@ifc-lite/cli 0.28.0 → 0.28.1

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.
@@ -0,0 +1,299 @@
1
+ /* This Source Code Form is subject to the terms of the Mozilla Public
2
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
3
+ * file, You can obtain one at https://mozilla.org/MPL/2.0/. */
4
+ /**
5
+ * Reading a STEP record's top-level argument list out of its text, for a
6
+ * caller that then edits one slot BY INDEX.
7
+ *
8
+ * The failure this exists to prevent is silent. A scanner that loses track of
9
+ * quote state or paren depth still produces parts; they are just not the
10
+ * record's arguments any more, because the commas it swallowed took every
11
+ * following slot with them. The write lands on whatever the mis-scan
12
+ * accumulated and the caller reports a success that did not happen
13
+ * (LTplus-AG/ifc-lite#2470, #4125 for the `mutate` instance). So this returns
14
+ * null rather than parts it does not believe in.
15
+ *
16
+ * The mis-scan nothing structural notices is an undoubled apostrophe, what an
17
+ * authoring tool emits when it forgets to double one. The scan closes the
18
+ * string at it and reopens on the next quote, so the text between the two is
19
+ * read inside-out and every comma in it is swallowed. TWO of them leave quote
20
+ * parity EVEN and paren depth at ZERO, so the scan ends clean on text that
21
+ * split wrong: `'guid',$,'John's wall',$,$,$,$,'A's',.NOTDEFINED.` is nine
22
+ * attributes read as four, and writing slot 2 deletes attributes 3 to 7.
23
+ *
24
+ * What always shows is a token ending where a token cannot end: the phantom
25
+ * terminator leaves the rest of the content where a separator belongs
26
+ * (`'John's wall'` closes after `John`, leaving a bare `s`). So this validates
27
+ * the list as a grammar rather than counting characters. An argument is ONE
28
+ * token, after a token only whitespace and then `,` or `)` may follow, and that
29
+ * (In the STEP examples below, `*\/` is a JavaScript escape so the example does not
30
+ * terminate this comment block. The backslash is not in the STEP text, and STEP
31
+ * gives `\` its own meaning, so strip it before copying an example into a test.)
32
+ *
33
+ * holds AT EVERY DEPTH. Applied to the top level alone it misses a phantom that
34
+ * swallows a `)`, a `(` and the comma between them, which leaves the depth
35
+ * balanced and each surviving part passing a token check on its own:
36
+ * `IFCLABEL('a's'),$,IFCLABEL('b's'),$` is four attributes read as two. Any
37
+ * two top-level slots that each hold a string inside parens can do it, and real
38
+ * entities have that shape (`IfcPerson`'s MiddleNames / PrefixTitles,
39
+ * `IfcPropertyTableValue`'s DefiningValues / DefinedValues).
40
+ *
41
+ * What the rule cannot see is a corruption whose bytes are themselves a valid
42
+ * argument list: a stray apostrophe at the END of a string's content emits a
43
+ * doubled quote by accident, so `''','''` is two arguments to its author and
44
+ * one to anyone reading the text. A generated sweep accepted 814 of 80,035 corruptions
45
+ * with a split other than the author's, down from 1,066 before the rule reached into
46
+ * lists, and every survivor it looked at had that shape; separating them needs the
47
+ * schema, not the text. That harness is NOT committed, so those figures record one run
48
+ * and cannot be reproduced from this tree.
49
+ *
50
+ * A STEP block comment is refused rather than understood, because `mutate` must
51
+ * not rewrite a record it cannot read: `/` breaks a bare run, so a part
52
+ * carrying a comment outside a string is never one token, whatever the comment
53
+ * CONTAINS. Whitespace in the comment is not what gives it away:
54
+ * `$,/*renamed*\/,$` has none and was read as three slots for two attributes,
55
+ * the same phantom-slot shift as the rest of #4125 (see {@link isTokenBreak}
56
+ * for why breaking on the OPENER is enough).
