sensemaking 0.12.2 → 0.13.0
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/cjs/cli/download.js +4 -4
- package/dist/cjs/cli/download.js.map +1 -1
- package/dist/cjs/cli/init.js +4 -4
- package/dist/cjs/cli/init.js.map +1 -1
- package/dist/cjs/cli/map.js +4 -4
- package/dist/cjs/cli/map.js.map +1 -1
- package/dist/cjs/cli/named.js +2 -2
- package/dist/cjs/cli/named.js.map +1 -1
- package/dist/cjs/cli/path.js +6 -6
- package/dist/cjs/cli/path.js.map +1 -1
- package/dist/cjs/cli/peek.js +4 -4
- package/dist/cjs/cli/peek.js.map +1 -1
- package/dist/cjs/cli/related.js +4 -4
- package/dist/cjs/cli/related.js.map +1 -1
- package/dist/cjs/cli/search.js +4 -4
- package/dist/cjs/cli/search.js.map +1 -1
- package/dist/cjs/cli/shared.d.cts +2 -2
- package/dist/cjs/cli/shared.d.ts +2 -2
- package/dist/cjs/cli/shared.js +5 -5
- package/dist/cjs/cli/shared.js.map +1 -1
- package/dist/cjs/cli/status.js +16 -16
- package/dist/cjs/cli/status.js.map +1 -1
- package/dist/cjs/cli/types.d.cts +1 -1
- package/dist/cjs/cli/types.d.ts +1 -1
- package/dist/cjs/cli.js +3 -3
- package/dist/cjs/cli.js.map +1 -1
- package/dist/cjs/commands/index.d.cts +10 -0
- package/dist/cjs/commands/index.d.ts +10 -0
- package/dist/cjs/commands/index.js +42 -0
- package/dist/cjs/commands/index.js.map +1 -0
- package/dist/cjs/commands/map.d.cts +21 -0
- package/dist/cjs/commands/map.d.ts +21 -0
- package/dist/cjs/commands/map.js +90 -0
- package/dist/cjs/commands/map.js.map +1 -0
- package/dist/cjs/commands/peek.d.cts +20 -0
- package/dist/cjs/commands/peek.d.ts +20 -0
- package/dist/cjs/commands/peek.js +175 -0
- package/dist/cjs/commands/peek.js.map +1 -0
- package/dist/cjs/commands/related.d.cts +6 -0
- package/dist/cjs/commands/related.d.ts +6 -0
- package/dist/cjs/commands/related.js +232 -0
- package/dist/cjs/commands/related.js.map +1 -0
- package/dist/cjs/commands/scope.d.cts +10 -0
- package/dist/cjs/commands/scope.d.ts +10 -0
- package/dist/cjs/commands/scope.js +124 -0
- package/dist/cjs/commands/scope.js.map +1 -0
- package/dist/cjs/commands/search.d.cts +12 -0
- package/dist/cjs/commands/search.d.ts +12 -0
- package/dist/cjs/{commands.js → commands/search.js} +20 -358
- package/dist/cjs/commands/search.js.map +1 -0
- package/dist/cjs/commands/status.d.cts +9 -0
- package/dist/cjs/commands/status.d.ts +9 -0
- package/dist/cjs/commands/status.js +25 -0
- package/dist/cjs/commands/status.js.map +1 -0
- package/dist/cjs/config/access.d.cts +19 -0
- package/dist/cjs/config/access.d.ts +19 -0
- package/dist/cjs/config/access.js +149 -0
- package/dist/cjs/config/access.js.map +1 -0
- package/dist/cjs/config/index.d.cts +5 -0
- package/dist/cjs/config/index.d.ts +5 -0
- package/dist/cjs/config/index.js +77 -0
- package/dist/cjs/config/index.js.map +1 -0
- package/dist/cjs/config/load.d.cts +8 -0
- package/dist/cjs/config/load.d.ts +8 -0
- package/dist/cjs/config/load.js +448 -0
- package/dist/cjs/config/load.js.map +1 -0
- package/dist/cjs/config/resolve.d.cts +6 -0
- package/dist/cjs/config/resolve.d.ts +6 -0
- package/dist/cjs/config/resolve.js +50 -0
- package/dist/cjs/config/resolve.js.map +1 -0
- package/dist/{esm/config.d.ts → cjs/config/types.d.cts} +2 -32
- package/dist/cjs/{config.d.cts → config/types.d.ts} +2 -32
- package/dist/cjs/config/types.js +39 -0
- package/dist/cjs/config/types.js.map +1 -0
- package/dist/cjs/config/validate.d.cts +4 -0
- package/dist/cjs/config/validate.d.ts +4 -0
- package/dist/cjs/config/validate.js +363 -0
- package/dist/cjs/config/validate.js.map +1 -0
- package/dist/cjs/db/index.d.cts +4 -0
- package/dist/cjs/db/index.d.ts +4 -0
- package/dist/cjs/db/index.js +42 -0
- package/dist/cjs/db/index.js.map +1 -0
- package/dist/cjs/{db.d.cts → db/open.d.cts} +2 -8
- package/dist/cjs/{db.d.ts → db/open.d.ts} +2 -8
- package/dist/cjs/db/open.js +223 -0
- package/dist/cjs/db/open.js.map +1 -0
- package/dist/cjs/db/reconcile.d.cts +9 -0
- package/dist/cjs/db/reconcile.d.ts +9 -0
- package/dist/cjs/{db.js → db/reconcile.js} +24 -261
- package/dist/cjs/db/reconcile.js.map +1 -0
- package/dist/cjs/db/shared.d.cts +5 -0
- package/dist/cjs/db/shared.d.ts +5 -0
- package/dist/cjs/db/shared.js +45 -0
- package/dist/cjs/db/shared.js.map +1 -0
- package/dist/cjs/features/embed.d.cts +1 -1
- package/dist/cjs/features/embed.d.ts +1 -1
- package/dist/cjs/features/embed.js +4 -3
- package/dist/cjs/features/embed.js.map +1 -1
- package/dist/cjs/features/index.d.cts +1 -1
- package/dist/cjs/features/index.d.ts +1 -1
- package/dist/cjs/features/index.js +4 -2
- package/dist/cjs/features/index.js.map +1 -1
- package/dist/cjs/features/links.js +38 -20
- package/dist/cjs/features/links.js.map +1 -1
- package/dist/cjs/features/tags.d.cts +2 -0
- package/dist/cjs/features/tags.d.ts +2 -0
- package/dist/cjs/features/tags.js +188 -0
- package/dist/cjs/features/tags.js.map +1 -0
- package/dist/cjs/features/types.d.cts +2 -2
- package/dist/cjs/features/types.d.ts +2 -2
- package/dist/cjs/index.d.cts +6 -6
- package/dist/cjs/index.d.ts +6 -6
- package/dist/cjs/index.js +15 -15
- package/dist/cjs/index.js.map +1 -1
- package/dist/cjs/scan.d.cts +3 -1
- package/dist/cjs/scan.d.ts +3 -1
- package/dist/cjs/scan.js +9 -6
- package/dist/cjs/scan.js.map +1 -1
- package/dist/cjs/sql-functions.d.cts +2 -0
- package/dist/cjs/sql-functions.d.ts +2 -0
- package/dist/cjs/sql-functions.js +64 -0
- package/dist/cjs/sql-functions.js.map +1 -0
- package/dist/cjs/watch.d.cts +1 -1
- package/dist/cjs/watch.d.ts +1 -1
- package/dist/cjs/watch.js +12 -12
- package/dist/cjs/watch.js.map +1 -1
- package/dist/esm/cli/download.js +1 -1
- package/dist/esm/cli/download.js.map +1 -1
- package/dist/esm/cli/init.js +1 -1
- package/dist/esm/cli/init.js.map +1 -1
- package/dist/esm/cli/map.js +1 -1
- package/dist/esm/cli/map.js.map +1 -1
- package/dist/esm/cli/named.js +1 -1
- package/dist/esm/cli/named.js.map +1 -1
- package/dist/esm/cli/path.js +1 -1
- package/dist/esm/cli/path.js.map +1 -1
- package/dist/esm/cli/peek.js +1 -1
- package/dist/esm/cli/peek.js.map +1 -1
- package/dist/esm/cli/related.js +1 -1
- package/dist/esm/cli/related.js.map +1 -1
- package/dist/esm/cli/search.js +1 -1
- package/dist/esm/cli/search.js.map +1 -1
- package/dist/esm/cli/shared.d.ts +2 -2
- package/dist/esm/cli/shared.js +2 -2
- package/dist/esm/cli/shared.js.map +1 -1
- package/dist/esm/cli/status.js +3 -3
- package/dist/esm/cli/status.js.map +1 -1
- package/dist/esm/cli/types.d.ts +1 -1
- package/dist/esm/cli/types.js.map +1 -1
- package/dist/esm/cli.js +1 -1
- package/dist/esm/cli.js.map +1 -1
- package/dist/esm/commands/index.d.ts +10 -0
- package/dist/esm/commands/index.js +8 -0
- package/dist/esm/commands/index.js.map +1 -0
- package/dist/esm/commands/map.d.ts +21 -0
- package/dist/esm/commands/map.js +56 -0
- package/dist/esm/commands/map.js.map +1 -0
- package/dist/esm/commands/peek.d.ts +20 -0
- package/dist/esm/commands/peek.js +63 -0
- package/dist/esm/commands/peek.js.map +1 -0
- package/dist/esm/commands/related.d.ts +6 -0
- package/dist/esm/commands/related.js +47 -0
- package/dist/esm/commands/related.js.map +1 -0
- package/dist/esm/commands/scope.d.ts +10 -0
- package/dist/esm/commands/scope.js +60 -0
- package/dist/esm/commands/scope.js.map +1 -0
- package/dist/esm/commands/search.d.ts +12 -0
- package/dist/esm/commands/search.js +258 -0
- package/dist/esm/commands/search.js.map +1 -0
- package/dist/esm/commands/status.d.ts +9 -0
- package/dist/esm/commands/status.js +16 -0
- package/dist/esm/commands/status.js.map +1 -0
- package/dist/esm/config/access.d.ts +19 -0
- package/dist/esm/config/access.js +89 -0
- package/dist/esm/config/access.js.map +1 -0
- package/dist/esm/config/index.d.ts +5 -0
- package/dist/esm/config/index.js +4 -0
- package/dist/esm/config/index.js.map +1 -0
- package/dist/esm/config/load.d.ts +8 -0
