesting too deep (max ${MAX_NESTING}); body kept as raw`, line: openLineNo });
321
335
  block.raw = body;
322
336
  }
323
337
  else {
@@ -347,7 +361,7 @@ function scanBlocks(lines, base, ctx, depth = 0) {
347
361
  // §7: native chart — resolved in a second pass (data=#id may be
348
362
  // defined later in the document).
349
363
  if (body.length > 0 && body.some((l) => l.trim() !== "")) {
350
- diags.push({ severity: "warning", message: "geml-chart body is ignored; the chart spec lives in attributes", line: openLineNo });
364
+ diags.push({ severity: "warning", code: "ignored-diagram-body", message: "geml-chart body is ignored; the chart spec lives in attributes", line: openLineNo });
351
365
  }
352
366
  (ctx.charts ??= []).push({ block, line: openLineNo });
353
367
  }
@@ -357,18 +371,18 @@ function scanBlocks(lines, base, ctx, depth = 0) {
357
371
  // ("view config travels with the data"). Body is empty.
358
372
  const src = attrs.attrs["src"];
359
373
  if (typeof src !== "string" || src === "") {
360
- diags.push({ severity: "warning", message: "geml-code-graph: missing `src=` (nothing to render)", line: openLineNo });
374
+ diags.push({ severity: "warning", code: "code-graph-missing-src", message: "geml-code-graph: missing `src=` (nothing to render)", line: openLineNo });
361
375
  }
362
376
  else if (ctx.resolveDoc && ctx.resolveDoc(src) === null) {
363
- diags.push({ severity: "warning", message: `geml-code-graph: cannot resolve document \`${src}\``, line: openLineNo });
377
+ diags.push({ severity: "warning", code: "code-graph-unresolvable-document", message: `geml-code-graph: cannot resolve document \`${src}\``, line: openLineNo });
364
378
  }
365
379
  if (body.length > 0 && body.some((l) => l.trim() !== "")) {
366
- diags.push({ severity: "warning", message: "geml-code-graph body is ignored; the embed is configured by `src=` alone", line: openLineNo });
380
+ diags.push({ severity: "warning", code: "ignored-diagram-body", message: "geml-code-graph body is ignored; the embed is configured by `src=` alone", line: openLineNo });
367
381
  }
368
382
  }
369
383
  else if (typeof fmt === "string" && !DIAGRAM_RENDERERS.has(fmt)) {
370
384
  // §7: warn on a diagram format with no registered renderer.
371
- diags.push({ severity: "warning", message: `no registered renderer for diagram format \`${fmt}\`; body kept raw`, line: openLineNo });
385
+ diags.push({ severity: "warning", code: "unknown-diagram-format", message: `no registered renderer for diagram format \`${fmt}\`; body kept raw`, line: openLineNo });
372
386
  }
373
387
  }
374
388
  }
@@ -436,7 +450,7 @@ function parseData(lines) {
436
450
  // resolving `other.geml#id` references.
437
451
  function gatherIds(source) {
438
452
  const ctx = { diags: [], ids: new Map(), refs: [], meta: new Map() };
439
- scanBlocks(source.replace(/\r\n?/g, "\n").split("\n"), 0, ctx);
453
+ scanBlocks(normalizeSource(source).split("\n"), 0, ctx);
440
454
  return new Set(ctx.ids.keys());
441
455
  }
442
456
  // Pre-scan for `=== meta` blocks (at any fence depth) and merge their
@@ -468,14 +482,14 @@ function validateRefs(ctx, opts) {
468
482
  if (!ref.doc)
469
483
  continue;
470
484
  if (!opts.resolveDoc) {
471
- ctx.diags.push({ severity: "warning", message: `cross-document reference \`${ref.doc}${ref.anchor ? "#" + ref.anchor : ""}\` not checked (no document resolver)`, line: ref.line });
485
+ ctx.diags.push({ severity: "warning", code: "unchecked-cross-document-reference", message: `cross-document reference \`${ref.doc}${ref.anchor ? "#" + ref.anchor : ""}\` not checked (no document resolver)`, line: ref.line });
472
486
  continue;
473
487
  }
474
488
  let ids = docIds.get(ref.doc);
475
489
  if (ids === undefined) {
476
490
  const src = opts.resolveDoc(ref.doc);
477
491
  if (src === null) {
478
- ctx.diags.push({ severity: "error", message: `cannot resolve document \`${ref.doc}\``, line: ref.line });
492
+ ctx.diags.push({ severity: "error", code: "unresolvable-document", message: `cannot resolve document \`${ref.doc}\``, line: ref.line });
479
493
  docIds.set(ref.doc, new Set());
480
494
  continue;
481
495
  }
@@ -483,14 +497,16 @@ function validateRefs(ctx, opts) {
483
497
  docIds.set(ref.doc, ids);
484
498
  }
485
499
  if (ref.anchor !== undefined && !ids.has(ref.anchor)) {
486
- ctx.diags.push({ severity: "error", message: `unresolved reference \`${ref.doc}#${ref.anchor}\``, line: ref.line });
500
+ ctx.diags.push({ severity: "error", code: "unresolved-cross-document-reference", message: `unresolved reference \`${ref.doc}#${ref.anchor}\``, line: ref.line });
487
501
  }
488
502
  continue;
489
503
  }
490
504
  // internal, autoref, footnote — anchor must be a known id in this document.
491
505
  if (ref.anchor !== undefined && !ctx.ids.has(ref.anchor)) {
492
- const what = ref.kind === "footnote" ? `footnote \`[^${ref.anchor}]\`` : `reference \`#${ref.anchor}\``;
493
- ctx.diags.push({ severity: "error", message: `unresolved ${what}`, line: ref.line });
506
+ const footnote = ref.kind === "footnote";
507
+ const what = footnote ? `footnote \`[^${ref.anchor}]\`` : `reference \`#${ref.anchor}\``;
508
+ const code = footnote ? "unresolved-footnote" : "unresolved-reference";
509
+ ctx.diags.push({ severity: "error", code, message: `unresolved ${what}`, line: ref.line });
494
510
  }
495
511
  }
496
512
  }
@@ -501,13 +517,15 @@ function resolveCharts(ctx) {
501
517
  const ref = typeof block.attrs["data"] === "string" ? block.attrs["data"] : "";
502
518
  const id = ref.replace(/^#/, "");
503
519
  if (id === "") {
504
- ctx.diags.push({ severity: "error", message: "geml-chart: missing `data=#id`", line });
520
+ ctx.diags.push({ severity: "error", code: "chart-missing-data", message: "geml-chart: missing `data=#id`", line });
505
521
  continue;
506
522
  }
507
523
  const table = ctx.tables?.get(id);
508
524
  if (!table) {
509
- const what = ctx.ids.has(id) ? `data target \`#${id}\` is not a table` : `unresolved reference \`#${id}\``;
510
- ctx.diags.push({ severity: "error", message: `geml-chart: ${what}`, line });
525
+ const known = ctx.ids.has(id);
526
+ const what = known ? `data target \`#${id}\` is not a table` : `unresolved reference \`#${id}\``;
527
+ const code = known ? "chart-data-not-a-table" : "unresolved-reference";
528
+ ctx.diags.push({ severity: "error", code, message: `geml-chart: ${what}`, line });
511
529
  continue;
512
530
  }
513
531
  if (table.src !== undefined) {
@@ -524,7 +542,7 @@ function resolveCharts(ctx) {
524
542
  }
525
543
  }
526
544
  export function parse(source, opts = {}) {
527
- const lines = source.replace(/\r\n?/g, "\n").split("\n");
545
+ const lines = normalizeSource(source).split("\n");
528
546
  const ctx = { diags: [], ids: new Map(), refs: [], meta: collectMeta(lines), resolveDoc: opts.resolveDoc };
529
547
  const children = scanBlocks(lines, 0, ctx);
530
548
  resolveCharts(ctx);
@@ -548,7 +566,7 @@ function idOfHeading(braces, text, line, ctx) {
548
566
  function fenceClose(lines, i, open) {
549
567
  const openLen = open[1].length;
550
568
  const id = open[3] ? parseAttrs(open[3]).id : undefined;
551
- const labeled = id !== undefined ? new RegExp(`^={3,}[ \\t]+#${id}[ \\t]*$`) : null;
569
+ const labeled = id !== undefined ? new RegExp(`^={3,}[ \\t]+#${reLit(id)}[ \\t]*$`) : null;
552
570
  for (let j = i + 1; j < lines.length; j++) {
553
571
  if (isCloseFence(lines[j], openLen) || (labeled && labeled.test(lines[j])))
554
572
  return { end: j + 1, closed: true };
@@ -635,7 +653,7 @@ function collectSpans(lines, base, out, ctx, depth = 0) {
635
653
  // with the physical lines produced by splitLines(source).
636
654
  export function blockSpans(source) {
637
655
  const out = new Map();
638
- const lines = source.replace(/\r\n?/g, "\n").split("\n");
656
+ const lines = normalizeSource(source).split("\n");
639
657
  // Inert context: heading auto-ids slug the interpolated text (parser parity);
640
658
  // its diagnostics are discarded — the span scan never reports.
641
659
  const ctx = { diags: [], ids: new Map(), refs: [], meta: collectMeta(lines) };
@@ -719,52 +737,86 @@ function parseStamp(s) {
719
737
  return new Date(Date.UTC(+y, +mo - 1, +d, +h, +mi, +se));
720
738
  }
721
739
  const VERSION = "1.0"; // GEML spec version this CLI targets
722
- const PARSER_VERSION = "1.3.2"; // reference implementation; keep in sync with package.json
723
- const USAGE = `geml — GEML reference CLI
724
-
725
- Usage:
726
- geml <file.geml|-> parse -> document-model JSON (stdout)
727
- geml get <file.geml|-> #id [--json][--head] print ONE block by id (a heading id = its section;
728
- --head narrows any id to its head line; --json = model node)
729
- geml set <file.geml|-> #id [--from f][-o f][--head] replace ONE block by id (new content: --from/stdin)
730
- geml revert <file.geml> #id [--to <sel>][--head] restore ONE block to a past revision (sel: -N|latest|id)
731
- geml check <file.geml|-> [--root d][--json] validate only: diagnostics + exit code
732
- (--root widens cross-doc refs to dir d, e.g. the repo root)
733
- geml render <file.geml|-> [-o out.html] render to one self-contained HTML file
734
- geml fmt <file.geml|-> [-o out.geml] re-serialize to canonical GEML
735
- geml convert <file.md|-> [-o out.geml] Markdown -> GEML
736
- geml export <file.geml|-> [-o out.md] GEML -> Markdown (lossy)
737
- geml history <commit|verify|show|restore|log> <file.geml> [...]