57
+ *
58
+ * Several readers in this repo DO skip comments; diverging from them is
59
+ * deliberate. The nearest is `validate.ts` in this very package, which skips
60
+ * `/* ... *\/` while counting top-level attribute indices; it returns indices
61
+ * rather than parts, so it never has to put the bytes back. `source-header.ts`'s
62
+ * `splitTopLevel` is closer still, since it returns PARTS, but it drops the
63
+ * comment bytes, so it cannot satisfy the round-trip contract below either.
64
+ * And `packages/parser`'s
65
+ * `entity-extractor.ts` reads comments at every depth through
66
+ * `StepTextScan.skipLexicalAt`, and says so as policy: one comment-skip rule for
67
+ * decoded STEP text rather than a fourth hand-rolled copy. The difference is what
68
+ * the two produce. That one extracts VALUES and collapses a comment to a space;
69
+ * this one must satisfy `parts.join(',') === input` byte for byte, because its
70
+ * caller rewrites the user's file. Skipping a comment under that contract would
71
+ * silently delete it from their file, and keeping it inside the part means the
72
+ * part is not one token. Doing both needs a richer return type than `string[]`,
73
+ * which is a different module. (`STEP_TRIVIA` itself is legal anywhere whitespace
74
+ * is. Its call sites are what place it, and they all put it immediately before a
75
+ * `\(`.)
76
+ *
77
+ * What this module does NOT cover, so its refusal is not read as more than it is:
78
+ * `mutate` finds records with a line regex that allows only whitespace before the
79
+ * `(`, so a comment after the class keyword is skipped BEFORE reaching this code,
80
+ * and the run reports success having changed nothing; and `lastIndexOf(')')` can
81
+ * slice into a legal trailing comment. Both are #4163, one on each SIDE of the
82
+ * argument list (before the `(`, after the `)`), and both are the same
83
+ * line-versus-record shape as the wrapped-record case #4163 also covers.
84
+ *
85
+ * `packages/export/src/step-argument-parser.ts` is the nearest copy, and this is
86
+ * deliberately its near-twin rather than an import: `@ifc-lite/export`'s
87
+ * `exports` map exposes only `.`, so reaching it would mean adding a published
88
+ * export (and an `api-surface` entry) to a v4.0.0 package to fix a CLI bug.
89
+ * #4125 tracks consolidating them (`derive-variants.mts` and
90
+ * `placement-datum.test.ts` are two the issue does not name), and that is a real merge rather
91
+ * than a no-op swap: its `splitTopLevelStepArguments` has the three structural
92
+ * checks and NO per-part check at all, which is exactly the top-level-only
93
+ * version the two paragraphs above show is insufficient. Measured, it splits
94
+ * `$,/* renamed *\/ $` into two parts and `'guid',$,'Name',/* c *\/,$` into five.
95
+ * The three structural checks are the contract the two DO share.
96
+ */
97
+ /**
98
+ * Split a STEP argument list on top-level commas, or return null when the text
99
+ * cannot be scanned as one. `input` is the text BETWEEN a record's outermost
100
+ * parentheses; the module header says why null rather than parts.
101
+ *
102
+ * Nothing is trimmed and nothing is normalised, so for input this accepts
103
+ * `parts.join(',')` reproduces `input` byte for byte and a caller that replaces
104
+ * one part leaves every other byte of the record alone. Rejected: a quote still
105
+ * open at the end, a paren depth that does not return to zero, a depth that
106
+ * ever goes NEGATIVE (a stray closing paren balanced by a later opening one),
107
+ * and a part that is not one token at every depth (see {@link isLoneStepToken}).