- package/dist/esm/config/load.js +242 -0
- package/dist/esm/config/load.js.map +1 -0
- package/dist/esm/config/resolve.d.ts +6 -0
- package/dist/esm/config/resolve.js +35 -0
- package/dist/esm/config/resolve.js.map +1 -0
- package/dist/{cjs/config.d.ts → esm/config/types.d.ts} +2 -32
- package/dist/esm/config/types.js +15 -0
- package/dist/esm/config/types.js.map +1 -0
- package/dist/esm/config/validate.d.ts +4 -0
- package/dist/esm/config/validate.js +226 -0
- package/dist/esm/config/validate.js.map +1 -0
- package/dist/esm/db/index.d.ts +4 -0
- package/dist/esm/db/index.js +5 -0
- package/dist/esm/db/index.js.map +1 -0
- package/dist/esm/{db.d.ts → db/open.d.ts} +2 -8
- package/dist/esm/db/open.js +160 -0
- package/dist/esm/db/open.js.map +1 -0
- package/dist/esm/db/reconcile.d.ts +9 -0
- package/dist/esm/{db.js → db/reconcile.js} +12 -216
- package/dist/esm/db/reconcile.js.map +1 -0
- package/dist/esm/db/shared.d.ts +5 -0
- package/dist/esm/db/shared.js +20 -0
- package/dist/esm/db/shared.js.map +1 -0
- package/dist/esm/features/embed.d.ts +1 -1
- package/dist/esm/features/embed.js +2 -1
- package/dist/esm/features/embed.js.map +1 -1
- package/dist/esm/features/index.d.ts +1 -1
- package/dist/esm/features/index.js +3 -1
- package/dist/esm/features/index.js.map +1 -1
- package/dist/esm/features/links.js +36 -21
- package/dist/esm/features/links.js.map +1 -1
- package/dist/esm/features/tags.d.ts +2 -0
- package/dist/esm/features/tags.js +80 -0
- package/dist/esm/features/tags.js.map +1 -0
- package/dist/esm/features/types.d.ts +2 -2
- package/dist/esm/features/types.js.map +1 -1
- package/dist/esm/index.d.ts +6 -6
- package/dist/esm/index.js +3 -3
- package/dist/esm/index.js.map +1 -1
- package/dist/esm/scan.d.ts +3 -1
- package/dist/esm/scan.js +5 -2
- package/dist/esm/scan.js.map +1 -1
- package/dist/esm/sql-functions.d.ts +2 -0
- package/dist/esm/sql-functions.js +47 -0
- package/dist/esm/sql-functions.js.map +1 -0
- package/dist/esm/watch.d.ts +1 -1
- package/dist/esm/watch.js +2 -2
- package/dist/esm/watch.js.map +1 -1
- package/package.json +2 -1
- package/schema.json +4 -0
- package/skills/sense/SKILL.md +3 -3
- package/skills/sense-bases/EXAMPLES.md +103 -0
- package/skills/sense-bases/SKILL.md +137 -0
- package/skills/sense-setup/SKILL.md +1 -1
- package/dist/cjs/commands.d.cts +0 -58
- package/dist/cjs/commands.d.ts +0 -58
- package/dist/cjs/commands.js.map +0 -1
- package/dist/cjs/config.js +0 -919
- package/dist/cjs/config.js.map +0 -1
- package/dist/cjs/db.js.map +0 -1
- package/dist/esm/commands.d.ts +0 -58
- package/dist/esm/commands.js +0 -486
- package/dist/esm/commands.js.map +0 -1
- package/dist/esm/config.js +0 -599
- package/dist/esm/config.js.map +0 -1
- package/dist/esm/db.js.map +0 -1
|
@@ -0,0 +1,226 @@
|
|
|
1
|
+
import { SenseError } from '../errors.js';
|
|
2
|
+
// Shape check for hand-edited files, so a typo names itself instead of surfacing as a
|
|
3
|
+
// TypeError. Unknown top-level keys warn (forward compat); unknown keys inside a block error.
|
|
4
|
+
const KNOWN_KEYS = new Set([
|
|
5
|
+
'$schema',
|
|
6
|
+
'version',
|
|
7
|
+
'presets',
|
|
8
|
+
'features',
|
|
9
|
+
'embed',
|
|
10
|
+
'content',
|
|
11
|
+
'queries'
|
|
12
|
+
]);
|
|
13
|
+
const KNOWN_PRESET_KEYS = new Set([
|
|
14
|
+
'include',
|
|
15
|
+
'exclude',
|
|
16
|
+
'k',
|
|
17
|
+
'semantic',
|
|
18
|
+
'where'
|
|
19
|
+
]);
|
|
20
|
+
const KNOWN_FEATURE_KEYS = new Set([
|
|
21
|
+
'links',
|
|
22
|
+
'sections',
|
|
23
|
+
'tags',
|
|
24
|
+
'rank'
|
|
25
|
+
]);
|
|
26
|
+
const KNOWN_EMBED_KEYS = new Set([
|
|
27
|
+
'model',
|
|
28
|
+
'type',
|
|
29
|
+
'url',
|
|
30
|
+
'key'
|
|
31
|
+
]);
|
|
32
|
+
const KNOWN_CONTENT_KEYS = new Set([
|
|
33
|
+
'tokenize'
|
|
34
|
+
]);
|
|
35
|
+
const SAVED_SEARCH_KEYS = new Set([
|
|
36
|
+
'search',
|
|
37
|
+
'preset',
|
|
38
|
+
'include',
|
|
39
|
+
'exclude',
|
|
40
|
+
'where',
|
|
41
|
+
'k'
|
|
42
|
+
]);
|
|
43
|
+
export function unknownConfigKeys(cfg) {
|
|
44
|
+
return Object.keys(cfg).filter((k)=>!KNOWN_KEYS.has(k));
|
|
45
|
+
}
|
|
46
|
+
function isNonEmptyStringArray(value) {
|
|
47
|
+
return Array.isArray(value) && value.length > 0 && value.every((g)=>typeof g === 'string' && g.length > 0);
|
|
48
|
+
}
|
|
49
|
+
// Pre-v3 shape check, just enough for migrateConfig to run safely (it reads scan.include
|
|
50
|
+
// directly). The full v3 shape is checked by validateConfig once migration has produced it.
|
|
51
|
+
export function validateLegacyScan(parsed, configPath) {
|
|
52
|
+
const cfg = typeof parsed === 'object' && parsed !== null && !Array.isArray(parsed) ? parsed : null;
|
|
53
|
+
const scan = cfg === null || cfg === void 0 ? void 0 : cfg.scan;
|
|
54
|
+
if (!cfg || !scan || !isNonEmptyStringArray(scan.include)) {
|
|
55
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: scan.include must be a non-empty array of glob strings`);
|
|
56
|
+
}
|
|
57
|
+
}
|
|
58
|
+
// Only the shape is checked here. Whether the linked SQLite accepts the tokenizer is settled
|
|
59
|
+
// by probing it in db.ts, so this never has to carry a table of which version added what.
|
|
60
|
+
function validateContentBlock(value, configPath) {
|
|
61
|
+
if (typeof value !== 'object' || value === null || Array.isArray(value)) {
|
|
62
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: content must be an object of { tokenize?: string }`);
|
|
63
|
+
}
|
|
64
|
+
const block = value;
|
|
65
|
+
const unknown = Object.keys(block).filter((k)=>!KNOWN_CONTENT_KEYS.has(k));
|
|
66
|
+
if (unknown.length > 0) {
|
|
67
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: content has unknown key(s) ${unknown.join(', ')}; content takes tokenize`);
|
|
68
|
+
}
|
|
69
|
+
if (block.tokenize !== undefined && (typeof block.tokenize !== 'string' || block.tokenize.trim() === '')) {
|
|
70
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: content.tokenize must be a non-empty string, e.g. "trigram" or "unicode61 tokenchars '-_'"`);
|
|
71
|
+
}
|
|
72
|
+
}
|
|
73
|
+
function validateFeaturesBlock(value, configPath) {
|
|
74
|
+
if (typeof value !== 'object' || value === null || Array.isArray(value)) {
|
|
75
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: features must be an object of name -> boolean`);
|
|
76
|
+
}
|
|
77
|
+
const block = value;
|
|
78
|
+
const unknown = Object.keys(block).filter((k)=>!KNOWN_FEATURE_KEYS.has(k));
|
|
79
|
+
if (unknown.length > 0) {
|
|
80
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: features has unknown key(s) ${unknown.join(', ')}; embed is not a features key -- see the top-level "embed" block`);
|
|
81
|
+
}
|
|
82
|
+
for (const [name, v] of Object.entries(block)){
|
|
83
|
+
if (typeof v !== 'boolean') {
|
|
84
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: features.${name} must be a boolean`);
|
|
85
|
+
}
|
|
86
|
+
}
|
|
87
|
+
}
|
|
88
|
+
function validateEmbedBlock(value, configPath) {
|
|
89
|
+
if (typeof value !== 'object' || value === null || Array.isArray(value)) {
|
|
90
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: embed must be an object of { model?, type?: "static"|"api", url?, key? }`);
|
|
91
|
+
}
|
|
92
|
+
const embed = value;
|
|
93
|
+
const unknown = Object.keys(embed).filter((k)=>!KNOWN_EMBED_KEYS.has(k));
|
|
94
|
+
if (unknown.length > 0) {
|
|
95
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: embed has unknown key(s) ${unknown.join(', ')}`);
|
|
96
|
+
}
|
|
97
|
+
if (embed.model !== undefined && typeof embed.model !== 'string') {
|
|
98
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: embed.model must be a string`);
|
|
99
|
+
}
|
|
100
|
+
// The block exists to name a model; provider settings without one would read as "vectors
|
|
101
|
+
// configured" while embedding stays off, which is the ambiguity v4 removed.