738
- geml codemap <build|verify|render|serve|refresh|find|mcp> [...] code-graph toolkit (alias: codegraph)
739
- geml --help | --version [--json]
740
-
741
- Use '-' as the file to read from stdin.
742
- Exit codes:
743
- 0 ok
744
- 1 document/operation error
745
- 2 command usage error.
740
+ const PARSER_VERSION = "1.4.3"; // reference implementation; keep in sync with package.json
741
+ const USAGE = `geml — GEML reference CLI
742
+
743
+ Usage:
744
+ geml <file.geml|-> [--to <fmt>] [--from <fmt>] [-o out] transform a document (default: --to json)
745
+ --to <output>: json | html | md | geml
746
+ --to md -> Markdown (lossy)
747
+ --to html -> self-contained HTML
748
+ --to geml -> canonical re-format
749
+ --to json -> document-model JSON (default)
750
+ --from <input>: geml | md | json (overrides extension; html is output-only)
751
+ geml notes.md -> GEML (md inferred from extension)
752
+ geml model.json --to geml -> GEML (round-trips a prior --to json)
753
+ geml - --from md read Markdown on stdin
754
+ geml get <file.geml|-> [#id] [--json] [--head] with #id: print that block
755
+ (a heading id = its whole section; --head = head line;
756
+ --json = model node). Without #id: list all addressable
757
+ ids (--json = array).
758
+ geml set <file.geml|-> #id [--head|--body] [--in f[#src]|-] [-o f] replace ONE block by id
759
+ (--in F takes F's block #id, F#src takes #src, else stdin raw;
760
+ default = whole block · --head = head line · --body = body)
761
+ geml add <file.geml|-> (--append | --before #id | --after #id) [--in f[#src]|-] [-o f] insert a fragment
762
+ (1+ blocks and/or prose; content keeps its own ids, a clash is refused)
763
+ geml delete <file.geml|-> #id [#id2 …] [-o f] remove one or more blocks
764
+ (a missing id is skipped; a dangling reference is a warning, not a refusal)
765
+ geml rename <file.geml|-> #old #new [-o f] rename an id and every reference to it (id-boundary safe)
766
+ geml revert <file.geml> #id [--rev <sel>] [--head] undo one block to a past revision (splice / resurrect / remove)
767
+ (sel: 0 | -N | id-prefix | changed; default -1)
768
+ geml check <file.geml|-> [--root d] [--json] validate only: diagnostics + exit code
769
+ (--root widens cross-doc refs to dir d, e.g. the repo root)
770
+ geml history <commit|verify|show|restore|log> <file.geml> [...] .gemlhistory version sidecar
771
+ geml codemap <build|verify|render|serve|refresh|find|mcp> [...] code-graph toolkit (alias: codegraph)
772
+ geml mcp --workspace <dir> [--no-history] serve document CRUD over MCP (stdio)
773
+ (9 tools: list/read/check/history + write/add/delete/rename/revert;
774
+ every write is validated before it reaches disk)
775
+ geml --help | --version [--json]
776
+
777
+ Use '-' as the file to read from stdin.
778
+ Mutations (set/add/delete/rename) write the whole updated document in place for a
779
+ file, or to stdout for '-' input; -o redirects it (-o - = stdout).
780
+ Exit codes:
781
+ 0 ok
782
+ 1 document/operation error
783
+ 2 command usage error.
746
784
  `;
747
785
  // One-line usage for each subcommand — the single source for both the error
748
786
  // shown on misuse and the `<cmd> --help` text.
749
787
  const SUBHELP = {
750
- get: "usage: geml get <file.geml|-> #id [--json] [--head] (a heading id = its whole section; --head narrows any id to its head line)",
751
- set: "usage: geml set <file.geml|-> #id [--from FILE] [-o out.geml] [--head]",
788
+ get: "usage: geml get <file.geml|-> [#id] [--json] [--head] (with #id: that block, a heading id = its whole section, --head = its head line; without #id: list every addressable id, --json = array)",
789
+ set: "usage: geml set <file.geml|-> #id [--head|--body] [--in F | --in F#src | --in -] [-o out.geml] (content: --in F takes F's block #id, --in F#src takes #src, else stdin raw; default = whole block, --head = head line — both normalize the id to #id — --body = body; guarded splice, refused if it breaks the doc)",
790
+ add: "usage: geml add <file.geml|-> (--append | --before #id | --after #id) [--in F | --in F#src | --in -] [-o out.geml] (insert a GEML fragment — 1+ blocks and/or prose — at a position; --in F takes all of F, --in F#src takes #src, else stdin raw; content keeps its own ids, a collision is refused)",
791
+ delete: "usage: geml delete <file.geml|-> #id [#id2 …] [-o out.geml] (remove one or more blocks; a missing id is skipped with a note, not an error; a reference left dangling is a warning, not a refusal — delete never fails on a live reference)",
792
+ rename: "usage: geml rename <file.geml|-> #old #new [-o out.geml] (rewrite an id's declaration AND every reference — [[#id]], [text](#id), chart data=#id, footnote [^id] — id-boundary safe, skipping raw block bodies; #new must be free; refused if it breaks the doc)",
752
793
  check: "usage: geml check <file.geml|-> [--root <dir>] [--json] (--root: resolve cross-doc refs within <dir> instead of the file's own directory)",
753
- render: "usage: geml render <file.geml|-> [-o out.html]",
754
- convert: "usage: geml convert <file.md|-> [-o out.geml]",
755
- export: "usage: geml export <file.geml|-> [-o out.md]",
756
- fmt: "usage: geml fmt <file.geml|-> [-o out.geml]",
757
- revert: "usage: geml revert <file.geml> #id [--to <sel>] [--changed] [--dry-run] [-o out] [--head] (sel: -N | latest | id-prefix; default -1)",
794
+ revert: "usage: geml revert <file.geml> #id [--rev <sel>] [--append|--before #x|--after #x] [--head] [--dry-run] [-o out] (reconcile #id to a revision: splice / resurrect / remove; sel: 0 | -N | id-prefix | changed; default -1)",
758
795
  history: "usage: geml history <commit|verify|show|restore|log> <file.geml> [...]",
759
- codemap: `usage: geml codemap build [--root <repo>] # auto-detect languages, run the indexer(s), and merge into one codemap (--root defaults to the current directory)
760
- geml codemap build (--db <graph.db> | --adapter joern|scip --raw <in>)+ [--root <repo>] [--out .geml-code-graph] [--container module|dir|file] [--lang <JAVASRC|NEWC|…>] [--joern <path>] [--history [-m msg]]
761
- geml codemap verify [dir] geml check + profile reference checks
762
- geml codemap render [dir] every doc -> sibling .html (open index.html from disk)
763
- geml codemap serve [dir] [--port 8140] [--watch] [--background|--stop] live viewer: pages render from .geml on request; --watch re-runs the recipe when sources change
764
- geml codemap refresh [dir] [--force] [--commit] [--background|--hook] re-run the recorded build recipe (_index/refresh.json); --commit lands it as its own commit
765
- geml codemap find <name> [dir] locate a symbol by substring name -> doc#id + src (stdout, no browser)
766
- geml codemap mcp stdio MCP server (GEML_GRAPH_DIR or graph_dir arg)
796
+ codemap: `usage: geml codemap build [--root <repo>] # auto-detect languages, run the indexer(s), and merge into one codemap (--root defaults to the current directory)
797
+ geml codemap build (--db <graph.db> | --adapter joern|scip --raw <in>)+ [--root <repo>] [--out .geml-code-graph] [--container module|dir|file] [--lang <JAVASRC|NEWC|…>] [--joern <path>] [--history [-m msg]]
798
+ geml codemap verify [dir] geml check + profile reference checks
799
+ geml codemap render [dir] every doc -> sibling .html (open index.html from disk)
800
+ geml codemap serve [dir] [--port 8140] [--watch] [--background|--stop] live viewer: pages render from .geml on request; --watch re-runs the recipe when sources change
801
+ geml codemap refresh [dir] [--force] [--commit] [--background|--hook] re-run the recorded build recipe (_index/refresh.json); --commit lands it as its own commit
802
+ geml codemap find <name> [dir] locate a symbol by substring name -> doc#id + src (stdout, no browser)
803
+ geml codemap mcp stdio MCP server (GEML_GRAPH_DIR or graph_dir arg)
767
804
  (<dir> for verify/render/serve/refresh/find defaults to ./.geml-code-graph; codegraph and code-graph are accepted as aliases of codemap)`,
805
+ mcp: `usage: geml mcp --workspace <dir> [--no-history]
806
+
807
+ Serve GEML document CRUD over the MCP stdio transport (JSON-RPC 2.0).
808
+ Nine tools: geml_list_ids · geml_read_block · geml_check · geml_history_log
809
+ geml_write_block · geml_add_block · geml_delete_block
810
+ geml_rename_id · geml_revert_block
811
+
812
+ --workspace <dir> REQUIRED. Root holding the .geml documents. Every path a
813
+ client names is confined here; a client cannot widen it.
814
+ --no-history Skip the .gemlhistory commit taken before each write
815
+ (default: commit, so geml_revert_block always has a
816
+ revision to undo to).
817
+
818
+ Register with a client:
819
+ claude mcp add geml-docs -- geml mcp --workspace /abs/path/to/docs`,
768
820
  };
769
821
  // Set from argv at dispatch time; when true, errors are emitted as a JSON
770
822
  // envelope so an agent that standardizes on --json never has to parse text.
@@ -779,6 +831,19 @@ function fail(msg, code = 2) {
779
831
  console.error(`error: ${msg}`);
780
832
  process.exit(code);
781
833
  }
834
+ // Refuse a mutation whose RESULT would be broken (the pre-write check every
835
+ // mutation runs). Prose mode is the long-standing wording: the first error,
836
+ // phrased by the call site. `--json` additionally carries the FULL diagnostic
837
+ // list with the stable codes of spec Appendix A, so a programmatic caller —
838
+ // `geml mcp` above all — reports what actually broke instead of re-parsing
839
+ // English out of stderr.