108
+ *
109
+ * An EMPTY top-level slot (`a,,b`, or a trailing comma) is NOT rejected: it is
110
+ * invalid STEP that costs no alignment, since an empty argument is one part
111
+ * exactly as the entity parser counts it, so every index still names the
112
+ * attribute it is meant to. An empty INPUT is not an empty slot: `#1=IFCFOO();`
113
+ * has no arguments, so it splits to `[]` and any slot request then fails the
114
+ * caller's bounds check.
115
+ */
116
+ export function splitTopLevelStepArgs(input) {
117
+ if (input === '')
118
+ return [];
119
+ const parts = [];
120
+ let current = '';
121
+ let depth = 0;
122
+ let inString = false;
123
+ for (let i = 0; i < input.length; i++) {
124
+ const char = input[i];
125
+ if (char === "'") {
126
+ current += char;
127
+ // A doubled quote INSIDE a string is an escaped apostrophe. Outside one
128
+ // the first quote opens a string and the second is read on its own next
129
+ // pass, so `''` there is the empty string rather than an escape.
130
+ if (inString && input[i + 1] === "'") {
131
+ current += input[i + 1];
132
+ i++;
133
+ continue;
134
+ }
135
+ inString = !inString;
136
+ continue;
137
+ }
138
+ if (!inString) {
139
+ if (char === '(') {
140
+ depth++;
141
+ }
142
+ else if (char === ')') {
143
+ depth--;
144
+ if (depth < 0)
145
+ return null;
146
+ }
147
+ else if (char === ',' && depth === 0) {
148
+ parts.push(current);
149
+ current = '';
150
+ continue;
151
+ }
152
+ }
153
+ current += char;
154
+ }
155
+ if (inString || depth !== 0)
156
+ return null;
157
+ parts.push(current);
158
+ // These three are the contract this shares with
159
+ // `packages/export/src/step-argument-parser.ts`, which has exactly them and no
160
+ // per-part check. That parity, not speed, is why they are kept: they add no
161
+ // coverage at all, and deleting any one of them, or all three, fails no test
162
+ // (measured). Only the `depth < 0` check above is an early exit; these two run
163
+ // after the whole scan and save only a walk over the parts. Kept as
164
+ // the shared spelling until #4125 merges the copies.
165
+ return parts.every(isLoneStepToken) ? parts : null;
166
+ }
167
+ /**
168
+ * Is `part` exactly ONE STEP argument, all the way down? Surrounding whitespace
169
+ * is ignored, because a record may carry it and this must not refuse a record
170
+ * it could rewrite. An empty part passes, as an empty slot is accepted above.
171
+ */
172
+ function isLoneStepToken(part) {
173
+ const text = part.trim();
174
+ return text === '' || skipToken(text, 0) === text.length;
175
+ }
176
+ /**
177
+ * Index just past the one token starting at `from`, or -1 when the text there
178
+ * is not one token. The forms, per ISO 10303-21: a string (`'...'`, `''`
179
+ * escaping an apostrophe), a list (`(...)`), or a bare run (`$`, `*`, `#123`, a
180
+ * number, an enumeration `.T.`, a binary `"0F"`) optionally applied to a list
181
+ * (`IFCINTEGER(3)`).
182
+ *
183
+ * Deliberately loose about what a bare run CONTAINS and strict only about where
184
+ * it may end: this is not a STEP validator and must not become one. A list's
185
+ * elements are held to the same rule, so a fragment left by a phantom string
186
+ * terminator is caught wherever it lands. Nesting is carried in `depth` rather
187
+ * than by recursing, so a pathologically nested record (this text comes from a
188
+ * file) refuses instead of overflowing the stack.
189
+ */
190
+ function skipToken(text, from) {
191
+ let i = from;
192
+ let depth = 0;
193
+ let expectToken = true;
194
+ for (;;) {
195
+ i += countWhitespace(text, i);
196
+ const char = text[i];
197
+ if (expectToken) {
198
+ // A list element may be empty (`(1,,2)`), as a top-level slot may be.