|
|
102
|
+
if (embed.model === undefined || embed.model === '') {
|
|
103
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: embed.model is required when the "embed" block is present (it is what turns vectors on); remove the block to index without vectors`);
|
|
104
|
+
}
|
|
105
|
+
if (embed.type !== undefined && embed.type !== 'static' && embed.type !== 'api') {
|
|
106
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: embed.type must be "static" or "api"`);
|
|
107
|
+
}
|
|
108
|
+
if (embed.url !== undefined && typeof embed.url !== 'string') {
|
|
109
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: embed.url must be a string`);
|
|
110
|
+
}
|
|
111
|
+
if (embed.key !== undefined && typeof embed.key !== 'string') {
|
|
112
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: embed.key must be a string`);
|
|
113
|
+
}
|
|
114
|
+
}
|
|
115
|
+
function validatePreset(name, value, configPath) {
|
|
116
|
+
if (typeof value !== 'object' || value === null || Array.isArray(value)) {
|
|
117
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name} must be an object`);
|
|
118
|
+
}
|
|
119
|
+
const preset = value;
|
|
120
|
+
const unknown = Object.keys(preset).filter((k)=>!KNOWN_PRESET_KEYS.has(k));
|
|
121
|
+
if (unknown.length > 0) {
|
|
122
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name} has unknown key(s) ${unknown.join(', ')}; a preset takes include, exclude, k, semantic, where`);
|
|
123
|
+
}
|
|
124
|
+
if (!isNonEmptyStringArray(preset.include)) {
|
|
125
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.include must be a non-empty array of glob strings`);
|
|
126
|
+
}
|
|
127
|
+
if (preset.exclude !== undefined && !isNonEmptyStringArray(preset.exclude)) {
|
|
128
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.exclude must be a non-empty array of glob strings`);
|
|
129
|
+
}
|
|
130
|
+
if (preset.semantic !== undefined && typeof preset.semantic !== 'boolean') {
|
|
131
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.semantic must be a boolean`);
|
|
132
|
+
}
|
|
133
|
+
if (preset.k !== undefined && (typeof preset.k !== 'number' || !Number.isInteger(preset.k) || preset.k <= 0)) {
|
|
134
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.k must be a positive integer`);
|
|
135
|
+
}
|
|
136
|
+
if (preset.where !== undefined && typeof preset.where !== 'string') {
|
|
137
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.where must be a SQL condition string`);
|
|
138
|
+
}
|
|
139
|
+
}
|
|
140
|
+
// A queries.<name> entry: { sql } or a saved search { search, preset?, include?, exclude?, where?, k? }.
|
|
141
|
+
function validateSavedQuery(name, value, configPath) {
|
|
142
|
+
// A bare string used to mean SQL. It now fails rather than being inferred: an entry says
|
|
143
|
+
// which of the two verbs it runs, the same choice the CLI makes explicit.
|
|
144
|
+
if (typeof value === 'string') {
|
|
145
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must say which verb it runs: { "sql": ${JSON.stringify(value.length > 40 ? `${value.slice(0, 37)}...` : value)} } to run it as SQL, or { "search": "..." } for a ranked search`);
|
|
146
|
+
}
|
|
147
|
+
if (typeof value !== 'object' || value === null || Array.isArray(value)) {
|
|
148
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must be { sql } or { search, preset?, include?, exclude?, where?, k? }`);
|
|
149
|
+
}
|
|
150
|
+
const entry = value;
|
|
151
|
+
if ('sql' in entry) {
|
|
152
|
+
const unknown = Object.keys(entry).filter((k)=>k !== 'sql');
|
|
153
|
+
if (unknown.length > 0) {
|
|
154
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} has unknown key(s) ${unknown.join(', ')}; a { sql } query takes only sql`);
|
|
155
|
+
}
|
|
156
|
+
if (typeof entry.sql !== 'string' || entry.sql.trim() === '') {
|
|
157
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.sql must be a non-empty string`);
|
|
158
|
+
}
|
|
159
|
+
return;
|
|
160
|
+
}
|
|
161
|
+
if ('search' in entry) {
|
|
162
|
+
const unknown = Object.keys(entry).filter((k)=>!SAVED_SEARCH_KEYS.has(k));
|
|
163
|
+
if (unknown.length > 0) {
|
|
164
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} has unknown key(s) ${unknown.join(', ')}; a saved search takes search, preset, include, exclude, where, k`);
|
|
165
|
+
}
|
|
166
|
+
// A saved entry saves a question; a scope without a question is just flags.
|
|
167
|
+
if (typeof entry.search !== 'string' || entry.search.trim() === '') {
|
|
168
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.search must be non-empty text`);
|
|
169
|
+
}
|
|
170
|
+
if (entry.preset !== undefined && (typeof entry.preset !== 'string' || entry.preset.length === 0)) {
|
|
171
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.preset must be a preset name`);
|
|
172
|
+
}
|
|
173
|
+
if (entry.include !== undefined && !isNonEmptyStringArray(entry.include)) {
|
|
174
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.include must be a non-empty array of glob strings`);
|
|
175
|
+
}
|
|
176
|
+
if (entry.exclude !== undefined && !isNonEmptyStringArray(entry.exclude)) {
|
|
177
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.exclude must be a non-empty array of glob strings`);
|
|
178
|
+
}
|
|
179
|
+
if (entry.where !== undefined && typeof entry.where !== 'string') {
|
|
180
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.where must be a SQL condition string`);
|
|
181
|
+
}
|
|
182
|
+
if (entry.k !== undefined && (typeof entry.k !== 'number' || !Number.isInteger(entry.k) || entry.k <= 0)) {
|
|
183
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.k must be a positive integer`);
|
|
184
|
+
}
|
|
185
|
+
return;
|
|
186
|
+
}
|
|
187
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must be { sql } or { search, preset?, include?, exclude?, where?, k? }`);
|
|
188
|
+
}
|
|
189
|
+
export function validateConfig(parsed, configPath) {
|
|
190
|
+
if (typeof parsed !== 'object' || parsed === null || Array.isArray(parsed)) {
|
|
191
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: config must be a JSON object`);
|
|
192
|
+
}
|
|
193
|
+
const cfg = parsed;
|
|
194
|
+
// `checks` is rejected by name, not warned: silence would hide that assertions are gone.
|
|
195
|
+
if (cfg.checks !== undefined) {
|
|
196
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: checks was removed in v3 -- sense check no longer asserts on saved queries; a returned row set is the reader's judgment`);
|
|
197
|
+
}
|
|
198
|
+
const presets = cfg.presets;
|
|
199
|
+
if (!presets || typeof presets !== 'object' || Array.isArray(presets) || Object.keys(presets).length === 0) {
|
|
200
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: presets must be a non-empty object of preset name -> { include, exclude?, k?, semantic?, where? }`);
|
|
201
|
+
}
|
|
202
|
+
for (const [name, value] of Object.entries(presets)){
|
|
203
|
+
validatePreset(name, value, configPath);
|
|
204
|
+
}
|
|
205
|
+
if (presets.default === undefined) {
|
|
206
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: presets must include a "default" preset`);
|
|
207
|
+
}
|
|
208
|
+
if (cfg.queries === undefined) cfg.queries = {};
|
|
209
|
+
if (typeof cfg.queries !== 'object' || cfg.queries === null || Array.isArray(cfg.queries)) {
|
|
210
|
+
throw new SenseError('CONFIG_INVALID', `${configPath}: queries must be an object of name -> { sql } or { search }`);
|
|
211
|
+
}
|
|
212
|
+
const queries = cfg.queries;
|
|
213
|
+
for (const [name, value] of Object.entries(queries)){
|
|
214
|
+
validateSavedQuery(name, value, configPath);
|
|
215
|
+
}
|
|
216
|
+
if (cfg.features !== undefined) {
|
|
217
|
+
validateFeaturesBlock(cfg.features, configPath);
|
|
218
|
+
}
|
|
219
|
+
if (cfg.embed !== undefined) {
|
|
220
|
+
validateEmbedBlock(cfg.embed, configPath);
|
|
221
|
+
}
|
|
222
|
+
if (cfg.content !== undefined) {
|
|
223
|
+
validateContentBlock(cfg.content, configPath);
|
|
224
|
+
}
|
|
225
|
+
return cfg;
|
|
226
|
+