840
+ function refuseBroken(prose, errs) {
841
+ if (jsonMode) {
842
+ console.error(JSON.stringify({ error: prose, code: 1, diagnostics: errs }));
843
+ process.exit(1);
844
+ }
845
+ fail(prose, 1);
846
+ }
782
847
  // Read a file, or stdin when the path is "-". On failure emit a clean error.
783
848
  function readInput(file) {
784
849
  try {
@@ -942,10 +1007,10 @@ function runHistory(args) {
942
1007
  console.log(`restored ${file} to ${rev}`);
943
1008
  }
944
1009
  else if (sub === "log") {
945
- // Newest-first, with the `--to` selector for each row in the first column
946
- // (`latest` for the tip, then `-1`, `-2`, …) so the output is copy-paste.
1010
+ // Newest-first, with the `--rev` selector for each row in the first column
1011
+ // (`0` for the tip, then `-1`, `-2`, …) so the output is copy-paste.
947
1012
  for (const r of listRevisions(historyPath)) {
948
- const sel = r.current ? "latest" : `-${r.offset}`;
1013
+ const sel = r.current ? "0" : `-${r.offset}`;
949
1014
  console.log(`${sel.padEnd(7)} ${r.id} ${r.author ?? "-"} ${r.summary ?? ""}`.replace(/\s+$/, ""));
950
1015
  }
951
1016
  }
@@ -957,39 +1022,98 @@ function runHistory(args) {
957
1022
  fail(historyError(e, file, historyPath));
958
1023
  }
959
1024
  }
960
- // `geml convert <file.md|-> [-o out.geml]` — Markdown -> GEML.
961
- function runConvert(args) {
962
- const file = args.find((a) => a === "-" || (!a.startsWith("-") && a !== flag(args, "-o")));
1025
+ function runTransform(argv) {
1026
+ const out = flag(argv, "-o") ?? flag(argv, "--out");
1027
+ const fromRaw = flag(argv, "--from");
1028
+ const toRaw = flag(argv, "--to");
1029
+ const [file] = positionals(argv, ["-o", "--out", "--from", "--to"]);
963
1030
  if (!file)
964
- fail(SUBHELP.convert);
965
- const { geml, notes } = mdToGeml(readInput(file));
966
- for (const n of notes)
967
- console.error(`note: ${n}`);
968
- const outPath = flag(args, "-o") ?? flag(args, "--out");
969
- if (outPath) {
970
- writeFileSync(outPath, geml);
971
- console.error(`wrote ${outPath}`);
1031
+ fail("no input file (use '-' to read from stdin)", 2);
1032
+ // A bare `--to`/`--from` (no following value) is a mistyped flag, not a
1033
+ // silent fall-through to the default — flag() would return undefined and we
1034
+ // must not quietly ignore it.
1035
+ if (argv.includes("--from") && fromRaw === undefined)
1036
+ fail("--from needs a format (geml | md | json)", 2);
1037
+ if (argv.includes("--to") && toRaw === undefined)
1038
+ fail("--to needs a format (json | html | md | geml)", 2);
1039
+ // Input format: an explicit --from wins (for any input, file or stdin), else
1040
+ // the file extension, else GEML (covers .geml, unknown extensions, and stdin).
1041
+ let inFmt;
1042
+ if (fromRaw !== undefined) {
1043
+ if (fromRaw !== "geml" && fromRaw !== "md" && fromRaw !== "json") {
1044
+ fail(`--from: unknown input format '${fromRaw}' (want geml | md | json)`, 2);
1045
+ }
1046
+ inFmt = fromRaw;
1047
+ }
1048
+ else if (/\.(md|markdown)$/i.test(file)) {
1049
+ inFmt = "md";
1050
+ }
1051
+ else if (/\.json$/i.test(file)) {
1052
+ inFmt = "json";
972
1053
  }
973
1054
  else {
974
- process.stdout.write(geml);
1055
+ inFmt = "geml";
975
1056
  }
976
- }
977
- // `geml export <file.geml|-> [-o out.md]` — GEML -> Markdown (lossy). Writes
978
- // the output even with diagnostics, prints any lossy-projection notes, and
979
- // exits non-zero on a parse error same contract as render.
980
- function runExport(args) {
981
- const out = flag(args, "-o") ?? flag(args, "--out");
982
- const file = args.find((a) => a === "-" || (!a.startsWith("-") && a !== out));
983
- if (!file)
984
- fail(SUBHELP.export);
985
- const doc = parse(readInput(file), { resolveDoc: resolverFor(file) });
986
- const { md, notes } = gemlToMd(doc);
987
- if (out) {
988
- writeFileSync(out, md);
989
- console.error(`wrote ${out}`);
1057
+ // Output format: an explicit --to wins, else md input -> geml, geml -> json.
1058
+ let outFmt;
1059
+ if (toRaw !== undefined) {
1060
+ if (toRaw !== "json" && toRaw !== "html" && toRaw !== "md" && toRaw !== "geml") {
1061
+ fail(`--to: unknown output format '${toRaw}' (want json | html | md | geml)`, 2);
1062
+ }
1063
+ outFmt = toRaw;
990
1064
  }
991
- else
992
- process.stdout.write(md);
1065
+ else {
1066
+ outFmt = inFmt === "geml" ? "json" : "geml"; // geml->json; md/json->geml
1067
+ }
1068
+ const src = readInput(file);
1069
+ // md -> geml is a direct projection, not a parse/serialize round-trip: emit
1070
+ // the converter's GEML verbatim (the old `convert`; no diagnostics to raise).
1071
+ if (inFmt === "md" && outFmt === "geml") {
1072
+ const { geml, notes } = mdToGeml(src);
1073
+ writeOut(geml, out);
1074
+ for (const n of notes)
1075
+ console.error(`note: ${n}`);
1076
+ return;
1077
+ }
1078
+ // Otherwise load a document — a md input is converted to GEML first — and
1079
+ // project it to the target.
1080
+ let notes = [];
1081
+ let doc;
1082
+ if (inFmt === "json") {
1083
+ doc = loadModelJson(src, file); // the inverse of `--to json`
1084
+ }
1085
+ else if (inFmt === "md") {
1086
+ const conv = mdToGeml(src);
1087
+ notes = conv.notes;
1088
+ doc = parse(conv.geml, { resolveDoc: resolverFor(file) });
1089
+ }
1090
+ else {
1091
+ doc = parse(src, { resolveDoc: resolverFor(file) });
1092
+ }
1093
+ let output;
1094
+ switch (outFmt) {
1095
+ case "json":
1096
+ output = JSON.stringify(doc, null, 2) + "\n"; // == the former bare parse
1097
+ break;
1098
+ case "geml":
1099
+ output = serialize(doc); // == the former `fmt`
1100
+ break;
1101
+ case "html":
1102
+ output = renderHtml(doc, {
1103
+ source: file === "-" ? "stdin" : basename(file),
1104
+ // geml-code-graph embeds load + parse sibling codemap docs on demand.
1105
+ loadDoc: resolverFor(file),
1106
+ parseDoc: (s) => parse(s),
1107
+ });
1108
+ break;
1109
+ case "md": {
1110
+ const r = gemlToMd(doc); // == the former `export`
1111
+ notes = notes.concat(r.notes);
1112
+ output = r.md;
1113
+ break;
1114
+ }
1115
+ }
1116
+ writeOut(output, out);
993
1117
  for (const n of notes)
994
1118
  console.error(`note: ${n}`);
995
1119
  for (const d of doc.diagnostics)
@@ -997,55 +1121,57 @@ function runExport(args) {
997
1121
  if (doc.diagnostics.some((d) => d.severity === "error"))
998
1122
  process.exit(1);
999
1123
  }
1000
- // `geml render <file.geml> [-o out.html]` GEML -> one self-contained,
1001
- // interactive HTML artifact (the P0 runtime). Writes the file even when there
1002
- // are diagnostics (a viewer should still show what it can), but exits non-zero
1003
- // on any error so CI and agents get a hard signal.
1004
- function runRender(args) {
1005
- const out = flag(args, "-o") ?? flag(args, "--out");
1006
- const file = args.find((a) => a === "-" || (!a.startsWith("-") && a !== out));
1007
- if (!file)
1008
- fail(SUBHELP.render);
1009
- const doc = parse(readInput(file), { resolveDoc: resolverFor(file) });
1010
- const html = renderHtml(doc, {
1011
- source: file === "-" ? "stdin" : basename(file),
1012
- // geml-code-graph embeds load + parse sibling codemap documents on demand.
1013
- loadDoc: resolverFor(file),
1014
- parseDoc: (s) => parse(s),
1015
- });
1016
- if (out) {
1017
- writeFileSync(out, html);
1018
- console.error(`wrote ${out}`);
1124
+ // Load a document-model JSON (the exact output of `--to json`) back into a
1125
+ // Document, so `--from json --to geml` is the inverse of a prior `--to json`.
1126
+ // The model is trusted as-is no re-parse so a clean round-trip is byte-stable
1127
+ // with `--to geml`. Anything that is not a document model is refused, and any
1128
+ // carried diagnostics are preserved (so a broken doc's JSON stays flagged).
1129
+ function loadModelJson(src, file) {
1130
+ let obj;
1131
+ try {
1132
+ obj = JSON.parse(src);
1019
1133
  }
1020
- else
1021
- process.stdout.write(html);
1022
- for (const d of doc.diagnostics)
1023
- console.error(`${d.severity}: ${d.message} (line ${d.line})`);
1024
- if (doc.diagnostics.some((d) => d.severity === "error"))
1025
- process.exit(1);
1134
+ catch (e) {
1135
+ fail(`--from json: ${file === "-" ? "stdin" : file} is not valid JSON (${e.message})`, 1);
1136
+ }
1137
+ const d = obj;
1138
+ if (!d || typeof d !== "object" || d.kind !== "document" || !Array.isArray(d.children)) {
1139
+ fail(`--from json: not a GEML document-model JSON (expected {"kind":"document","children":[…]})`, 1);
1140
+ }
1141
+ const doc = d;
1142
+ if (!Array.isArray(doc.diagnostics))
1143
+ doc.diagnostics = [];
1144
+ return doc;
1026
1145
  }
1027
- // `geml fmt <file.geml> [-o out.geml]` re-serialize the document model into
1028
- // canonical GEML. Because `serialize` is the inverse of `parse`, `fmt` is a
1029
- // pretty-printer whose output parses back to the same model (round-trip stable).