199
+ if (depth > 0 && (char === ',' || char === ')')) {
200
+ expectToken = false;
201
+ continue;
202
+ }
203
+ if (char === "'") {
204
+ i = skipString(text, i);
205
+ if (i < 0)
206
+ return -1;
207
+ }
208
+ else if (char === '(') {
209
+ depth++;
210
+ i++;
211
+ continue; // the list's first element is still a token to read
212
+ }
213
+ else {
214
+ const start = i;
215
+ while (i < text.length && !isTokenBreak(text[i]))
216
+ i++;
217
+ // Nothing consumed: the end of the text, or a separator where a token
218
+ // belongs.
219
+ if (i === start)
220
+ return -1;
221
+ const afterKeyword = i + countWhitespace(text, i);
222
+ if (text[afterKeyword] === '(') {
223
+ depth++;
224
+ i = afterKeyword + 1;
225
+ continue;
226
+ }
227
+ }
228
+ expectToken = false;
229
+ continue;
230
+ }
231
+ // A token just closed, so only `,` or `)` may follow. At depth zero the
232
+ // token was the whole argument.
233
+ if (depth === 0)
234
+ return i;
235
+ if (char === ')') {
236
+ depth--;
237
+ i++;
238
+ continue;
239
+ }
240
+ if (char !== ',')
241
+ return -1;
242
+ i++;
243
+ expectToken = true;
244
+ }
245
+ }
246
+ /** Index just past the string starting at `from`, or -1 if it never closes. */
247
+ function skipString(text, from) {
248
+ for (let i = from + 1; i < text.length; i++) {
249
+ if (text[i] !== "'")
250
+ continue;
251
+ if (text[i + 1] === "'")
252
+ i++;
253
+ else
254
+ return i + 1;
255
+ }
256
+ return -1;
257
+ }
258
+ /**
259
+ * Can this character only begin or separate another token?
260
+ *
261
+ * `/` is in the set for the STEP block comment, which opens with `/*`. Breaking
262
+ * on the OPENER is enough, and is why there is no comment state machine here: a
263
+ * comment can only sit where a token or a separator belongs, and either way the
264
+ * `/` ends the bare run short of the part's end, so `isLoneStepToken` refuses
265
+ * the part however the comment is written. `*` is deliberately NOT in the set,
266
+ * because it is the derived-attribute marker and a token in its own right, so
267
+ * breaking on it would refuse `$,$,*`.
268
+ */
269
+ function isTokenBreak(char) {
270
+ return (char === "'" ||
271
+ char === '(' ||
272
+ char === ')' ||
273
+ char === ',' ||
274
+ char === '/' ||
275
+ char === ' ' ||
276
+ char === '\t');
277
+ }
278
+ /**
279
+ * How many whitespace characters run from `from`.
280
+ *
281
+ * Space and tab only, where every NAMED STEP whitespace set in this repo
282
+ * (`is_step_space`, `isSpaceByte`, `isAsciiSpace`, `STEP_TRIVIA`) is the six
283
+ * characters ` \t\n\r\x0b\x0c`. That is safe here only because the caller
284
+ * splits on newlines before this ever runs. `\n` and `\r` are in neither this
285
+ * set nor {@link isTokenBreak}, so a bare run spans them silently and two tokens
286
+ * separated by a newline come back as ONE part (measured: `'$\n$'` splits to
287
+ * `["$\n$"]`, where `'$ $'` and `'$\t$'` are refused). That is the same
288
+ * token-ending-where-it-cannot defect this module exists to catch, unreachable only
289
+ * because the caller pre-splits on newlines. #4163's multi-line work would
290
+ * start feeding real multi-line argument lists through here; widen both sets then
291
+ * rather than relying on that accident.