}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/config/validate.ts"],"sourcesContent":["import { SenseError } from '../errors.ts';\nimport type { Config } from './types.ts';\n\n// Shape check for hand-edited files, so a typo names itself instead of surfacing as a\n// TypeError. Unknown top-level keys warn (forward compat); unknown keys inside a block error.\nconst KNOWN_KEYS = new Set(['$schema', 'version', 'presets', 'features', 'embed', 'content', 'queries']);\nconst KNOWN_PRESET_KEYS = new Set(['include', 'exclude', 'k', 'semantic', 'where']);\nconst KNOWN_FEATURE_KEYS = new Set(['links', 'sections', 'tags', 'rank']);\nconst KNOWN_EMBED_KEYS = new Set(['model', 'type', 'url', 'key']);\nconst KNOWN_CONTENT_KEYS = new Set(['tokenize']);\nconst SAVED_SEARCH_KEYS = new Set(['search', 'preset', 'include', 'exclude', 'where', 'k']);\n\nexport function unknownConfigKeys(cfg: Record<string, unknown>): string[] {\n return Object.keys(cfg).filter((k) => !KNOWN_KEYS.has(k));\n}\n\nfunction isNonEmptyStringArray(value: unknown): value is string[] {\n return Array.isArray(value) && value.length > 0 && value.every((g) => typeof g === 'string' && g.length > 0);\n}\n\n// Pre-v3 shape check, just enough for migrateConfig to run safely (it reads scan.include\n// directly). The full v3 shape is checked by validateConfig once migration has produced it.\nexport function validateLegacyScan(parsed: unknown, configPath: string): void {\n const cfg = typeof parsed === 'object' && parsed !== null && !Array.isArray(parsed) ? (parsed as Record<string, unknown>) : null;\n const scan = cfg?.scan as { include?: unknown } | undefined;\n if (!cfg || !scan || !isNonEmptyStringArray(scan.include)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: scan.include must be a non-empty array of glob strings`);\n }\n}\n\n// Only the shape is checked here. Whether the linked SQLite accepts the tokenizer is settled\n// by probing it in db.ts, so this never has to carry a table of which version added what.\nfunction validateContentBlock(value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: content must be an object of { tokenize?: string }`);\n }\n const block = value as Record<string, unknown>;\n const unknown = Object.keys(block).filter((k) => !KNOWN_CONTENT_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: content has unknown key(s) ${unknown.join(', ')}; content takes tokenize`);\n }\n if (block.tokenize !== undefined && (typeof block.tokenize !== 'string' || block.tokenize.trim() === '')) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: content.tokenize must be a non-empty string, e.g. \"trigram\" or \"unicode61 tokenchars '-_'\"`);\n }\n}\n\nfunction validateFeaturesBlock(value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: features must be an object of name -> boolean`);\n }\n const block = value as Record<string, unknown>;\n const unknown = Object.keys(block).filter((k) => !KNOWN_FEATURE_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: features has unknown key(s) ${unknown.join(', ')}; embed is not a features key -- see the top-level \"embed\" block`);\n }\n for (const [name, v] of Object.entries(block)) {\n if (typeof v !== 'boolean') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: features.${name} must be a boolean`);\n }\n }\n}\n\nfunction validateEmbedBlock(value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed must be an object of { model?, type?: \"static\"|\"api\", url?, key? }`);\n }\n const embed = value as Record<string, unknown>;\n const unknown = Object.keys(embed).filter((k) => !KNOWN_EMBED_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed has unknown key(s) ${unknown.join(', ')}`);\n }\n if (embed.model !== undefined && typeof embed.model !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.model must be a string`);\n }\n // The block exists to name a model; provider settings without one would read as \"vectors\n // configured\" while embedding stays off, which is the ambiguity v4 removed.\n if (embed.model === undefined || embed.model === '') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.model is required when the \"embed\" block is present (it is what turns vectors on); remove the block to index without vectors`);\n }\n if (embed.type !== undefined && embed.type !== 'static' && embed.type !== 'api') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.type must be \"static\" or \"api\"`);\n }\n if (embed.url !== undefined && typeof embed.url !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.url must be a string`);\n }\n if (embed.key !== undefined && typeof embed.key !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.key must be a string`);\n }\n}\n\nfunction validatePreset(name: string, value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name} must be an object`);\n }\n const preset = value as Record<string, unknown>;\n const unknown = Object.keys(preset).filter((k) => !KNOWN_PRESET_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name} has unknown key(s) ${unknown.join(', ')}; a preset takes include, exclude, k, semantic, where`);\n }\n if (!isNonEmptyStringArray(preset.include)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.include must be a non-empty array of glob strings`);\n }\n if (preset.exclude !== undefined && !isNonEmptyStringArray(preset.exclude)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.exclude must be a non-empty array of glob strings`);\n }\n if (preset.semantic !== undefined && typeof preset.semantic !== 'boolean') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.semantic must be a boolean`);\n }\n if (preset.k !== undefined && (typeof preset.k !== 'number' || !Number.isInteger(preset.k) || preset.k <= 0)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.k must be a positive integer`);\n }\n if (preset.where !== undefined && typeof preset.where !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.where must be a SQL condition string`);\n }\n}\n\n// A queries.<name> entry: { sql } or a saved search { search, preset?, include?, exclude?, where?, k? }.\nfunction validateSavedQuery(name: string, value: unknown, configPath: string): void {\n // A bare string used to mean SQL. It now fails rather than being inferred: an entry says\n // which of the two verbs it runs, the same choice the CLI makes explicit.\n if (typeof value === 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must say which verb it runs: { \"sql\": ${JSON.stringify(value.length > 40 ? `${value.slice(0, 37)}...` : value)} } to run it as SQL, or { \"search\": \"...\" } for a ranked search`);\n }\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must be { sql } or { search, preset?, include?, exclude?, where?, k? }`);\n }\n const entry = value as Record<string, unknown>;\n if ('sql' in entry) {\n const unknown = Object.keys(entry).filter((k) => k !== 'sql');\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} has unknown key(s) ${unknown.join(', ')}; a { sql } query takes only sql`);\n }\n if (typeof entry.sql !== 'string' || entry.sql.trim() === '') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.sql must be a non-empty string`);\n }\n return;\n }\n if ('search' in entry) {\n const unknown = Object.keys(entry).filter((k) => !SAVED_SEARCH_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} has unknown key(s) ${unknown.join(', ')}; a saved search takes search, preset, include, exclude, where, k`);\n }\n // A saved entry saves a question; a scope without a question is just flags.\n if (typeof entry.search !== 'string' || entry.search.trim() === '') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.search must be non-empty text`);\n }\n if (entry.preset !== undefined && (typeof entry.preset !== 'string' || entry.preset.length === 0)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.preset must be a preset name`);\n }\n if (entry.include !== undefined && !isNonEmptyStringArray(entry.include)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.include must be a non-empty array of glob strings`);\n }\n if (entry.exclude !== undefined && !isNonEmptyStringArray(entry.exclude)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.exclude must be a non-empty array of glob strings`);\n }\n if (entry.where !== undefined && typeof entry.where !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.where must be a SQL condition string`);\n }\n if (entry.k !== undefined && (typeof entry.k !== 'number' || !Number.isInteger(entry.k) || entry.k <= 0)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.k must be a positive integer`);\n }\n return;\n }\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must be { sql } or { search, preset?, include?, exclude?, where?, k? }`);\n}\n\nexport function validateConfig(parsed: unknown, configPath: string): Config {\n if (typeof parsed !== 'object' || parsed === null || Array.isArray(parsed)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: config must be a JSON object`);\n }\n const cfg = parsed as Record<string, unknown>;\n\n // `checks` is rejected by name, not warned: silence would hide that assertions are gone.\n if (cfg.checks !== undefined) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: checks was removed in v3 -- sense check no longer asserts on saved queries; a returned row set is the reader's judgment`);\n }\n\n const presets = cfg.presets as Record<string, unknown> | undefined;\n if (!presets || typeof presets !== 'object' || Array.isArray(presets) || Object.keys(presets).length === 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets must be a non-empty object of preset name -> { include, exclude?, k?, semantic?, where? }`);\n }\n for (const [name, value] of Object.entries(presets)) {\n validatePreset(name, value, configPath);\n }\n if (presets.default === undefined) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets must include a \"default\" preset`);\n }\n\n if (cfg.queries === undefined) cfg.queries = {};\n if (typeof cfg.queries !