1030
- function runFmt(args) {
1031
- const out = flag(args, "-o") ?? flag(args, "--out");
1032
- const file = args.find((a) => a === "-" || (!a.startsWith("-") && a !== out));
1033
- if (!file)
1034
- fail(SUBHELP.fmt);
1035
- const doc = parse(readInput(file), { resolveDoc: resolverFor(file) });
1036
- const text = serialize(doc);
1146
+ // Write to `-o out` (with a `wrote` note on stderr) or to stdout.
1147
+ function writeOut(text, out) {
1037
1148
  if (out) {
1038
1149
  writeFileSync(out, text);
1039
1150
  console.error(`wrote ${out}`);
1040
1151
  }
1041
1152
  else
1042
1153
  process.stdout.write(text);
1043
- // A broken document must not be reported as a clean format. Surface the
1044
- // diagnostics and exit non-zero, matching parse/render/check.
1045
- for (const d of doc.diagnostics)
1046
- console.error(`${d.severity}: ${d.message} (line ${d.line})`);
1047
- if (doc.diagnostics.some((d) => d.severity === "error"))
1048
- process.exit(1);
1154
+ }
1155
+ // Output-target rule shared by the MUTATION verbs (set, and — soon — add,
1156
+ // delete, rename, revert): a real file input with no `-o` is edited IN PLACE
1157
+ // (it's the obvious target, and it's what lets an agent chain edits without
1158
+ // re-reading a path back out of stdout); stdin (`file === "-"`) has no such
1159
+ // target, so it falls back to stdout. `-o` always wins when given: `-o -`
1160
+ // explicitly requests stdout (even for a file input), `-o <path>` writes
1161
+ // there. Every write announces itself with `wrote <path>` on stderr; stdout
1162
+ // stays reserved for the document bytes so it's still pipeable.
1163
+ function resolveOutTarget(file, oFlag) {
1164
+ const toFile = (path) => ({
1165
+ write(text) { writeFileSync(path, text); console.error(`wrote ${path}`); },
1166
+ });
1167
+ const toStdout = { write(text) { process.stdout.write(text); } };
1168
+ if (oFlag === "-")
1169
+ return toStdout;
1170
+ if (oFlag !== undefined)
1171
+ return toFile(oFlag);
1172
+ if (file === "-")
1173
+ return toStdout;
1174
+ return toFile(file);
1049
1175
  }
1050
1176
  // Positional args (a file, an id) are the non-flag tokens that aren't the value
1051
1177
  // of a value-taking flag. `-` (stdin) is a positional, not a flag. An id may be
@@ -1069,6 +1195,47 @@ function positionals(args, valued) {
1069
1195
  }
1070
1196
  return out;
1071
1197
  }
1198
+ // `geml get <file>` with no id: list every addressable id — the document's
1199
+ // table of contents. Default output is one id per line with its kind (and, for
1200
+ // a heading, its level and text); `--json` is a machine-readable array so an
1201
+ // agent can pick its next `get #id` target. Ids are listed in document order
1202
+ // (the registration order parse() records), covering the same set `get #id`
1203
+ // resolves against: typed blocks, headings, and footnote definitions.
1204
+ function listIds(source, file, json) {
1205
+ const doc = parse(source, { resolveDoc: resolverFor(file) });
1206
+ const rows = doc.ids.map((id) => {
1207
+ const site = findBlockSite(doc.children, id);
1208
+ const b = site?.siblings[site.index];
1209
+ if (b?.kind === "heading")
1210
+ return { id, kind: "heading", level: b.level, text: b.text };
1211
+ if (b?.kind === "block") {
1212
+ const row = { id, kind: b.type };
1213
+ if (b.classes.includes("footnote"))
1214
+ row.footnote = true; // §5.2 footnote definition
1215
+ return row;
1216
+ }
1217
+ return { id, kind: b?.kind ?? "unknown" };
1218
+ });
1219
+ if (json) {
1220
+ console.log(JSON.stringify(rows, null, 2));
1221
+ return;
1222
+ }
1223
+ if (rows.length === 0) {
1224
+ console.error(`no addressable ids in ${file === "-" ? "stdin" : file}`);
1225
+ return;
1226
+ }
1227
+ // Align the id and kind columns; append a heading's level+text or a footnote flag.
1228
+ const idW = Math.max(...rows.map((r) => r.id.length + 1));
1229
+ const kindW = Math.max(...rows.map((r) => r.kind.length));
1230
+ for (const r of rows) {
1231
+ let line = `#${r.id}`.padEnd(idW + 1) + " " + r.kind.padEnd(kindW);
1232
+ if (r.kind === "heading")
1233
+ line += ` h${r.level} ${r.text}`;
1234
+ else if (r.footnote)
1235
+ line += " footnote";
1236
+ console.log(line.replace(/\s+$/, ""));
1237
+ }
1238
+ }
1072
1239
  // `geml get <file.geml|-> #id [--json]` — print ONE block, addressed by id,
1073
1240
  // without loading the rest of the document into context. Default output is the
1074
1241
  // block's exact source bytes: a typed block's full `=== … ===` span, a
@@ -1081,8 +1248,14 @@ function runGet(args) {
1081
1248
  const json = args.includes("--json");
1082
1249
  const headOnly = args.includes("--head");
1083
1250
  const [file, rawId] = positionals(args, []);
1084
- if (!file || !rawId)
1251
+ if (!file)
1085
1252
  fail(SUBHELP.get);
1253
+ // No id: list every addressable id — the document's "table of contents", so
1254
+ // an agent can discover what `get #id` can target without pulling the model.
1255
+ if (!rawId) {
1256
+ listIds(readInput(file), file, json);
1257
+ return;
1258
+ }
1086
1259
  const id = rawId.replace(/^#/, "");
1087
1260
  const source = readInput(file);
1088
1261
  if (json) {
@@ -1118,42 +1291,407 @@ function runGet(args) {
1118
1291
  const span = headOnly ? narrowToHead(found) : found;
1119
1292
  process.stdout.write(splitLines(source).slice(span.start, span.end).join(""));
1120
1293
  }
1121
- // `geml set <file.geml|-> #id [--from FILE] [-o out]` replace ONLY that
1122
- // block's source span with new content (from --from or stdin), preserving every
1123
- // other byte. Prints the full updated document, or writes in place with -o. The
1124
- // splice is re-parsed and rejected if it broke the doc: `set` never writes a
1125
- // corrupt file.
1294
+ const NO_CONTENT = "no replacement content (use --in FILE or pipe it on stdin)";
1295
+ // `geml set <file.geml|-> #id [--head|--body] [--in F|F#src|-] [-o out]`
1296
+ // replace ONE existing block, addressed by #id, with new content, preserving
1297
+ // every other byte. Two content CHANNELS × three MODES:
1298
+ //
1299
+ // channels · `--in F[#src]` extracts a BLOCK from GEML file F (F is always
1300
+ // read as GEML — extension ignored, no md conversion): `--in F`
1301
+ // takes the block whose id == the target #id; `--in F#src` takes
1302
+ // #src. stdin (default, or `--in -`) is raw bytes.
1303
+ // modes · default replaces the WHOLE block, `--head` only the head line,
1304
+ // `--body` only the body. Default and `--head` NORMALIZE the
1305
+ // content's id to #id (its source id is irrelevant); `--body`
1306
+ // keeps the target's head verbatim, so #id is preserved naturally.
1307
+ //
1308
+ // Output follows resolveOutTarget (file -> in place, stdin -> stdout, `-o`/`-o -`
1309
+ // override) and every splice is guarded — re-parsed and rejected if it broke
1310
+ // the doc, so `set` never writes a corrupt file.
1126
1311
  function runSet(args) {
1127
1312
  const out = flag(args, "-o") ?? flag(args, "--out");
1128
- const from = flag(args, "--from");
1313
+ const from = flag(args, "--in");
1129
1314
  const headOnly = args.includes("--head");
1130
- const [file, rawId] = positionals(args, ["-o", "--out", "--from"]);
1131
- if (!file || !rawId)
1315
+ const bodyOnly = args.includes("--body");
1316
+ if (headOnly && bodyOnly)
1317
+ fail("--head and --body are mutually exclusive", 2);
1318
+ const [file, rawId] = positionals(args, ["-o", "--out", "--in"]);
1319
+ if (!file)
1132
1320
  fail(SUBHELP.set);
1321
+ // No id: there is no block to replace. Point the way to discovery, not a bare
1322
+ // usage line — `geml get <file>` lists every id `set` can target.
1323
+ if (!rawId)
1324
+ fail(`no #id given — run 'geml get ${file === "-" ? "<file>" : file}' to list addressable ids`, 2);
1133
1325
  const id = rawId.replace(/^#/, "");
1134
- // Both the document and the replacement can't come from stdin. Reject that up
1135
- // front before consuming stdin so the document read below is unambiguous.
1136
- if (file === "-" && from === undefined) {
1137
- fail("reading the document from stdin needs --from for the new content", 2);
1326
+ // The raw channel is stdin `--in` omitted or `--in -`; anything else sources
1327
+ // a block from a file. Document and content can't BOTH be stdin: reject that
1328
+ // up front, before consuming stdin, so the document read below is unambiguous.
1329
+ const rawChannel = from === undefined || from === "-";
1330
+ if (file === "-" && rawChannel) {
1331
+ fail("reading the document from stdin needs --in for the new content", 2);
1138
1332
  }
1139
1333
  const source = readInput(file);
1140
- // New content: an explicit --from file, else stdin.
1141
- let replacement;
1142
- if (from !== undefined) {
1143
- replacement = readInput(from);
1334
+ if (bodyOnly) {
1335
+ runSetBody(source, id, from, rawChannel, file, out);
1336
+ return;
1337
+ }
1338
+ // default / --head: content is a whole block (default) or a bare head line.