292
+ */
293
+ function countWhitespace(text, from) {
294
+ let n = 0;
295
+ while (from + n < text.length && (text[from + n] === ' ' || text[from + n] === '\t'))
296
+ n++;
297
+ return n;
298
+ }
299
+ //# sourceMappingURL=step-args.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"step-args.js","sourceRoot":"","sources":["../../src/commands/step-args.ts"],"names":[],"mappings":"AAAA;;+DAE+D;AAE/D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4FG;AAEH;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAM,UAAU,qBAAqB,CAAC,KAAa;IACjD,IAAI,KAAK,KAAK,EAAE;QAAE,OAAO,EAAE,CAAC;IAE5B,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,OAAO,GAAG,EAAE,CAAC;IACjB,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,IAAI,QAAQ,GAAG,KAAK,CAAC;IAErB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QAEtB,IAAI,IAAI,KAAK,GAAG,EAAE,CAAC;YACjB,OAAO,IAAI,IAAI,CAAC;YAChB,wEAAwE;YACxE,wEAAwE;YACxE,iEAAiE;YACjE,IAAI,QAAQ,IAAI,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,GAAG,EAAE,CAAC;gBACrC,OAAO,IAAI,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;gBACxB,CAAC,EAAE,CAAC;gBACJ,SAAS;YACX,CAAC;YACD,QAAQ,GAAG,CAAC,QAAQ,CAAC;YACrB,SAAS;QACX,CAAC;QAED,IAAI,CAAC,QAAQ,EAAE,CAAC;YACd,IAAI,IAAI,KAAK,GAAG,EAAE,CAAC;gBACjB,KAAK,EAAE,CAAC;YACV,CAAC;iBAAM,IAAI,IAAI,KAAK,GAAG,EAAE,CAAC;gBACxB,KAAK,EAAE,CAAC;gBACR,IAAI,KAAK,GAAG,CAAC;oBAAE,OAAO,IAAI,CAAC;YAC7B,CAAC;iBAAM,IAAI,IAAI,KAAK,GAAG,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;gBACvC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;gBACpB,OAAO,GAAG,EAAE,CAAC;gBACb,SAAS;YACX,CAAC;QACH,CAAC;QAED,OAAO,IAAI,IAAI,CAAC;IAClB,CAAC;IAED,IAAI,QAAQ,IAAI,KAAK,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACzC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IACpB,gDAAgD;IAChD,+EAA+E;IAC/E,4EAA4E;IAC5E,6EAA6E;IAC7E,+EAA+E;IAC/E,oEAAoE;IACpE,qDAAqD;IACrD,OAAO,KAAK,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC;AACrD,CAAC;AAED;;;;GAIG;AACH,SAAS,eAAe,CAAC,IAAY;IACnC,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IACzB,OAAO,IAAI,KAAK,EAAE,IAAI,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC,KAAK,IAAI,CAAC,MAAM,CAAC;AAC3D,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,SAAS,SAAS,CAAC,IAAY,EAAE,IAAY;IAC3C,IAAI,CAAC,GAAG,IAAI,CAAC;IACb,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,IAAI,WAAW,GAAG,IAAI,CAAC;IAEvB,SAAS,CAAC;QACR,CAAC,IAAI,eAAe,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QAC9B,MAAM,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QAErB,IAAI,WAAW,EAAE,CAAC;YAChB,sEAAsE;YACtE,IAAI,KAAK,GAAG,CAAC,IAAI,CAAC,IAAI,KAAK,GAAG,IAAI,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;gBAChD,WAAW,GAAG,KAAK,CAAC;gBACpB,SAAS;YACX,CAAC;YACD,IAAI,IAAI,KAAK,GAAG,EAAE,CAAC;gBACjB,CAAC,GAAG,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;gBACxB,IAAI,CAAC,GAAG,CAAC;oBAAE,OAAO,CAAC,CAAC,CAAC;YACvB,CAAC;iBAAM,IAAI,IAAI,KAAK,GAAG,EAAE,CAAC;gBACxB,KAAK,EAAE,CAAC;gBACR,CAAC,EAAE,CAAC;gBACJ,SAAS,CAAC,oDAAoD;YAChE,CAAC;iBAAM,CAAC;gBACN,MAAM,KAAK,GAAG,CAAC,CAAC;gBAChB,OAAO,CAAC,GAAG,IAAI,CAAC,MAAM,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;oBAAE,CAAC,EAAE,CAAC;gBACtD,sEAAsE;gBACtE,WAAW;gBACX,IAAI,CAAC,KAAK,KAAK;oBAAE,OAAO,CAAC,CAAC,CAAC;gBAC3B,MAAM,YAAY,GAAG,CAAC,GAAG,eAAe,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;gBAClD,IAAI,IAAI,CAAC,YAAY,CAAC,KAAK,GAAG,EAAE,CAAC;oBAC/B,KAAK,EAAE,CAAC;oBACR,CAAC,GAAG,YAAY,GAAG,CAAC,CAAC;oBACrB,SAAS;gBACX,CAAC;YACH,CAAC;YACD,WAAW,GAAG,KAAK,CAAC;YACpB,SAAS;QACX,CAAC;QAED,wEAAwE;QACxE,gCAAgC;QAChC,IAAI,KAAK,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QAC1B,IAAI,IAAI,KAAK,GAAG,EAAE,CAAC;YACjB,KAAK,EAAE,CAAC;YACR,CAAC,EAAE,CAAC;YACJ,SAAS;QACX,CAAC;QACD,IAAI,IAAI,KAAK,GAAG;YAAE,OAAO,CAAC,CAAC,CAAC;QAC5B,CAAC,EAAE,CAAC;QACJ,WAAW,GAAG,IAAI,CAAC;IACrB,CAAC;AACH,CAAC;AAED,+EAA+E;AAC/E,SAAS,UAAU,CAAC,IAAY,EAAE,IAAY;IAC5C,KAAK,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5C,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG;YAAE,SAAS;QAC9B,IAAI,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,GAAG;YAAE,CAAC,EAAE,CAAC;;YACxB,OAAO,CAAC,GAAG,CAAC,CAAC;IACpB,CAAC;IACD,OAAO,CAAC,CAAC,CAAC;AACZ,CAAC;AAED;;;;;;;;;;GAUG;AACH,SAAS,YAAY,CAAC,IAAY;IAChC,OAAO,CACL,IAAI,KAAK,GAAG;QACZ,IAAI,KAAK,GAAG;QACZ,IAAI,KAAK,GAAG;QACZ,IAAI,KAAK,GAAG;QACZ,IAAI,KAAK,GAAG;QACZ,IAAI,KAAK,GAAG;QACZ,IAAI,KAAK,IAAI,CACd,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,SAAS,eAAe,CAAC,IAAY,EAAE,IAAY;IACjD,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,OAAO,IAAI,GAAG,CAAC,GAAG,IAAI,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,KAAK,GAAG,IAAI,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,KAAK,IAAI,CAAC;QAAE,CAAC,EAAE,CAAC;IAC1F,OAAO,CAAC,CAAC;AACX,CAAC"}
@@ -0,0 +1,124 @@
1
+ /**
2
+ * Spatial-structure relations for an extracted subset: which ones join it, and
3
+ * what text each one emits.
4
+ *
5
+ * `IfcRelContainedInSpatialStructure` is one-to-MANY, and a real exporter
6
+ * writes exactly ONE of them per storey, naming every product in that storey.
7
+ * An all-or-nothing rule (keep the relation only when every id it mentions is
8
+ * kept) therefore drops containment on ANY strict subset of a real model: the
9
+ * extracted products land outside the spatial tree and a viewer shows the
10
+ * storey with nothing under it. So the `RelatedElements` SET is rewritten down
11
+ * to the kept members instead of the relation being dropped.