== 'object' || cfg.queries === null || Array.isArray(cfg.queries)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries must be an object of name -> { sql } or { search }`);\n }\n const queries = cfg.queries as Record<string, unknown>;\n for (const [name, value] of Object.entries(queries)) {\n validateSavedQuery(name, value, configPath);\n }\n\n if (cfg.features !== undefined) {\n validateFeaturesBlock(cfg.features, configPath);\n }\n if (cfg.embed !== undefined) {\n validateEmbedBlock(cfg.embed, configPath);\n }\n if (cfg.content !== undefined) {\n validateContentBlock(cfg.content, configPath);\n }\n\n return cfg as unknown as Config;\n}\n"],"names":["SenseError","KNOWN_KEYS","Set","KNOWN_PRESET_KEYS","KNOWN_FEATURE_KEYS","KNOWN_EMBED_KEYS","KNOWN_CONTENT_KEYS","SAVED_SEARCH_KEYS","unknownConfigKeys","cfg","Object","keys","filter","k","has","isNonEmptyStringArray","value","Array","isArray","length","every","g","validateLegacyScan","parsed","configPath","scan","include","validateContentBlock","block","unknown","join","tokenize","undefined","trim","validateFeaturesBlock","name","v","entries","validateEmbedBlock","embed","model","type","url","key","validatePreset","preset","exclude","semantic","Number","isInteger","where","validateSavedQuery","JSON","stringify","slice","entry","sql","search","validateConfig","checks","presets","default","queries","features","content"],"mappings":"AAAA,SAASA,UAAU,QAAQ,eAAe;AAG1C,sFAAsF;AACtF,8FAA8F;AAC9F,MAAMC,aAAa,IAAIC,IAAI;IAAC;IAAW;IAAW;IAAW;IAAY;IAAS;IAAW;CAAU;AACvG,MAAMC,oBAAoB,IAAID,IAAI;IAAC;IAAW;IAAW;IAAK;IAAY;CAAQ;AAClF,MAAME,qBAAqB,IAAIF,IAAI;IAAC;IAAS;IAAY;IAAQ;CAAO;AACxE,MAAMG,mBAAmB,IAAIH,IAAI;IAAC;IAAS;IAAQ;IAAO;CAAM;AAChE,MAAMI,qBAAqB,IAAIJ,IAAI;IAAC;CAAW;AAC/C,MAAMK,oBAAoB,IAAIL,IAAI;IAAC;IAAU;IAAU;IAAW;IAAW;IAAS;CAAI;AAE1F,OAAO,SAASM,kBAAkBC,GAA4B;IAC5D,OAAOC,OAAOC,IAAI,CAACF,KAAKG,MAAM,CAAC,CAACC,IAAM,CAACZ,WAAWa,GAAG,CAACD;AACxD;AAEA,SAASE,sBAAsBC,KAAc;IAC3C,OAAOC,MAAMC,OAAO,CAACF,UAAUA,MAAMG,MAAM,GAAG,KAAKH,MAAMI,KAAK,CAAC,CAACC,IAAM,OAAOA,MAAM,YAAYA,EAAEF,MAAM,GAAG;AAC5G;AAEA,yFAAyF;AACzF,4FAA4F;AAC5F,OAAO,SAASG,mBAAmBC,MAAe,EAAEC,UAAkB;IACpE,MAAMf,MAAM,OAAOc,WAAW,YAAYA,WAAW,QAAQ,CAACN,MAAMC,OAAO,CAACK,UAAWA,SAAqC;IAC5H,MAAME,OAAOhB,gBAAAA,0BAAAA,IAAKgB,IAAI;IACtB,IAAI,CAAChB,OAAO,CAACgB,QAAQ,CAACV,sBAAsBU,KAAKC,OAAO,GAAG;QACzD,MAAM,IAAI1B,WAAW,kBAAkB,GAAGwB,WAAW,wDAAwD,CAAC;IAChH;AACF;AAEA,6FAA6F;AAC7F,0FAA0F;AAC1F,SAASG,qBAAqBX,KAAc,EAAEQ,UAAkB;IAC9D,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,oDAAoD,CAAC;IAC5G;IACA,MAAMI,QAAQZ;IACd,MAAMa,UAAUnB,OAAOC,IAAI,CAACiB,OAAOhB,MAAM,CAAC,CAACC,IAAM,CAACP,mBAAmBQ,GAAG,CAACD;IACzE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,6BAA6B,EAAEK,QAAQC,IAAI,CAAC,MAAM,wBAAwB,CAAC;IAClI;IACA,IAAIF,MAAMG,QAAQ,KAAKC,aAAc,CAAA,OAAOJ,MAAMG,QAAQ,KAAK,YAAYH,MAAMG,QAAQ,CAACE,IAAI,OAAO,EAAC,GAAI;QACxG,MAAM,IAAIjC,WAAW,kBAAkB,GAAGwB,WAAW,4FAA4F,CAAC;IACpJ;AACF;AAEA,SAASU,sBAAsBlB,KAAc,EAAEQ,UAAkB;IAC/D,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,+CAA+C,CAAC;IACvG;IACA,MAAMI,QAAQZ;IACd,MAAMa,UAAUnB,OAAOC,IAAI,CAACiB,OAAOhB,MAAM,CAAC,CAACC,IAAM,CAACT,mBAAmBU,GAAG,CAACD;IACzE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,8BAA8B,EAAEK,QAAQC,IAAI,CAAC,MAAM,gEAAgE,CAAC;IAC3K;IACA,KAAK,MAAM,CAACK,MAAMC,EAAE,IAAI1B,OAAO2B,OAAO,CAACT,OAAQ;QAC7C,IAAI,OAAOQ,MAAM,WAAW;YAC1B,MAAM,IAAIpC,WAAW,kBAAkB,GAAGwB,WAAW,WAAW,EAAEW,KAAK,kBAAkB,CAAC;QAC5F;IACF;AACF;AAEA,SAASG,mBAAmBtB,KAAc,EAAEQ,UAAkB;IAC5D,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,0EAA0E,CAAC;IAClI;IACA,MAAMe,QAAQvB;IACd,MAAMa,UAAUnB,OAAOC,IAAI,CAAC4B,OAAO3B,MAAM,CAAC,CAACC,IAAM,CAACR,iBAAiBS,GAAG,CAACD;IACvE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,2BAA2B,EAAEK,QAAQC,IAAI,CAAC,OAAO;IACxG;IACA,IAAIS,MAAMC,KAAK,KAAKR,aAAa,OAAOO,MAAMC,KAAK,KAAK,UAAU;QAChE,MAAM,IAAIxC,WAAW,kBAAkB,GAAGwB,WAAW,8BAA8B,CAAC;IACtF;IACA,yFAAyF;IACzF,4EAA4E;IAC5E,IAAIe,MAAMC,KAAK,KAAKR,aAAaO,MAAMC,KAAK,KAAK,IAAI;QACnD,MAAM,IAAIxC,WAAW,kBAAkB,GAAGwB,WAAW,oIAAoI,CAAC;IAC5L;IACA,IAAIe,MAAME,IAAI,KAAKT,aAAaO,MAAME,IAAI,KAAK,YAAYF,MAAME,IAAI,KAAK,OAAO;QAC/E,MAAM,IAAIzC,WAAW,kBAAkB,GAAGwB,WAAW,sCAAsC,CAAC;IAC9F;IACA,IAAIe,MAAMG,GAAG,KAAKV,aAAa,OAAOO,MAAMG,GAAG,KAAK,UAAU;QAC5D,MAAM,IAAI1C,WAAW,kBAAkB,GAAGwB,WAAW,4BAA4B,CAAC;IACpF;IACA,IAAIe,MAAMI,GAAG,KAAKX,aAAa,OAAOO,MAAMI,GAAG,KAAK,UAAU;QAC5D,MAAM,IAAI3C,WAAW,kBAAkB,GAAGwB,WAAW,4BAA4B,CAAC;IACpF;AACF;AAEA,SAASoB,eAAeT,IAAY,EAAEnB,KAAc,EAAEQ,UAAkB;IACtE,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kBAAkB,CAAC;IAC3F;IACA,MAAMU,SAAS7B;IACf,MAAMa,UAAUnB,OAAOC,IAAI,CAACkC,QAAQjC,MAAM,CAAC,CAACC,IAAM,CAACV,kBAAkBW,GAAG,CAACD;IACzE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,oBAAoB,EAAEN,QAAQC,IAAI,CAAC,MAAM,qDAAqD,CAAC;IACvK;IACA,IAAI,CAACf,sBAAsB8B,OAAOnB,OAAO,GAAG;QAC1C,MAAM,IAAI1B,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;IAC3H;IACA,IAAIU,OAAOC,OAAO,KAAKd,aAAa,CAACjB,sBAAsB8B,OAAOC,OAAO,GAAG;QAC1E,MAAM,IAAI9C,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;IAC3H;IACA,IAAIU,OAAOE,QAAQ,KAAKf,aAAa,OAAOa,OAAOE,QAAQ,KAAK,WAAW;QACzE,MAAM,IAAI/C,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,2BAA2B,CAAC;IACpG;IACA,IAAIU,OAAOhC,CAAC,KAAKmB,aAAc,CAAA,OAAOa,OAAOhC,CAAC,KAAK,YAAY,CAACmC,OAAOC,SAAS,CAACJ,OAAOhC,CAAC,KAAKgC,OAAOhC,CAAC,IAAI,CAAA,GAAI;QAC5G,MAAM,IAAIb,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,6BAA6B,CAAC;IACtG;IACA,IAAIU,OAAOK,KAAK,KAAKlB,aAAa,OAAOa,OAAOK,KAAK,KAAK,UAAU;QAClE,MAAM,IAAIlD,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,qCAAqC,CAAC;IAC9G;AACF;AAEA,yGAAyG;AACzG,SAASgB,mBAAmBhB,IAAY,EAAEnB,KAAc,EAAEQ,UAAkB;IAC1E,yFAAyF;IACzF,0EAA0E;IAC1E,IAAI,OAAOR,UAAU,UAAU;QAC7B,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,uCAAuC,EAAEiB,KAAKC,SAAS,CAACrC,MAAMG,MAAM,GAAG,KAAK,GAAGH,MAAMsC,KAAK,CAAC,GAAG,IAAI,GAAG,CAAC,GAAGtC,OAAO,+DAA+D,CAAC;IACxP;IACA,IAAI,OAAOA,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,uEAAuE,CAAC;IAChJ;IACA,MAAMoB,QAAQvC;IACd,IAAI,SAASuC,OAAO;QAClB,MAAM1B,UAAUnB,OAAOC,IAAI,CAAC4C,OAAO3C,MAAM,CAAC,CAACC,IAAMA,MAAM;QACvD,IAAIgB,QAAQV,MAAM,GAAG,GAAG;YACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,oBAAoB,EAAEN,QAAQC,IAAI,CAAC,MAAM,gCAAgC,CAAC;QAClJ;QACA,IAAI,OAAOyB,MAAMC,GAAG,KAAK,YAAYD,MAAMC,GAAG,CAACvB,IAAI,OAAO,IAAI;YAC5D,MAAM,IAAIjC,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,+BAA+B,CAAC;QACxG;QACA;IACF;IACA,IAAI,YAAYoB,OAAO;QACrB,MAAM1B,UAAUnB,OAAOC,IAAI,CAAC4C,OAAO3C,MAAM,CAAC,CAACC,IAAM,CAACN,kBAAkBO,GAAG,CAACD;QACxE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;YACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,oBAAoB,EAAEN,QAAQC,IAAI,CAAC,MAAM,iEAAiE,CAAC;QACnL;QACA,4EAA4E;QAC5E,IAAI,OAAOyB,MAAME,MAAM,KAAK,YAAYF,MAAME,MAAM,CAACxB,IAAI,OAAO,IAAI;YAClE,MAAM,IAAIjC,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,8BAA8B,CAAC;QACvG;QACA,IAAIoB,MAAMV,MAAM,KAAKb,aAAc,CAAA,OAAOuB,MAAMV,MAAM,KAAK,YAAYU,MAAMV,MAAM,CAAC1B,MAAM,KAAK,CAAA,GAAI;YACjG,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,6BAA6B,CAAC;QACtG;QACA,IAAIoB,MAAM7B,OAAO,KAAKM,aAAa,CAACjB,sBAAsBwC,MAAM7B,OAAO,GAAG;YACxE,MAAM,IAAI1B,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;QAC3H;QACA,IAAIoB,MAAMT,OAAO,KAAKd,aAAa,CAACjB,sBAAsBwC,MAAMT,OAAO,GAAG;YACxE,MAAM,IAAI9C,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;QAC3H;QACA,IAAIoB,MAAML,KAAK,KAAKlB,aAAa,OAAOuB,MAAML,KAAK,KAAK,UAAU;YAChE,MAAM,IAAIlD,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,qCAAqC,CAAC;QAC9G;QACA,IAAIoB,MAAM1C,CAAC,KAAKmB,aAAc,CAAA,OAAOuB,MAAM1C,CAAC,KAAK,YAAY,CAACmC,OAAOC,SAAS,CAACM,MAAM1C,CAAC,KAAK0C,MAAM1C,CAAC,IAAI,CAAA,GAAI;YACxG,MAAM,IAAIb,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,6BAA6B,CAAC;QACtG;QACA;IACF;IACA,MAAM,IAAInC,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,uEAAuE,CAAC;AAChJ;AAEA,OAAO,SAASuB,eAAenC,MAAe,EAAEC,UAAkB;IAChE,IAAI,OAAOD,WAAW,YAAYA,WAAW,QAAQN,MAAMC,OAAO,CAACK,SAAS;QAC1E,MAAM,IAAIvB,WAAW,kBAAkB,GAAGwB,WAAW,8BAA8B,CAAC;IACtF;IACA,MAAMf,MAAMc;IAEZ,yFAAyF;IACzF,IAAId,IAAIkD,MAAM,KAAK3B,WAAW;QAC5B,MAAM,IAAIhC,WAAW,kBAAkB,GAAGwB,WAAW,yHAAyH,CAAC;IACjL;IAEA,MAAMoC,UAAUnD,IAAImD,OAAO;IAC3B,IAAI,CAACA,WAAW,OAAOA,YAAY,YAAY3C,MAAMC,OAAO,CAAC0C,YAAYlD,OAAOC,IAAI,CAACiD,SAASzC,MAAM,KAAK,GAAG;QAC1G,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,mGAAmG,CAAC;IAC3J;IACA,KAAK,MAAM,CAACW,MAAMnB,MAAM,IAAIN,OAAO2B,OAAO,CAACuB,SAAU;QACnDhB,eAAeT,MAAMnB,OAAOQ;IAC9B;IACA,IAAIoC,QAAQC,OAAO,KAAK7B,WAAW;QACjC,MAAM,IAAIhC,WAAW,kBAAkB,GAAGwB,WAAW,yCAAyC,CAAC;IACjG;IAEA,IAAIf,IAAIqD,OAAO,KAAK9B,WAAWvB,IAAIqD,OAAO,GAAG,CAAC;IAC9C,IAAI,OAAOrD,IAAIqD,OAAO,KAAK,YAAYrD,IAAIqD,OAAO,KAAK,QAAQ7C,MAAMC,OAAO,CAACT,IAAIqD,OAAO,GAAG;QACzF,MAAM,IAAI9D,WAAW,kBAAkB,GAAGwB,WAAW,4DAA4D,CAAC;IACpH;IACA,MAAMsC,UAAUrD,IAAIqD,OAAO;IAC3B,KAAK,MAAM,CAAC3B,MAAMnB,MAAM,IAAIN,OAAO2B,OAAO,CAACyB,SAAU;QACnDX,mBAAmBhB,MAAMnB,OAAOQ;IAClC;IAEA,IAAIf,IAAIsD,QAAQ,KAAK/B,WAAW;QAC9BE,sBAAsBzB,IAAIsD,QAAQ,EAAEvC;IACtC;IACA,IAAIf,IAAI8B,KAAK,KAAKP,WAAW;QAC3BM,mBAAmB7B,IAAI8B,KAAK,EAAEf;IAChC;IACA,IAAIf,IAAIuD,OAAO,KAAKhC,WAAW;QAC7BL,qBAAqBlB,IAAIuD,OAAO,EAAExC;IACpC;IAEA,OAAOf;AACT"}
|
|
@@ -0,0 +1,5 @@
|
|
|
1
|
+
// Rows -> SQLite: core schema, reconcile loop, has(). Parsing lives in scan.ts;
|
|
2
|
+
// everything beyond frontmatter + content lives in src/features/.