1339
+ let content;
1340
+ if (rawChannel) {
1341
+ content = readInput("-");
1342
+ if (content === "")
1343
+ fail(NO_CONTENT, 1);
1344
+ // Default mode wants exactly ONE block. Pure prose has no head to carry the
1345
+ // id (steer to --body); multiple blocks are `add`'s job. --head takes a
1346
+ // lone head line, so it skips the whole-block shape check.
1347
+ if (!headOnly) {
1348
+ const shape = contentShape(content);
1349
+ if (shape === "empty")
1350
+ fail(NO_CONTENT, 1);
1351
+ if (shape === "prose")
1352
+ fail(`content is prose, not a block — use --body to set the body of #${id}`, 1);
1353
+ if (shape === "multi")
1354
+ fail("set replaces ONE block, but the content has multiple blocks (use add)", 1);
1355
+ }
1144
1356
  }
1145
1357
  else {
1146
- replacement = readInput("-");
1147
- if (replacement === "")
1148
- fail("no replacement content (use --from FILE or pipe it on stdin)", 1);
1358
+ content = extractBlock(from, id, headOnly ? "head" : "whole");
1149
1359
  }
1150
- const updated = spliceBlock(source, id, replacement, file, headOnly);
1151
- if (out) {
1152
- writeFileSync(out, updated);
1153
- console.error(`wrote ${out}`);
1360
+ const normalized = normalizeBlockId(content, id);
1361
+ const updated = spliceBlock(source, id, normalized, file, headOnly);
1362
+ resolveOutTarget(file, out).write(updated);
1363
+ }
1364
+ // `--body`: swap ONLY the target block's body, keeping its head (and #id) and,
1365
+ // for a typed block, its close fence. Assembles head + new body + close and
1366
+ // reuses the guarded spliceBlock — the head carries #id, so the id survives
1367
+ // with no normalization needed.
1368
+ function runSetBody(source, id, from, rawChannel, file, out) {
1369
+ const found = blockSpans(source).get(id);
1370
+ if (!found)
1371
+ fail(`no block with id \`${id}\``, 1);
1372
+ const lines = splitLines(source);
1373
+ const headLine = lines[found.start] ?? "";
1374
+ const headText = stripEol(headLine);
1375
+ // A typed block keeps its closing fence; a heading section has none.
1376
+ let closeLine = null;
1377
+ const open = FENCE_OPEN.exec(headText);
1378
+ if (open) {
1379
+ const lastText = stripEol(lines[found.end - 1] ?? "").replace(/[ \t]+$/, "");
1380
+ const bid = open[3] ? parseAttrs(open[3]).id : undefined;
1381
+ const labeled = bid !== undefined && new RegExp(`^={3,}[ \\t]+#${reLit(bid)}[ \\t]*$`).test(lastText);
1382
+ if (isCloseFence(lastText, open[1].length) || labeled)
1383
+ closeLine = lines[found.end - 1] ?? "";
1384
+ }
1385
+ let body;
1386
+ if (rawChannel) {
1387
+ body = readInput("-");
1388
+ if (body === "")
1389
+ fail(NO_CONTENT, 1);
1390
+ }
1391
+ else {
1392
+ body = extractBlock(from, id, "body");
1154
1393
  }
1394
+ let head = headLine;
1395
+ if (head !== "" && !/(\r\n|\r|\n)$/.test(head))
1396
+ head += "\n";
1397
+ let b = body.replace(/\r\n?/g, "\n");
1398
+ if (closeLine !== null && b !== "" && !b.endsWith("\n"))
1399
+ b += "\n";
1400
+ const replacement = closeLine !== null ? head + b + closeLine : head + b;
1401
+ // A typed block (closeLine !== null) must stay ONE block: enforce the
1402
+ // block-count invariant so a `===` fence in the raw body can't close it early
1403
+ // and inject siblings (SEC F2). A heading section body has no close fence and
1404
+ // may legitimately contain blocks, so it is not count-guarded.
1405
+ const updated = spliceBlock(source, id, replacement, file, false, closeLine !== null);
1406
+ resolveOutTarget(file, out).write(updated);
1407
+ }
1408
+ // `geml add <file|-> (--append | --before #x | --after #x) [--in F|F#src|-] [-o]`
1409
+ // — insert a GEML fragment (1+ blocks and/or prose) at a position. Unlike `set`,
1410
+ // `add` names no target id, so content keeps its OWN ids (no normalization); an
1411
+ // id colliding with the document (or duplicated within the fragment) makes the
1412
+ // re-parse fail and nothing is written. Bare prose is a valid fragment.
1413
+ function runAdd(args) {
1414
+ const out = flag(args, "-o") ?? flag(args, "--out");
1415
+ const from = flag(args, "--in");
1416
+ const before = flag(args, "--before");
1417
+ const after = flag(args, "--after");
1418
+ const append = args.includes("--append");
1419
+ const posCount = (append ? 1 : 0) + (before !== undefined ? 1 : 0) + (after !== undefined ? 1 : 0);
1420
+ if (posCount !== 1)
1421
+ fail("add needs exactly one position: --append | --before #id | --after #id", 2);
1422
+ const [file] = positionals(args, ["-o", "--out", "--in", "--before", "--after"]);
1423
+ if (!file)
1424
+ fail(SUBHELP.add);
1425
+ const rawChannel = from === undefined || from === "-";
1426
+ if (file === "-" && rawChannel)
1427
+ fail("reading the document from stdin needs --in for the new content", 2);
1428
+ const source = readInput(file);
1429
+ // Content: --in F#src -> block #src; --in F -> all of F (a multi-block
1430
+ // fragment is fine here); stdin -> raw. No id-normalization: add keeps ids.
1431
+ let content;
1432
+ if (rawChannel)
1433
+ content = readInput("-");
1434
+ else if (from.includes("#"))
1435
+ content = extractBlock(from, "", "whole");
1155
1436
  else
1156
- process.stdout.write(updated);
1437
+ content = readInput(from);
1438
+ if (content.trim() === "")
1439
+ fail("no content to add (use --in FILE or pipe it on stdin)", 1);
1440
+ // Resolve the physical-line insertion point.
1441
+ const lines = splitLines(source);
1442
+ let at;
1443
+ if (append) {
1444
+ at = lines.length;
1445
+ }
1446
+ else {
1447
+ const anchorId = (before ?? after).replace(/^#/, "");
1448
+ const span = blockSpans(source).get(anchorId);
1449
+ if (!span)
1450
+ fail(`no block with id \`${anchorId}\` in ${file === "-" ? "stdin" : file}`, 1);
1451
+ at = before !== undefined ? span.start : span.end;
1452
+ }
1453
+ const updated = insertFragment(source, lines, at, content, file);
1454
+ resolveOutTarget(file, out).write(updated);
1455
+ }
1456
+ // Splice `fragment` into `source` at physical-line index `at` (splitLines
1457
+ // coords), separating it from adjacent content with a single blank line so
1458
+ // blocks don't fuse, then GUARD: the re-parse must be error-free (a colliding
1459
+ // or duplicate id surfaces as an error diagnostic) and no pre-existing id may
1460
+ // vanish. Returns the updated text; on any violation fail()s and writes nothing.
1461
+ function insertFragment(source, lines, at, fragment, file) {
1462
+ const beforeIds = parse(source, { resolveDoc: resolverFor(file) }).ids;
1463
+ const before = lines.slice(0, at);
1464
+ const after = lines.slice(at);
1465
+ // The preceding line must end in a newline so the fragment starts on its own.
1466
+ if (before.length && !/(\r\n|\r|\n)$/.test(before[before.length - 1])) {
1467
+ before[before.length - 1] += "\n";
1468
+ }
1469
+ let frag = fragment.replace(/\r\n?/g, "\n");
1470
+ if (!frag.endsWith("\n"))
1471
+ frag += "\n";
1472
+ // A single blank separator on each side that has adjacent content and isn't
1473
+ // already blank — keeps a following head / preceding block from fusing.
1474
+ const blank = (s) => stripEol(s).trim() === "";
1475
+ const sepBefore = before.length && !blank(before[before.length - 1]) ? "\n" : "";
1476
+ const sepAfter = after.length && !blank(after[0]) ? "\n" : "";
1477
+ const updated = before.join("") + sepBefore + frag + sepAfter + after.join("");
1478
+ const reparsed = parse(updated, { resolveDoc: resolverFor(file) });
1479
+ const errs = reparsed.diagnostics.filter((d) => d.severity === "error");
1480
+ if (errs.length) {
1481
+ const first = errs[0];
1482
+ refuseBroken(`adding the content would break the document: ${first.message} (line ${first.line}); not written`, errs);
1483
+ }
1484
+ const now = new Set(reparsed.ids);
1485
+ const dropped = beforeIds.find((x) => !now.has(x));
1486
+ if (dropped !== undefined)
1487
+ fail(`adding the content would drop block \`#${dropped}\`; not written`, 1);
1488
+ return updated;
1489
+ }
1490
+ // `geml delete <file|-> #id [#id2 …] [-o]` — remove one or more blocks. A
1491
+ // missing id is SKIPPED with a note (declarative "ensure absent", not an
1492
+ // error). Unlike set/add, delete's write is LENIENT: removing a complete block
1493
+ // can't break the parse structurally, but it may leave a reference dangling —
1494
+ // that is a WARNING, never a refusal (delete is reversible via revert + history,
1495
+ // and `geml check` still flags the dangling ref afterward). Contained/overlapping
1496
+ // spans (a nested block inside a deleted heading section) are handled by deleting
1497
+ // the UNION of target lines, so a line is never spliced twice.
1498
+ function runDelete(args) {
1499
+ const out = flag(args, "-o") ?? flag(args, "--out");
1500
+ const pos = positionals(args, ["-o", "--out"]);
1501
+ const file = pos[0];
1502
+ if (!file)
1503
+ fail(SUBHELP.delete);
1504
+ const ids = pos.slice(1).map((s) => s.replace(/^#/, ""));
1505
+ if (ids.length === 0)
1506
+ fail("delete needs at least one #id (run 'geml get <file>' to list ids)", 2);
1507
+ const source = readInput(file);
1508
+ const spans = blockSpans(source);
1509
+ const toDelete = new Set();
1510
+ let found = 0;
1511
+ for (const id of ids) {
1512
+ const span = spans.get(id);
1513
+ if (!span) {
1514
+ console.error(`skipped #${id}: no such block`);
1515
+ continue;
1516
+ }
1517
+ found++;
1518
+ for (let i = span.start; i < span.end; i++)
1519
+ toDelete.add(i);
1520
+ }
1521
+ if (found === 0) {
1522
+ resolveOutTarget(file, out).write(source);
1523
+ return;
1524
+ } // nothing to remove
1525
+ const updated = splitLines(source).filter((_, i) => !toDelete.has(i)).join("");
1526
+ // Lenient guard: surface any resulting error diagnostic (a reference now
1527
+ // dangling) as a WARNING, but write regardless.