12
+ *
13
+ * The no-dangling-reference invariant is unchanged: every `#id` in an emitted
14
+ * record is an id the subset keeps, and since #4128 the caller's forward
15
+ * closure keeps only ids the file DEFINES, so kept implies defined. That covers
16
+ * the ids this module CHOOSES; it does not cover the references inside a kept
17
+ * record, which are emitted verbatim, so a source file that already dangles
18
+ * still dangles. That is the rest of the rules:
19
+ * - the RELATING object (`RelatingStructure` / `RelatingObject`, the spatial
20
+ * parent) is a hard requirement. A containment with no parent is
21
+ * meaningless, and it would dangle. A KNOWN residual gap follows from that,
22
+ * and it is not this module's to close: `buildSubset` force-keeps only
23
+ * `IfcProject` / `IfcSite` / `IfcBuilding` / `IfcBuildingStorey`, so a
24
+ * product contained in an `IfcSpace` or `IfcSpatialZone` still has an
25
+ * unkept parent and still loses containment. Widening that type list is the
26
+ * wrong close, because it force-keeps every space AND its forward closure in
27
+ * every extraction; the right one is a BACKWARD closure from each kept
28
+ * product up its containment and aggregation edges, which keeps exactly the
29
+ * seeds' ancestors and needs no type list at all. Either way it is a change
30
+ * to how `buildSubset` seeds, not to the rule here. Filed as #4124.
31
+ * - an empty intersection drops the relation.
32
+ * - every other non-set reference must be kept too. That is attribute 1,
33
+ * `OwnerHistory`; usually it already is, because each kept product's own
34
+ * body names the same shared `IfcOwnerHistory` and the products' forward
35
+ * closure keeps it. An exporter that writes a PER-RELATIONSHIP
36
+ * `IfcOwnerHistory` referenced by nothing else hits this rule every time,
37
+ * and dropping there was the orphaned-storey symptom again (#4126). This
38
+ * function still takes no `parsed`, so it cannot close over such a
39
+ * reference; it REPORTS it in {@link SpatialRelationPlan.blockedOn} and the
40
+ * caller, which does have `parsed`, keeps it and replans. Purity is intact:
41
+ * `blockedOn` is a finding, not a mutation.
42
+ *
43
+ * A KNOWN gap: {@link keepWhole}, the fallback for a record this module could
44
+ * not read as its six attributes, still drops on the same private
45
+ * `OwnerHistory` and reports nothing. That is deliberate. Its references are
46
+ * read off the raw body with no established string boundaries, so a `#6` it
47
+ * names may be text rather than a reference, and forward-closing over it would
48
+ * re-create the hash-in-a-Name bug the scanner exists to avoid.
49
+ *
50
+ * A relation that loses NO member re-emits its source line verbatim, so an
51
+ * extraction that happened to keep every member does not churn.
52
+ *
53
+ * ## What this module re-implements, and why
54
+ *
55
+ * Three things here already exist in `@ifc-lite/export`, and all three are
56
+ * copied for ONE reason: that package's `exports` map exposes only `.`, and its
57
+ * `index.ts` re-exports none of them, so reaching any of them would mean adding
58
+ * a published export (and an `api-surface` entry) to a v4.0.0 package in order
59
+ * to fix a CLI bug.
60
+ *
61
+ * - `filterHiddenRefsFromRelationshipLine` (`reference-collector.ts`) is the
62
+ * same job, done better: it filters EVERY parenthesised attribute of ANY
63
+ * `IFCREL*` line against an `isExcluded` predicate, so it needs no slot
64
+ * table at all, and it carries an edge case this module has no equivalent
65
+ * of (`IfcRelConnectsStructuralMember`'s optional trailing placement).
66
+ * Adopting it would delete most of this file. The export barrier above is
67
+ * the whole reason it is not adopted here, and nothing else: fed only
68
+ * today's three types it would change no behaviour, because the CALLER
69
+ * picks which lines it sees. Widening the type set is the separate decision,
70
+ * and that is the one that changes which relations survive an extraction.