|
|
3
|
+
export { clearCache, DB_FILENAME, docCount, open, SCHEMA_VERSION } from './open.js';
|
|
4
|
+
export { reconcile } from './reconcile.js';
|
|
5
|
+
export { getMeta, setMeta } from './shared.js';
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/db/index.ts"],"sourcesContent":["// Rows -> SQLite: core schema, reconcile loop, has(). Parsing lives in scan.ts;\n// everything beyond frontmatter + content lives in src/features/.\n\nexport type { OpenResult } from './open.ts';\nexport { clearCache, DB_FILENAME, docCount, open, SCHEMA_VERSION } from './open.ts';\nexport { reconcile } from './reconcile.ts';\nexport { getMeta, setMeta } from './shared.ts';\n"],"names":["clearCache","DB_FILENAME","docCount","open","SCHEMA_VERSION","reconcile","getMeta","setMeta"],"mappings":"AAAA,gFAAgF;AAChF,kEAAkE;AAGlE,SAASA,UAAU,EAAEC,WAAW,EAAEC,QAAQ,EAAEC,IAAI,EAAEC,cAAc,QAAQ,YAAY;AACpF,SAASC,SAAS,QAAQ,iBAAiB;AAC3C,SAASC,OAAO,EAAEC,OAAO,QAAQ,cAAc"}
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { DatabaseSync } from 'node:sqlite';
|
|
2
|
-
import type {
|
|
2
|
+
import type { ResolvedConfig } from '../config/index.js';
|
|
3
3
|
export declare const DB_FILENAME = "cache.db";
|
|
4
|
-
export declare const SCHEMA_VERSION = "
|
|
4
|
+
export declare const SCHEMA_VERSION = "13";
|
|
5
5
|
export interface OpenResult {
|
|
6
6
|
db: DatabaseSync;
|
|
7
7
|
cfg: ResolvedConfig;
|
|
@@ -9,12 +9,6 @@ export interface OpenResult {
|
|
|
9
9
|
parsed: number;
|
|
10
10
|
warnings: string[];
|
|
11
11
|
}
|
|
12
|
-
export declare function getMeta(db: DatabaseSync, key: string): string | null;
|
|
13
|
-
export declare function setMeta(db: DatabaseSync, key: string, value: string | null): void;
|
|
14
12
|
export declare function docCount(db: DatabaseSync): number;
|
|
15
|
-
export declare function reconcile(db: DatabaseSync, cfg: Config, baseDir: string): {
|
|
16
|
-
parsed: number;
|
|
17
|
-
warnings: string[];
|
|
18
|
-
};
|
|
19
13
|
export declare function open(cfg: ResolvedConfig): OpenResult;
|
|
20
14
|
export declare function clearCache(cfg: ResolvedConfig): void;
|
|
@@ -0,0 +1,160 @@
|
|
|
1
|
+
// The Node floor (>=22.20) is explained here and nowhere else: 22.20 is the first release with
|
|
2
|
+
// both FTS5 and row-returning INSERT ... RETURNING. Raise it only for a load-bearing capability.
|
|
3
|
+
import { mkdirSync, rmSync } from 'node:fs';
|
|
4
|
+
import { join } from 'node:path';
|
|
5
|
+
import { DatabaseSync } from 'node:sqlite';
|
|
6
|
+
import { contentTokenize, featureSignature, STATE_DIR } from '../config/index.js';
|
|
7
|
+
import { SenseError } from '../errors.js';
|
|
8
|
+
import { activeFeatures } from '../features/index.js';
|
|
9
|
+
import { registerFunctions } from '../sql-functions.js';
|
|
10
|
+
import { changedSignatureKeys, rebuildContentTable, reconcile, signatureDiff } from './reconcile.js';
|
|
11
|
+
import { getMeta, setMeta } from './shared.js';
|
|
12
|
+
export const DB_FILENAME = 'cache.db';
|
|
13
|
+
// Cache shape version, independent of the config's own `version`. Bumping it rebuilds
|
|
14
|
+
// existing trees on first query.
|
|
15
|
+
export const SCHEMA_VERSION = '13';
|
|
16
|
+
// Stemming is English-only, but the segmentation underneath it is what decides coverage:
|
|
17
|
+
// unicode61 splits on spaces, so a language written without them (Chinese, Japanese, Thai)
|
|
18
|
+
// indexes a whole run as one token and word search finds nothing. `content.tokenize` is how
|
|
19
|
+
// such a tree picks trigram instead.
|
|
20
|
+
const DEFAULT_TOKENIZE = 'porter unicode61';
|
|
21
|
+
// FTS5 takes its tokenizer as a string literal inside DDL, where nothing can bind, so a
|
|
22
|
+
// configured value has to be concatenated. Probing a throwaway table is what makes that safe
|
|
23
|
+
// and is also the whole validation: anything the linked SQLite accepts passes, anything else
|
|
24
|
+
// fails here with SQLite's own message rather than against the real table. It means no table
|
|
25
|
+
// of which version added which tokenizer has to be maintained.
|
|
26
|
+
function resolveTokenize(db, cfg) {
|
|
27
|
+
const configured = contentTokenize(cfg);
|
|
28
|
+
if (configured === undefined) return DEFAULT_TOKENIZE;
|
|
29
|
+
const literal = configured.replace(/'/g, "''");
|
|
30
|
+
try {
|
|
31
|
+
db.exec('DROP TABLE IF EXISTS temp.sense_tokenize_probe');
|
|
32
|
+
db.exec(`CREATE VIRTUAL TABLE temp.sense_tokenize_probe USING fts5(x, tokenize = '${literal}')`);
|
|
33
|
+
db.exec('DROP TABLE IF EXISTS temp.sense_tokenize_probe');
|
|
34
|
+
} catch (err) {
|
|
35
|
+
throw new SenseError('CONFIG_INVALID', `content.tokenize "${configured}" is not a tokenizer this SQLite accepts (${err.message}); the built-in choices are unicode61, ascii, porter, and trigram, each with their own options`);
|
|
36
|
+
}
|
|
37
|
+
return literal;
|
|
38
|
+
}
|
|
39
|
+
// The tokenizer the content table was actually built with, from its own DDL -- the one
|
|
40
|
+
// record that cannot desynchronize from the table. NULL when the table does not exist yet.
|
|
41
|
+
function storedTokenize(db) {
|
|
42
|
+
const row = db.prepare(`SELECT sql FROM sqlite_master WHERE name = 'content'`).get();
|
|
43
|
+
if (!row) return null;
|
|
44
|
+
const m = row.sql.match(/tokenize = '((?:[^']|'')*)'/);
|
|
45
|
+
return m ? m[1] : null;
|
|
46
|
+
}
|
|
47
|
+
// Content is a separate table (not a column on frontmatter) so `SELECT * FROM frontmatter`
|
|
48
|
+
// can't dump file text into context. Features add their own tables after the core ones.
|
|
49
|
+
function ensureSchema(db, cfg, tokenize) {
|
|
50
|
+
db.exec(`CREATE TABLE IF NOT EXISTS frontmatter ("path" TEXT PRIMARY KEY, "_mtime" REAL, "_ctime" REAL, "_size" INTEGER, "_parse_error" TEXT)`);
|
|
51
|
+
// IF NOT EXISTS is safe against a tokenizer change: open() compares the table's own DDL
|
|
52
|
+
// against the resolved tokenizer before this runs, so a stale table is already gone by now.
|
|
53
|
+
// The three `_seg` sidecars are appended after path, never inserted: bm25(content, ...) and
|
|
54
|
+
// snippet(content, 2, ...) are documented against the first three columns and keep working
|
|
55
|
+
// (FTS5 defaults the weights it was not given). Each carries its field's exploded unspaced
|
|
56
|
+
// runs for text that needs it, and an empty string for text that does not.
|
|
57
|
+
db.exec(`CREATE VIRTUAL TABLE IF NOT EXISTS content USING fts5(title, summary, text, path UNINDEXED, title_seg, summary_seg, text_seg, tokenize = '${tokenize}')`);
|
|
58
|
+
// Coverage, not ownership: a path can appear under several presets. path leads the PK so the
|
|
59
|
+
// per-doc delete is an index hit -- keyed the other way, cold builds went quadratic.
|
|
60
|
+
db.exec(`CREATE TABLE IF NOT EXISTS preset_files ("path" TEXT, preset TEXT, PRIMARY KEY ("path", preset))`);
|
|
61
|
+
db.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');
|
|
62
|
+
for (const feature of activeFeatures(cfg))feature.schema(db);
|
|
63
|
+
if (getMeta(db, 'schema_version') === null) setMeta(db, 'schema_version', SCHEMA_VERSION);
|
|
64
|
+
if (getMeta(db, 'features') === null) setMeta(db, 'features', featureSignature(cfg));
|
|
65
|
+
}
|
|
66
|
+
export function docCount(db) {
|
|
67
|
+
const row = db.prepare('SELECT COUNT(*) AS n FROM frontmatter').get();
|
|
68
|
+
return row.n;
|
|
69
|
+
}
|
|
70
|
+
export function open(cfg) {
|
|
71
|
+
var _getMeta;
|
|
72
|
+
const stateDir = join(cfg.baseDir, STATE_DIR);
|
|
73
|
+
mkdirSync(stateDir, {
|
|
74
|
+
recursive: true
|
|
75
|
+
});
|
|
76
|
+
const dbPath = join(stateDir, DB_FILENAME);
|
|
77
|
+
const db = new DatabaseSync(dbPath);
|
|
78
|
+
db.exec('PRAGMA journal_mode = WAL');
|
|
79
|
+
// Covers a concurrent watcher's bulk reconcile (~5s for 500 files at 26k notes). A query
|
|
80
|
+
// that outwaits it still fails loudly.