1528
+ const reparsed = parse(updated, { resolveDoc: resolverFor(file) });
1529
+ for (const d of reparsed.diagnostics.filter((x) => x.severity === "error")) {
1530
+ console.error(`warning: ${d.message} (line ${d.line}) — left dangling by delete; run 'geml check' to see it as an error`);
1531
+ }
1532
+ resolveOutTarget(file, out).write(updated);
1533
+ }
1534
+ // `geml rename <file|-> #old #new [-o]` — the one verb that reaches OUTSIDE a
1535
+ // block: it rewrites #old's declaration AND every reference to it. #new must be
1536
+ // free; the guarded re-parse refuses anything that would break the doc.
1537
+ function runRename(args) {
1538
+ const out = flag(args, "-o") ?? flag(args, "--out");
1539
+ const [file, rawOld, rawNew] = positionals(args, ["-o", "--out"]);
1540
+ if (!file || !rawOld || !rawNew)
1541
+ fail(SUBHELP.rename);
1542
+ const oldId = rawOld.replace(/^#/, "");
1543
+ const newId = rawNew.replace(/^#/, "");
1544
+ if (oldId === newId)
1545
+ fail("#old and #new are the same id — nothing to rename", 2);
1546
+ const source = readInput(file);
1547
+ const before = parse(source, { resolveDoc: resolverFor(file) });
1548
+ if (!before.ids.includes(oldId))
1549
+ fail(`no block with id \`${oldId}\``, 1);
1550
+ if (before.ids.includes(newId))
1551
+ fail(`id \`${newId}\` already exists; not written`, 1);
1552
+ // Renaming an id that has recorded history breaks the revert-lineage for it
1553
+ // (revert keys by id and can't follow #old -> #new across the boundary). Warn
1554
+ // so the user knows a later `revert #new` won't reach pre-rename revisions.
1555
+ if (file !== "-") {
1556
+ const hp = historyPathFor(file);
1557
+ if (existsSync(hp)) {
1558
+ try {
1559
+ if (blockSpans(resolveContent(hp, "0").text).has(oldId)) {
1560
+ console.error(`warning: #${oldId} has history; revert across this rename is not tracked — see docs`);
1561
+ }
1562
+ }
1563
+ catch { /* unreadable/empty history: no warning */ }
1564
+ }
1565
+ }
1566
+ const updated = rewriteId(source, oldId, newId, file);
1567
+ const reparsed = parse(updated, { resolveDoc: resolverFor(file) });
1568
+ const errs = reparsed.diagnostics.filter((d) => d.severity === "error");
1569
+ if (errs.length) {
1570
+ const e = errs[0];
1571
+ refuseBroken(`rename would break the document: ${e.message} (line ${e.line}); not written`, errs);
1572
+ }
1573
+ if (!reparsed.ids.includes(newId))
1574
+ fail(`rename did not produce #${newId}; not written`, 1);
1575
+ if (reparsed.ids.includes(oldId))
1576
+ fail(`#${oldId} still present after rename; not written`, 1);
1577
+ // Every OTHER id must be untouched. The `#old` match boundary treats a char
1578
+ // outside [A-Za-z0-9_-] as an id terminator, but ids may contain e.g. `.`
1579
+ // (`#foo.bar`), so renaming `#foo` could silently rewrite the *different* id
1580
+ // `#foo.bar` -> `#baz.bar`. Reject when the set of ids other than the rename
1581
+ // pair changed at all (SEC/correctness: collateral id corruption).
1582
+ const othersBefore = before.ids.filter((id) => id !== oldId).sort().join("\n");
1583
+ const othersAfter = reparsed.ids.filter((id) => id !== newId).sort().join("\n");
1584
+ if (othersBefore !== othersAfter) {
1585
+ fail(`rename would also change other ids sharing the \`${oldId}\` prefix (e.g. \`#${oldId}…\`); not written`, 1);
1586
+ }
1587
+ resolveOutTarget(file, out).write(updated);
1588
+ }
1589
+ // Rewrite id `old` -> `new` everywhere it is a declaration or reference, id-
1590
+ // boundary-safe: `#old` is replaced only when NOT followed by an id char, so a
1591
+ // longer id like `#old2` / `#old-x` is untouched. Covers the declaration
1592
+ // (`{#old …}`, labeled close `=== #old`), block references (`[[#old]]`,
1593
+ // `[t](#old)`, chart `data=#old`) and footnotes (`[^old]`). RAW / data block
1594
+ // BODIES (code/diagram/math/table/meta) are skipped — a `#old` there is literal
1595
+ // text, not a reference. (Known residual: id-less raw bodies and inline
1596
+ // code/math spans in flow content — see design §8.)
1597
+ function rewriteId(source, oldId, newId, file) {
1598
+ const doc = parse(source, { resolveDoc: resolverFor(file) });
1599
+ const spans = blockSpans(source);
1600
+ const protectedLines = new Set();
1601
+ for (const b of doc.children) {
1602
+ if (b.kind === "block" && (b.mode === "raw" || b.mode === "data") && b.id) {
1603
+ const span = spans.get(b.id);
1604
+ if (span) {
1605
+ const br = bodyRange(source, span);
1606
+ for (let i = br.start; i < br.end; i++)
1607
+ protectedLines.add(i);
1608
+ }
1609
+ }
1610
+ }
1611
+ const esc = reLit(oldId);
1612
+ const hashRe = new RegExp(`#${esc}(?![A-Za-z0-9_-])`, "g");
1613
+ const fnRe = new RegExp(`(\\[\\^)${esc}(?![A-Za-z0-9_-])`, "g");
1614
+ const lines = splitLines(source);
1615
+ for (let i = 0; i < lines.length; i++) {
1616
+ if (protectedLines.has(i))
1617
+ continue;
1618
+ lines[i] = lines[i].replace(hashRe, `#${newId}`).replace(fnRe, `$1${newId}`);
1619
+ }
1620
+ return lines.join("");
1621
+ }
1622
+ // Extract one block from a GEML file for `--in`. `spec` is `F` (block whose id
1623
+ // == the target) or `F#src` (block #src) — the last `#` splits path from id, so
1624
+ // a `#` inside the path is tolerated; F is read as GEML regardless of extension
1625
+ // (blockSpans + splitLines, no parse — same slice `geml get` prints). `part`
1626
+ // selects the whole span, its head line, or its body. A missing file or absent
1627
+ // id is an operation error (exit 1); the caller writes nothing.
1628
+ function extractBlock(spec, targetId, part) {
1629
+ const hash = spec.lastIndexOf("#");
1630
+ const fragFile = hash >= 0 ? spec.slice(0, hash) : spec;
1631
+ const fragId = hash >= 0 ? spec.slice(hash + 1).replace(/^#/, "") : targetId;
1632
+ let text;
1633
+ try {
1634
+ text = readFileSync(fragFile, "utf8");
1635
+ }
1636
+ catch {
1637
+ fail(`cannot read ${fragFile}`, 1);
1638
+ }
1639
+ const span = blockSpans(text).get(fragId);
1640
+ if (!span)
1641
+ fail(`no block with id \`${fragId}\` in ${fragFile}`, 1);
1642
+ const lines = splitLines(text);
1643
+ if (part === "head")
1644
+ return lines.slice(span.start, span.start + 1).join("");
1645
+ if (part === "body") {
1646
+ const b = bodyRange(text, span);
1647
+ return lines.slice(b.start, b.end).join("");
1648
+ }
1649
+ return lines.slice(span.start, span.end).join("");
1650
+ }
1651
+ // Strip a single trailing terminator (`\r\n`, `\r`, or `\n`) from one line.
1652
+ function stripEol(line) {
1653
+ return line.replace(/(\r\n|\r|\n)$/, "");
1654
+ }
1655
+ // The body sub-range of a block span: [head+1, close) for a closed typed block,
1656
+ // otherwise [head+1, end) — a heading section (no close fence) or an
1657
+ // unterminated block whose span already runs to end-of-scope.
1658
+ function bodyRange(text, span) {
1659
+ const lines = splitLines(text);
1660
+ const open = FENCE_OPEN.exec(stripEol(lines[span.start] ?? ""));
1661
+ if (open) {
1662
+ const lastText = stripEol(lines[span.end - 1] ?? "").replace(/[ \t]+$/, "");
1663
+ const bid = open[3] ? parseAttrs(open[3]).id : undefined;
1664
+ const labeled = bid !== undefined && new RegExp(`^={3,}[ \\t]+#${reLit(bid)}[ \\t]*$`).test(lastText);
1665
+ const closed = isCloseFence(lastText, open[1].length) || labeled;
1666
+ return { start: span.start + 1, end: closed ? span.end - 1 : span.end };
1667
+ }
1668
+ return { start: span.start + 1, end: span.end };
1669
+ }
1670
+ // The shape of default-mode stdin content, section-aware: a heading OWNS its
1671
+ // section (`# H …blocks…` is ONE unit, not many), matching sectionEnd/blockSpans.
1672
+ // Used to reject pure prose (-> --body) and multi-block content (-> add) before
1673
+ // the splice — extraction via --in is inherently one block and skips this.