71
+ * Consolidation follow-up: #4125.
72
+ * - `splitTopLevelStepArguments` (`step-argument-parser.ts`): see
73
+ * {@link splitTopLevelArgs}.
74
+ * - `STRUCTURE_RELATIONS` (`merged-empty-containers.ts`): three lines, see
75
+ * below.
76
+ */
77
+ /**
78
+ * One parsed STEP record, as `extract-entities.ts`'s `parseStep` produces it.
79
+ * Declared HERE and imported there, rather than the other way round, so the
80
+ * dependency runs one way: `extract-entities.ts` imports this module, never
81
+ * the reverse.
82
+ */
83
+ export interface StepRecord {
84
+ id: number;
85
+ type: string;
86
+ /** Argument text between the outermost parentheses. */
87
+ body: string;
88
+ /** The verbatim `#id= TYPE(...);` text. */
89
+ full: string;
90
+ }
91
+ /**
92
+ * The assembled subset: the ids to emit, plus the record text to emit for the
93
+ * relations whose related-objects SET was filtered. An id absent from
94
+ * `rewritten` emits its source line unchanged.
95
+ */
96
+ export interface Subset {
97
+ keep: Set<number>;
98
+ rewritten: ReadonlyMap<number, string>;
99
+ }
100
+ /** What {@link planSpatialRelations} found: relations to add, and their text. */
101
+ export interface SpatialRelationPlan {
102
+ /** Relation ids that join the subset. */
103
+ add: number[];
104
+ /** Relation id → rewritten record text, for the ones that lost a member. */
105
+ rewritten: Map<number, string>;
106
+ /**
107
+ * Unkept non-SET references, in practice a relation-private `OwnerHistory`,
108
+ * of the relations that this plan dropped for THAT reason alone: their
109
+ * relating parent is kept and their member intersection is non-empty, so
110
+ * keeping these ids is all that stands between them and surviving. A caller
111
+ * holding the parsed model can close over them and replan (#4126). Every
112
+ * other drop is final and reports nothing here.
113
+ */
114
+ blockedOn: number[];
115
+ }
116
+ /**
117
+ * Decide every spatial-structure relation against a kept-id set.
118
+ *
119
+ * Pure in both arguments: adding relation ids to `keep` afterwards cannot
120
+ * change any verdict, because a relation of these types references products,
121
+ * spatial containers and an OwnerHistory, never another relation.
122
+ */
123
+ export declare function planSpatialRelations(instances: Iterable<StepRecord>, keep: ReadonlySet<number>): SpatialRelationPlan;
124
+ //# sourceMappingURL=subset-relations.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"subset-relations.d.ts","sourceRoot":"","sources":["../../src/commands/subset-relations.ts"],"names":[],"mappings":"AAIA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2EG;AAEH;;;;;GAKG;AACH,MAAM,WAAW,UAAU;IACzB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,uDAAuD;IACvD,IAAI,EAAE,MAAM,CAAC;IACb,2CAA2C;IAC3C,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;GAIG;AACH,MAAM,WAAW,MAAM;IACrB,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;IAClB,SAAS,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACxC;AAED,iFAAiF;AACjF,MAAM,WAAW,mBAAmB;IAClC,yCAAyC;IACzC,GAAG,EAAE,MAAM,EAAE,CAAC;IACd,4EAA4E;IAC5E,SAAS,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/B;;;;;;;OAOG;IACH,SAAS,EAAE,MAAM,EAAE,CAAC;CACrB;AA4CD;;;;;;GAMG;AACH,wBAAgB,oBAAoB,CAClC,SAAS,EAAE,QAAQ,CAAC,UAAU,CAAC,EAC/B,IAAI,EAAE,WAAW,CAAC,MAAM,CAAC,GACxB,mBAAmB,CAarB"}