|
|
81
|
+
db.exec('PRAGMA busy_timeout = 30000');
|
|
82
|
+
registerFunctions(db, contentTokenize(cfg) === undefined);
|
|
83
|
+
db.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');
|
|
84
|
+
// Before the rebuild branch below, never after: that branch deletes the cache, so a typo'd
|
|
85
|
+
// tokenizer validated later would cost a full re-index (and re-embed) to reach its own error.
|
|
86
|
+
const tokenize = resolveTokenize(db, cfg);
|
|
87
|
+
// Schema-version or feature-set mismatch: reconcile only reparses changed files, so an
|
|
88
|
+
// old cache can't be patched incrementally -- rebuild instead (cheap: nothing expensive lives here).
|
|
89
|
+
const version = getMeta(db, 'schema_version');
|
|
90
|
+
const features = getMeta(db, 'features');
|
|
91
|
+
const wantFeatures = featureSignature(cfg);
|
|
92
|
+
let tokenizeOnlyRebuild = false;
|
|
93
|
+
if (version !== null && version !== SCHEMA_VERSION || features !== null && features !== wantFeatures) {
|
|
94
|
+
// Indexing derives from presets, so a config edit rebuilding the cache must say so and
|
|
95
|
+
// name what changed -- silent rebuilds make derived indexing look like a hang or a bug.
|
|
96
|
+
if (version !== null && version !== SCHEMA_VERSION) {
|
|
97
|
+
console.error('sense: cache format changed (new sensemaking version); rebuilding the index');
|
|
98
|
+
db.close();
|
|
99
|
+
clearCache(cfg);
|
|
100
|
+
return open(cfg);
|
|
101
|
+
}
|
|
102
|
+
const changedKeys = changedSignatureKeys(features !== null && features !== void 0 ? features : '', wantFeatures);
|
|
103
|
+
// Only the tokenizer moved: frontmatter, links, sections, and embeddings are file-derived
|
|
104
|
+
// and tokenizer-independent, so they don't need re-deriving. Everything else -- a preset
|
|
105
|
+
// edit, an embed model change -- still takes the full clear/reopen below.
|
|
106
|
+
if (changedKeys.size === 1 && changedKeys.has('tokenize')) {
|
|
107
|
+
console.error('sense: config change (content tokenizer) rebuilds the text index; vectors, links, and sections are kept');
|
|
108
|
+
db.exec('DROP TABLE content');
|
|
109
|
+
tokenizeOnlyRebuild = true;
|
|
110
|
+
} else {
|
|
111
|
+
const changed = signatureDiff(features !== null && features !== void 0 ? features : '', wantFeatures);
|
|
112
|
+
console.error(`sense: config change (${changed}) rebuilds the index`);
|
|
113
|
+
db.close();
|
|
114
|
+
clearCache(cfg);
|
|
115
|
+
return open(cfg);
|
|
116
|
+
}
|
|
117
|
+
}
|
|
118
|
+
// Meta can lie after a crash between table creation and the signature write; the table's
|
|
119
|
+
// own DDL cannot. A mismatch here rebuilds no matter what meta says. A tokenize-only rebuild
|
|
120
|
+
// just dropped content above, so storedTokenize sees no table and this guard is skipped
|
|
121
|
+
// naturally rather than needing its own case.
|
|
122
|
+
const stored = storedTokenize(db);
|
|
123
|
+
if (stored !== null && stored !== tokenize) {
|
|
124
|
+
console.error('sense: cache was built with a different content tokenizer; rebuilding the index');
|
|
125
|
+
db.close();
|
|
126
|
+
clearCache(cfg);
|
|
127
|
+
return open(cfg);
|
|
128
|
+
}
|
|
129
|
+
ensureSchema(db, cfg, tokenize);
|
|
130
|
+
let rebuildWarnings = [];
|
|
131
|
+
if (tokenizeOnlyRebuild) {
|
|
132
|
+
rebuildWarnings = rebuildContentTable(db, cfg, cfg.baseDir);
|
|
133
|
+
setMeta(db, 'features', wantFeatures);
|
|
134
|
+
}
|
|
135
|
+
// 3x the largest reconcile this cache has recorded, floored at 30s and capped at 10min.
|
|
136
|
+
// Installed before reconcile() -- that call is the one that races a watcher's transaction.
|
|
137
|
+
const recordedMaxMs = Number((_getMeta = getMeta(db, 'reconcile_max_ms')) !== null && _getMeta !== void 0 ? _getMeta : '0');
|
|
138
|
+
db.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600000)}`);
|
|
139
|
+
const { parsed, warnings } = reconcile(db, cfg, cfg.baseDir);
|
|
140
|
+
return {
|
|
141
|
+
db,
|
|
142
|
+
cfg,
|
|
143
|
+
dbPath,
|
|
144
|
+
parsed,
|
|
145
|
+
warnings: [
|
|
146
|
+
...rebuildWarnings,
|
|
147
|
+
...warnings
|
|
148
|
+
]
|
|
149
|
+
};
|
|
150
|
+
}
|
|
151
|
+
// Deletes the cache directory, and only that. The rebuild is not this function's job: the next
|
|
152
|
+
// open() reconciles, which is what running any command already does, so a verb that bundled the
|
|
153
|
+
// two described the half that was not its own. Manual reset for a doubted cache; the schema and
|
|
154
|
+
// config-signature mismatches below reset themselves.
|
|
155
|
+
export function clearCache(cfg) {
|
|
156
|
+
rmSync(join(cfg.baseDir, STATE_DIR), {
|
|
157
|
+
recursive: true,
|
|
158
|
+
force: true
|
|
159
|
+
});
|
|
160
|
+
}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/db/open.ts"],"sourcesContent":["// The Node floor (>=22.20) is explained here and nowhere else: 22.20 is the first release with\n// both FTS5 and row-returning INSERT ... RETURNING. Raise it only for a load-bearing capability.\nimport { mkdirSync, rmSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { DatabaseSync } from 'node:sqlite';\nimport type { Config, ResolvedConfig } from '../config/index.ts';\nimport { contentTokenize, featureSignature, STATE_DIR } from '../config/index.ts';\nimport { SenseError } from '../errors.ts';\nimport { activeFeatures } from '../features/index.ts';\nimport { registerFunctions } from '../sql-functions.ts';\nimport { changedSignatureKeys, rebuildContentTable, reconcile, signatureDiff } from './reconcile.ts';\nimport { getMeta, setMeta } from './shared.ts';\n\nexport const DB_FILENAME = 'cache.db';\n// Cache shape version, independent of the config's own `version`. Bumping it rebuilds\n// existing trees on first query.\nexport const SCHEMA_VERSION = '13';\n\nexport interface OpenResult {\n db: DatabaseSync;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\n// Stemming is English-only, but the segmentation underneath it is what decides coverage:\n// unicode61 splits on spaces, so a language written without them (Chinese, Japanese, Thai)\n// indexes a whole run as one token and word search finds nothing. `content.tokenize` is how\n// such a tree picks trigram instead.\nconst DEFAULT_TOKENIZE = 'porter unicode61';\n\n// FTS5 takes its tokenizer as a string literal inside DDL, where nothing can bind, so a\n// configured value has to be concatenated. Probing a throwaway table is what makes that safe\n// and is also the whole validation: anything the linked SQLite accepts passes, anything else\n// fails here with SQLite's own message rather than against the real table. It means no table\n// of which version added which tokenizer has to be maintained.\nfunction resolveTokenize(db: DatabaseSync, cfg: Config): string {\n const configured = contentTokenize(cfg);\n if (configured === undefined) return DEFAULT_TOKENIZE;\n const literal = configured.replace(/'/g, \"''\");\n try {\n db.exec('DROP TABLE IF EXISTS temp.sense_tokenize_probe');\n db.exec(`CREATE VIRTUAL TABLE temp.sense_tokenize_probe USING fts5(x, tokenize = '${literal}')`);\n db.exec('DROP TABLE IF EXISTS temp.sense_tokenize_probe');\n } catch (err) {\n throw new SenseError('CONFIG_INVALID', `content.tokenize \"${configured}\" is not a tokenizer this SQLite accepts (${(err as Error).message}); the built-in choices are unicode61, ascii, porter, and trigram, each with their own options`);\n }\n return literal;\n}\n\n// The tokenizer the content table was actually built with, from its own DDL -- the one\n// record that cannot desynchronize from the table. NULL when the table does not exist yet.\nfunction storedTokenize(db: DatabaseSync): string | null {\n const row = db.prepare(`SELECT sql FROM sqlite_master WHERE name = 'content'`).get() as { sql: string } | undefined;\n if (!row) return null;\n const m = row.sql.match(/tokenize = '((?:[^']|'')*)'/);\n return m ? m[1] : null;\n}\n\n// Content is a separate table (not a column on frontmatter) so `SELECT * FROM frontmatter`\n// can't dump file text into context. Features add their own tables after the core ones.\nfunction ensureSchema(db: DatabaseSync, cfg: Config, tokenize: string): void {\n db.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" REAL, \"_ctime\" REAL, \"_size\" INTEGER, \"_parse_error\" TEXT)`);\n // IF NOT EXISTS is safe against a tokenizer change: open() compares the table's own DDL\n // against the resolved tokenizer before this runs, so a stale table is already gone by now.