1674
+ function contentShape(content) {
1675
+ const bs = parse(content).children;
1676
+ let blockUnits = 0, proseUnits = 0, i = 0;
1677
+ while (i < bs.length) {
1678
+ const b = bs[i];
1679
+ if (b.kind === "heading") {
1680
+ i = sectionEndIndex(bs, i);
1681
+ blockUnits++;
1682
+ }
1683
+ else if (b.kind === "block") {
1684
+ i++;
1685
+ blockUnits++;
1686
+ }
1687
+ else {
1688
+ i++;
1689
+ proseUnits++;
1690
+ }
1691
+ }
1692
+ if (blockUnits === 0)
1693
+ return proseUnits === 0 ? "empty" : "prose";
1694
+ return blockUnits + proseUnits === 1 ? "single" : "multi";
1157
1695
  }
1158
1696
  // Replace block #id's source span in `source` with `replacement`, preserving
1159
1697
  // every other byte, and GUARD the result: the re-parse must be error-free, #id
@@ -1161,11 +1699,12 @@ function runSet(args) {
1161
1699
  // can silently swallow a neighbour). Returns the updated document text; on any
1162
1700
  // violation it calls fail() and never returns a corrupt document. Shared by
1163
1701
  // `set` and `revert`.
1164
- function spliceBlock(source, id, replacement, file, headOnly = false) {
1702
+ function spliceBlock(source, id, replacement, file, headOnly = false, guardCount = false) {
1165
1703
  const found = blockSpans(source).get(id);
1166
1704
  if (!found)
1167
1705
  fail(`no block with id \`${id}\``, 1);
1168
- const beforeIds = parse(source, { resolveDoc: resolverFor(file) }).ids;
1706
+ const beforeDoc = parse(source, { resolveDoc: resolverFor(file) });
1707
+ const beforeIds = beforeDoc.ids;
1169
1708
  // Keep the bytes before and after the target span exactly; give the new block
1170
1709
  // a single trailing newline so the following block still starts on its own
1171
1710
  // line (unless it is the file's last line, which may legitimately lack one).
@@ -1191,7 +1730,7 @@ function spliceBlock(source, id, replacement, file, headOnly = false) {
1191
1730
  const errs = reparsed.diagnostics.filter((d) => d.severity === "error");
1192
1731
  if (errs.length) {
1193
1732
  const first = errs[0];
1194
- fail(`replacement would break the document: ${first.message} (line ${first.line}); not written`, 1);
1733
+ refuseBroken(`replacement would break the document: ${first.message} (line ${first.line}); not written`, errs);
1195
1734
  }
1196
1735
  const now = new Set(reparsed.ids);
1197
1736
  if (!now.has(id))
@@ -1200,22 +1739,46 @@ function spliceBlock(source, id, replacement, file, headOnly = false) {
1200
1739
  if (dropped !== undefined) {
1201
1740
  fail(`replacement would drop block \`#${dropped}\` (malformed content?); not written`, 1);
1202
1741
  }
1742
+ // For a typed block with a close fence, the body is opaque and swapping it
1743
+ // keeps exactly ONE block. A raw `--body` can embed a `===` fence of the
1744
+ // block's length that closes the target early and turns the remainder — plus
1745
+ // the close line we re-appended — into NEW sibling blocks, including an id-less
1746
+ // `=== meta` that redefines document metadata (the dropped-id check above
1747
+ // cannot see an id-less injection). Guarded callers refuse any count change.
1748
+ // (Not enforced for heading sections / whole-block set, whose replacement may
1749
+ // legitimately span several top-level blocks.)
1750
+ if (guardCount && reparsed.children.length !== beforeDoc.children.length) {
1751
+ fail(`replacement changes the block count (a fence in the body closed #${id} early and injected sibling block(s)?); not written`, 1);
1752
+ }
1203
1753
  return updated;
1204
1754
  }
1205
- // `geml revert <file.geml> #id [--to <sel>] [--changed] [--dry-run] [-o out] [--history PATH]`
1755
+ // `geml revert <file.geml> #id [--rev <sel>] [--dry-run] [-o out] [--history PATH]`
1206
1756
  // Restore ONE block to a past revision's version — a targeted, guarded splice
1207
- // that leaves the rest of the document untouched. <sel> (default `-1`): `-N` (N
1208
- // revisions back from current), `latest`, or an id prefix/suffix. `--changed`
1209
- // skips revisions that never touched the block, landing on its previous
1210
- // *distinct* version. `--dry-run` prints what would be spliced in, writing
1211
- // nothing. Writes in place by default (revert is a mutation); `-o` redirects.
1757
+ // that leaves the rest of the document untouched. <sel> (default `-1`): `0` (the
1758
+ // tip), `-N` (N revisions back), an id prefix/suffix, or `changed` — a content
1759
+ // selector that skips revisions which never touched the block, landing on its
1760
+ // previous *distinct* version. `--dry-run` prints what would be spliced in,
1761
+ // writing nothing. Writes in place by default (revert is a mutation); `-o` redirects.
1212
1762
  function runRevert(args) {
1213
- const changed = args.includes("--changed");
1214
1763
  const dryRun = args.includes("--dry-run");
1215
1764
  const headOnly = args.includes("--head");
1216
1765
  const out = flag(args, "-o") ?? flag(args, "--out");
1217
- const to = flag(args, "--to") ?? "-1";
1218
- const [file, rawId] = positionals(args, ["--to", "--history", "-o", "--out"]);
1766
+ const to = flag(args, "--rev") ?? "-1";
1767
+ // `--rev changed` is a CONTENT selector, not a position: skip commits that
1768
+ // never touched this block, landing on its previous *distinct* version. It is
1769
+ // just a `--rev` value, so it cannot conflict with a positional `-N`.
1770
+ const changed = to === "changed";
1771
+ // The former standalone `--changed` flag is now this value; refuse the old
1772
+ // spelling loudly rather than silently ignoring it (and reverting to -1).
1773
+ if (args.includes("--changed"))
1774
+ fail("--changed is now `--rev changed`", 2);
1775
+ const before = flag(args, "--before");
1776
+ const after = flag(args, "--after");
1777
+ const append = args.includes("--append");
1778
+ if ((append ? 1 : 0) + (before !== undefined ? 1 : 0) + (after !== undefined ? 1 : 0) > 1) {
1779
+ fail("revert takes at most one position: --append | --before #id | --after #id", 2);
1780
+ }
1781
+ const [file, rawId] = positionals(args, ["--rev", "--history", "-o", "--out", "--before", "--after"]);
1219
1782
  if (!file || !rawId)
1220
1783
  fail(SUBHELP.revert);
1221
1784
  if (file === "-")
@@ -1223,13 +1786,13 @@ function runRevert(args) {
1223
1786
  const id = rawId.replace(/^#/, "");
1224
1787
  const historyPath = flag(args, "--history") ?? historyPathFor(file);
1225
1788
  const source = readInput(file);
1226
- const found = blockSpans(source).get(id);
1227
- if (!found)
1228
- fail(`no block with id \`${id}\` in ${file}`, 1);
1229
- const curSpan = headOnly ? narrowToHead(found) : found;
1230
- const curBlock = splitLines(source).slice(curSpan.start, curSpan.end).join("");
1231
- // Extract block #id's source from a reconstructed revision (undefined if the
1232
- // block did not exist there). Under `--head`, extract only the head line.
1789
+ const curFull = blockSpans(source).get(id); // undefined => absent now
1790
+ const curBlock = curFull === undefined ? undefined : (() => {
1791
+ const span = headOnly ? narrowToHead(curFull) : curFull;
1792
+ return splitLines(source).slice(span.start, span.end).join("");
1793
+ })();
1794
+ // Extract #id's block from a reconstructed revision (undefined => absent
1795
+ // there). Under `--head`, extract only the head line.
1233
1796
  const pick = (text) => {
1234
1797
  const s = blockSpans(text).get(id);
1235
1798
  if (!s)
@@ -1241,7 +1804,7 @@ function runRevert(args) {
1241
1804
  const target = (() => {
1242
1805
  try {
1243
1806
  if (changed) {
1244
- const found = firstChangedContent(historyPath, curBlock, pick);
1807
+ const found = firstChangedContent(historyPath, curBlock ?? "", pick);
1245
1808
  if (!found)
1246
1809
  fail(`no earlier revision changes \`${id}\``, 1);
1247
1810
  return found;
@@ -1252,22 +1815,142 @@ function runRevert(args) {
1252
1815
  fail(historyError(e, file, historyPath), 1);
1253
1816
  }
1254
1817
  })();
1255
- const oldBlock = pick(target.text);
1256
- if (oldBlock === undefined)
1257
- fail(`block \`${id}\` does not exist at revision ${target.id}`, 1);
1258
- if (oldBlock === curBlock) {
1259
- console.error(`#${id} is unchanged at ${target.id}; nothing to revert${changed ? "" : " (try --to -2, or --changed)"}`);
1818
+ const oldBlock = pick(target.text); // undefined => absent at R
1819
+ // Common write path (bespoke message; -o path redirects; -o - -> stdout).
1820
+ const emit = (updated, verb) => {
1821
+ const dest = out ?? file;
1822
+ if (dest === "-")
1823
+ process.stdout.write(updated);
1824
+ else
1825
+ writeFileSync(dest, updated);
1826
+ console.error(`${verb}${dest === file ? "" : dest === "-" ? " -> stdout" : ` -> ${dest}`}`);
1827
+ };
1828
+ // Reconcile #id between now and revision R across the four presence cells.
1829
+ if (curBlock === undefined && oldBlock === undefined) {
1830
+ fail(`\`${id}\` exists in neither the document nor ${target.id} (try --rev changed)`, 1);
1831
+ }
1832
+ // both present -> SPLICE (undo set)
1833
+ if (curBlock !== undefined && oldBlock !== undefined) {
1834
+ if (oldBlock === curBlock) {
1835
+ console.error(`#${id} is unchanged at ${target.id}; nothing to revert${changed ? "" : " (try --rev -2, or --rev changed)"}`);
1836
+ // A no-op still has to PRODUCE the document when an output destination was
1837
+ // asked for: `-o` means "write the result somewhere", and the result of a
1838
+ // no-op revert is the unchanged document. Returning silently here left
1839
+ // `-o -` consumers with exit 0 and empty stdout, which reads as "success,
1840
+ // and the document is now empty".
1841
+ if (out !== undefined)
1842
+ emit(source, `#${id} unchanged`);
1843
+ return;
1844
+ }
1845
+ if (dryRun) {
1846
+ console.error(`would revert #${id} to ${target.id}:`);
1847
+ process.stdout.write(oldBlock.endsWith("\n") ? oldBlock : oldBlock + "\n");
1848
+ return;
1849
+ }
1850
+ emit(spliceBlock(source, id, oldBlock, file, headOnly), `reverted #${id} to ${target.id}`);
1851
+ return;
1852
+ }
1853
+ // --head is only meaningful for the splice cell (it can't resurrect or remove).
1854
+ if (headOnly) {
1855
+ fail("--head only applies when the block exists in both the document and the target revision", 2);
1856
+ }
1857
+ // absent now, present at R -> RESURRECT (undo delete)
1858
+ if (curBlock === undefined && oldBlock !== undefined) {
1859
+ // Guard: if the block we'd resurrect is the same (modulo id) as one already
1860
+ // present under a different id, #id was likely renamed away — resurrecting
1861
+ // would duplicate it. Point at `rename` instead of writing.
1862
+ const cmpKey = normalizeBlockId(oldBlock, "__cmp__");
1863
+ for (const [cid, cs] of blockSpans(source)) {
1864
+ if (cid === id)
1865
+ continue;
1866
+ const csrc = splitLines(source).slice(cs.start, cs.end).join("");
1867
+ if (normalizeBlockId(csrc, "__cmp__") === cmpKey) {
1868
+ fail(`#${id} looks renamed to #${cid}; use 'rename #${cid} #${id}' to undo the rename`, 1);
1869
+ }
1870
+ }
1871
+ const { at, where, warn } = resurrectPosition(source, target.text, id, before, after, append, file);
1872
+ if (dryRun) {
1873
+ console.error(`would resurrect #${id} from ${target.id} at ${where}:`);
1874
+ process.stdout.write(oldBlock.endsWith("\n") ? oldBlock : oldBlock + "\n");
1875
+ return;
1876
+ }
1877
+ if (warn)
1878
+ console.error(`warning: anchors for #${id} are gone; appended at end`);
1879
+ emit(insertFragment(source, splitLines(source), at, oldBlock, file), `resurrected #${id} from ${target.id} at ${where}`);
1260
1880
  return;
1261
1881
  }
1882
+ // present now, absent at R -> REMOVE (undo add)
1883
+ // Guard: if the block we'd remove is the same (modulo id) as one present at R
1884
+ // under a different id, #id was likely renamed IN — removing would delete a
1885
+ // renamed block. Point at `rename` instead (the dangerous direction).
1886
+ {
1887
+ const cmpKey = normalizeBlockId(curBlock, "__cmp__");
1888
+ for (const [rid, rs] of blockSpans(target.text)) {
1889
+ if (rid === id)
1890
+ continue;
1891
+ const rsrc = splitLines(target.text).slice(rs.start, rs.end).join("");
1892
+ if (normalizeBlockId(rsrc, "__cmp__") === cmpKey) {
1893
+ fail(`#${id} looks renamed from #${rid}; revert would delete it — use 'rename #${id} #${rid}'`, 1);
1894
+ }
1895
+ }
1896
+ }
1262
1897
  if (dryRun) {
1263
- console.error(`would revert #${id} to ${target.id}:`);
1264
- process.stdout.write(oldBlock.endsWith("\n") ? oldBlock : oldBlock + "\n");
1898
+ console.error(`would remove #${id} (absent at ${target.id})`);
1265
1899
  return;
1266
1900
  }
1267
- const updated = spliceBlock(source, id, oldBlock, file, headOnly);
1268
- const dest = out ?? file;
1269
- writeFileSync(dest, updated);
1270
- console.error(`reverted #${id} to ${target.id}${dest === file ? "" : ` -> ${dest}`}`);
1901
+ const span = curFull;
1902
+ const beforeIds = parse(source, { resolveDoc: resolverFor(file) }).ids;
1903
+ const updated = splitLines(source).filter((_, i) => i < span.start || i >= span.end).join("");
1904
+ const reparsed = parse(updated, { resolveDoc: resolverFor(file) });
1905
+ const errs = reparsed.diagnostics.filter((d) => d.severity === "error");
1906
+ if (errs.length) {
1907
+ const first = errs[0];
1908
+ refuseBroken(`removing #${id} would break the document: ${first.message} (line ${first.line}); not written`, errs);
1909
+ }
1910
+ const now = new Set(reparsed.ids);
1911
+ const dropped = beforeIds.find((x) => x !== id && !now.has(x));
1912
+ if (dropped !== undefined)
1913
+ fail(`removing #${id} would drop block \`#${dropped}\`; not written`, 1);
1914
+ emit(updated, `removed #${id} (absent at ${target.id})`);
1915
+ }
1916
+ // Choose the physical-line insertion point for a resurrected block. Explicit
1917
+ // --append/--before/--after win; otherwise infer from the block's neighbours in
1918
+ // revision R: the nearest id BEFORE it that still exists now (insert after it),
1919
+ // else the nearest id AFTER it that still exists (insert before it), else append
1920
+ // at end (warn=true). The deleted block's own former descendants are absent now
1921
+ // too, so they are naturally skipped as anchors.
1922
+ function resurrectPosition(source, revText, id, before, after, append, file) {
1923
+ const lines = splitLines(source);
1924
+ const here = blockSpans(source);
1925
+ if (append)
1926
+ return { at: lines.length, where: "end", warn: false };
1927
+ if (before !== undefined) {
1928
+ const a = before.replace(/^#/, "");
1929
+ const s = here.get(a);
1930
+ if (!s)
1931
+ fail(`no block with id \`${a}\` in ${file}`, 1);
1932
+ return { at: s.start, where: `before #${a}`, warn: false };
1933
+ }
1934
+ if (after !== undefined) {
1935
+ const a = after.replace(/^#/, "");
1936
+ const s = here.get(a);
1937
+ if (!s)
1938
+ fail(`no block with id \`${a}\` in ${file}`, 1);
1939
+ return { at: s.end, where: `after #${a}`, warn: false };
1940
+ }
1941
+ const revIds = [...blockSpans(revText).keys()];
1942
+ const idx = revIds.indexOf(id);
1943
+ for (let i = idx - 1; i >= 0; i--) {
1944
+ const s = here.get(revIds[i]);
1945
+ if (s)
1946
+ return { at: s.end, where: `after #${revIds[i]}`, warn: false };
1947
+ }
1948
+ for (let i = idx + 1; i < revIds.length; i++) {
1949
+ const s = here.get(revIds[i]);
1950
+ if (s)
1951
+ return { at: s.start, where: `before #${revIds[i]}`, warn: false };
1952
+ }
1953
+ return { at: lines.length, where: "end", warn: true };
1271
1954
  }
1272
1955
  // geml codemap <sub>: the code-graph toolkit ships as plain scripts in the
1273
1956
  // package's codemap/ directory (they are argv-driven programs, some
@@ -1291,6 +1974,15 @@ function runCodemap(args) {
1291
1974
  const r = spawnSync(process.execPath, [mod, ...args.slice(1)], { stdio: "inherit" });
1292
1975
  process.exit(r.status ?? 1);
1293
1976
  }
1977
+ // geml mcp: the document-CRUD MCP server. Runs as a child's MAIN module for the
1978
+ // same reason `codemap mcp` does — it owns stdin/stdout for the whole session
1979
+ // (the stdio transport), and dispatching by spawn keeps this module free of a
1980
+ // runtime import cycle (mcp.js imports the parser from here).
1981
+ function runMcp(args) {
1982
+ const mod = join(dirname(fileURLToPath(import.meta.url)), "mcp.js");
1983
+ const r = spawnSync(process.execPath, [mod, ...args], { stdio: "inherit" });
1984
+ process.exit(r.status ?? 1);
1985
+ }
1294
1986
  // npm's unix bin shim is a symlink named plain `geml`, so detect "run as a
1295
1987
  // CLI" by resolving argv[1] to its real path, not by its spelling.
1296
1988
  const entry = (() => {
@@ -1339,40 +2031,39 @@ if (entry && (entry === fileURLToPath(import.meta.url) || entry.endsWith("geml.t
1339
2031
  else if (cmd === "set") {
1340
2032
  runSet(argv.slice(1));
1341
2033
  }
2034
+ else if (cmd === "add") {
2035
+ runAdd(argv.slice(1));
2036
+ }
2037
+ else if (cmd === "delete") {
2038
+ runDelete(argv.slice(1));
2039
+ }
2040
+ else if (cmd === "rename") {
2041
+ runRename(argv.slice(1));
2042
+ }
1342
2043
  else if (cmd === "revert") {
1343
2044
  runRevert(argv.slice(1));
1344
2045
  }
1345
2046
  else if (cmd === "history") {
1346
2047
  runHistory(argv.slice(1));
1347
2048
  }
1348
- else if (cmd === "convert") {
1349
- runConvert(argv.slice(1));
1350
- }
1351
- else if (cmd === "export") {
1352
- runExport(argv.slice(1));
1353
- }
1354
- else if (cmd === "render") {
1355
- runRender(argv.slice(1));
1356
- }
1357
- else if (cmd === "fmt") {
1358
- runFmt(argv.slice(1));
1359
- }
1360
2049
  else if (cmd === "check") {
1361
2050
  runCheck(argv.slice(1));
1362
2051
  }
1363
2052
  else if (cmd === "codemap") {
1364
2053
  runCodemap(argv.slice(1));
1365
2054
  }
2055
+ else if (cmd === "mcp") {
2056
+ runMcp(argv.slice(1));
2057
+ }
1366
2058
  else if (cmd !== "-" && !/[.\/\\]/.test(cmd)) {
1367
2059
  // A bare word that is neither a known command nor a path is almost always
1368
- // a mistyped command — say so, don't try to read it as a file.
2060
+ // a mistyped command — say so, don't try to read it as a file. (The
2061
+ // reclaimed verbs render/export/fmt/convert land here too.)
1369
2062
  fail(`unknown command '${cmd}'. Run 'geml --help'.`);
1370
2063
  }
1371
2064
  else {
1372
- // Default: parse a file (or stdin via '-') to the document-model JSON.
1373
- const doc = parse(readInput(cmd), { resolveDoc: resolverFor(cmd) });
1374
- console.log(JSON.stringify(doc, null, 2));
1375
- if (doc.diagnostics.some((d) => d.severity === "error"))
1376
- process.exit(1);
2065
+ // A file (or stdin via '-') is the transform entry: `--to`/`--from`/`-o`,
2066
+ // default `--to json`. The single door for every format conversion.
2067
+ runTransform(argv);
1377
2068
  }
1378
2069
  }