\n // The three `_seg` sidecars are appended after path, never inserted: bm25(content, ...) and\n // snippet(content, 2, ...) are documented against the first three columns and keep working\n // (FTS5 defaults the weights it was not given). Each carries its field's exploded unspaced\n // runs for text that needs it, and an empty string for text that does not.\n db.exec(`CREATE VIRTUAL TABLE IF NOT EXISTS content USING fts5(title, summary, text, path UNINDEXED, title_seg, summary_seg, text_seg, tokenize = '${tokenize}')`);\n // Coverage, not ownership: a path can appear under several presets. path leads the PK so the\n // per-doc delete is an index hit -- keyed the other way, cold builds went quadratic.\n db.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n db.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n for (const feature of activeFeatures(cfg)) feature.schema(db);\n if (getMeta(db, 'schema_version') === null) setMeta(db, 'schema_version', SCHEMA_VERSION);\n if (getMeta(db, 'features') === null) setMeta(db, 'features', featureSignature(cfg));\n}\n\nexport function docCount(db: DatabaseSync): number {\n const row = db.prepare('SELECT COUNT(*) AS n FROM frontmatter').get() as { n: number };\n return row.n;\n}\n\nexport function open(cfg: ResolvedConfig): OpenResult {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, DB_FILENAME);\n\n const db = new DatabaseSync(dbPath);\n db.exec('PRAGMA journal_mode = WAL');\n // Covers a concurrent watcher's bulk reconcile (~5s for 500 files at 26k notes). A query\n // that outwaits it still fails loudly.\n db.exec('PRAGMA busy_timeout = 30000');\n registerFunctions(db, contentTokenize(cfg) === undefined);\n\n db.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');\n\n // Before the rebuild branch below, never after: that branch deletes the cache, so a typo'd\n // tokenizer validated later would cost a full re-index (and re-embed) to reach its own error.\n const tokenize = resolveTokenize(db, cfg);\n\n // Schema-version or feature-set mismatch: reconcile only reparses changed files, so an\n // old cache can't be patched incrementally -- rebuild instead (cheap: nothing expensive lives here).\n const version = getMeta(db, 'schema_version');\n const features = getMeta(db, 'features');\n const wantFeatures = featureSignature(cfg);\n let tokenizeOnlyRebuild = false;\n if ((version !== null && version !== SCHEMA_VERSION) || (features !== null && features !== wantFeatures)) {\n // Indexing derives from presets, so a config edit rebuilding the cache must say so and\n // name what changed -- silent rebuilds make derived indexing look like a hang or a bug.\n if (version !== null && version !== SCHEMA_VERSION) {\n console.error('sense: cache format changed (new sensemaking version); rebuilding the index');\n db.close();\n clearCache(cfg);\n return open(cfg);\n }\n const changedKeys = changedSignatureKeys(features ?? '', wantFeatures);\n // Only the tokenizer moved: frontmatter, links, sections, and embeddings are file-derived\n // and tokenizer-independent, so they don't need re-deriving. Everything else -- a preset\n // edit, an embed model change -- still takes the full clear/reopen below.\n if (changedKeys.size === 1 && changedKeys.has('tokenize')) {\n console.error('sense: config change (content tokenizer) rebuilds the text index; vectors, links, and sections are kept');\n db.exec('DROP TABLE content');\n tokenizeOnlyRebuild = true;\n } else {\n const changed = signatureDiff(features ?? '', wantFeatures);\n console.error(`sense: config change (${changed}) rebuilds the index`);\n db.close();\n clearCache(cfg);\n return open(cfg);\n }\n }\n\n // Meta can lie after a crash between table creation and the signature write; the table's\n // own DDL cannot. A mismatch here rebuilds no matter what meta says. A tokenize-only rebuild\n // just dropped content above, so storedTokenize sees no table and this guard is skipped\n // naturally rather than needing its own case.\n const stored = storedTokenize(db);\n if (stored !== null && stored !== tokenize) {\n console.error('sense: cache was built with a different content tokenizer; rebuilding the index');\n db.close();\n clearCache(cfg);\n return open(cfg);\n }\n\n ensureSchema(db, cfg, tokenize);\n\n let rebuildWarnings: string[] = [];\n if (tokenizeOnlyRebuild) {\n rebuildWarnings = rebuildContentTable(db, cfg, cfg.baseDir);\n setMeta(db, 'features', wantFeatures);\n }\n\n // 3x the largest reconcile this cache has recorded, floored at 30s and capped at 10min.\n // Installed before reconcile() -- that call is the one that races a watcher's transaction.\n const recordedMaxMs = Number(getMeta(db, 'reconcile_max_ms') ?? '0');\n db.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600_000)}`);\n\n const { parsed, warnings } = reconcile(db, cfg, cfg.baseDir);\n\n return { db, cfg, dbPath, parsed, warnings: [...rebuildWarnings, ...warnings] };\n}\n\n// Deletes the cache directory, and only that. The rebuild is not this function's job: the next\n// open() reconciles, which is what running any command already does, so a verb that bundled the\n// two described the half that was not its own. Manual reset for a doubted cache; the schema and\n// config-signature mismatches below reset themselves.\nexport function clearCache(cfg: ResolvedConfig): void {\n rmSync(join(cfg.baseDir, STATE_DIR), { recursive: true, force: true });\n}\n"],"names":["mkdirSync","rmSync","join","DatabaseSync","contentTokenize","featureSignature","STATE_DIR","SenseError","activeFeatures","registerFunctions","changedSignatureKeys","rebuildContentTable","reconcile","signatureDiff","getMeta","setMeta","DB_FILENAME","SCHEMA_VERSION","DEFAULT_TOKENIZE","resolveTokenize","db","cfg","configured","undefined","literal","replace","exec","err","message","storedTokenize","row","prepare","get","m","sql","match","ensureSchema","tokenize","feature","schema","docCount","n","open","stateDir","baseDir","recursive","dbPath","version","features","wantFeatures","tokenizeOnlyRebuild","console","error","close","clearCache","changedKeys","size","has","changed","stored","rebuildWarnings","recordedMaxMs","Number","Math","min","max","parsed","warnings","force"],"mappings":"AAAA,+FAA+F;AAC/F,iGAAiG;AACjG,SAASA,SAAS,EAAEC,MAAM,QAAQ,UAAU;AAC5C,SAASC,IAAI,QAAQ,YAAY;AACjC,SAASC,YAAY,QAAQ,cAAc;AAE3C,SAASC,eAAe,EAAEC,gBAAgB,EAAEC,SAAS,QAAQ,qBAAqB;AAClF,SAASC,UAAU,QAAQ,eAAe;AAC1C,SAASC,cAAc,QAAQ,uBAAuB;AACtD,SAASC,iBAAiB,QAAQ,sBAAsB;AACxD,SAASC,oBAAoB,EAAEC,mBAAmB,EAAEC,SAAS,EAAEC,aAAa,QAAQ,iBAAiB;AACrG,SAASC,OAAO,EAAEC,OAAO,QAAQ,cAAc;AAE/C,OAAO,MAAMC,cAAc,WAAW;AACtC,sFAAsF;AACtF,iCAAiC;AACjC,OAAO,MAAMC,iBAAiB,KAAK;AAUnC,yFAAyF;AACzF,2FAA2F;AAC3F,4FAA4F;AAC5F,qCAAqC;AACrC,MAAMC,mBAAmB;AAEzB,wFAAwF;AACxF,6FAA6F;AAC7F,6FAA6F;AAC7F,6FAA6F;AAC7F,+DAA+D;AAC/D,SAASC,gBAAgBC,EAAgB,EAAEC,GAAW;IACpD,MAAMC,aAAalB,gBAAgBiB;IACnC,IAAIC,eAAeC,WAAW,OAAOL;IACrC,MAAMM,UAAUF,WAAWG,OAAO,CAAC,MAAM;IACzC,IAAI;QACFL,GAAGM,IAAI,CAAC;QACRN,GAAGM,IAAI,CAAC,CAAC,yEAAyE,EAAEF,QAAQ,EAAE,CAAC;QAC/FJ,GAAGM,IAAI,CAAC;IACV,EAAE,OAAOC,KAAK;QACZ,MAAM,IAAIpB,WAAW,kBAAkB,CAAC,kBAAkB,EAAEe,WAAW,0CAA0C,EAAE,AAACK,IAAcC,OAAO,CAAC,8FAA8F,CAAC;IAC3O;IACA,OAAOJ;AACT;AAEA,uFAAuF;AACvF,2FAA2F;AAC3F,SAASK,eAAeT,EAAgB;IACtC,MAAMU,MAAMV,GAAGW,OAAO,CAAC,CAAC,oDAAoD,CAAC,EAAEC,GAAG;IAClF,IAAI,CAACF,KAAK,OAAO;IACjB,MAAMG,IAAIH,IAAII,GAAG,CAACC,KAAK,CAAC;IACxB,OAAOF,IAAIA,CAAC,CAAC,EAAE,GAAG;AACpB;AAEA,2FAA2F;AAC3F,wFAAwF;AACxF,SAASG,aAAahB,EAAgB,EAAEC,GAAW,EAAEgB,QAAgB;IACnEjB,GAAGM,IAAI,CAAC,CAAC,oIAAoI,CAAC;IAC9I,wFAAwF;IACxF,4FAA4F;IAC5F,4FAA4F;IAC5F,2FAA2F;IAC3F,2FAA2F;IAC3F,2EAA2E;IAC3EN,GAAGM,IAAI,CAAC,CAAC,0IAA0I,EAAEW,SAAS,EAAE,CAAC;IACjK,6FAA6F;IAC7F,qFAAqF;IACrFjB,GAAGM,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAC1GN,GAAGM,IAAI,CAAC;IACR,KAAK,MAAMY,WAAW9B,eAAea,KAAMiB,QAAQC,MAAM,CAACnB;IAC1D,IAAIN,QAAQM,IAAI,sBAAsB,MAAML,QAAQK,IAAI,kBAAkBH;IAC1E,IAAIH,QAAQM,IAAI,gBAAgB,MAAML,QAAQK,IAAI,YAAYf,iBAAiBgB;AACjF;AAEA,OAAO,SAASmB,SAASpB,EAAgB;IACvC,MAAMU,MAAMV,GAAGW,OAAO,CAAC,yCAAyCC,GAAG;IACnE,OAAOF,IAAIW,CAAC;AACd;AAEA,OAAO,SAASC,KAAKrB,GAAmB;QAwETP;IAvE7B,MAAM6B,WAAWzC,KAAKmB,IAAIuB,OAAO,EAAEtC;IACnCN,UAAU2C,UAAU;QAAEE,WAAW;IAAK;IACtC,MAAMC,SAAS5C,KAAKyC,UAAU3B;IAE9B,MAAMI,KAAK,IAAIjB,aAAa2C;IAC5B1B,GAAGM,IAAI,CAAC;IACR,yFAAyF;IACzF,uCAAuC;IACvCN,GAAGM,IAAI,CAAC;IACRjB,kBAAkBW,IAAIhB,gBAAgBiB,SAASE;IAE/CH,GAAGM,IAAI,CAAC;IAER,2FAA2F;IAC3F,8FAA8F;IAC9F,MAAMW,WAAWlB,gBAAgBC,IAAIC;IAErC,uFAAuF;IACvF,qGAAqG;IACrG,MAAM0B,UAAUjC,QAAQM,IAAI;IAC5B,MAAM4B,WAAWlC,QAAQM,IAAI;IAC7B,MAAM6B,eAAe5C,iBAAiBgB;IACtC,IAAI6B,sBAAsB;IAC1B,IAAI,AAACH,YAAY,QAAQA,YAAY9B,kBAAoB+B,aAAa,QAAQA,aAAaC,cAAe;QACxG,uFAAuF;QACvF,wFAAwF;QACxF,IAAIF,YAAY,QAAQA,YAAY9B,gBAAgB;YAClDkC,QAAQC,KAAK,CAAC;YACdhC,GAAGiC,KAAK;YACRC,WAAWjC;YACX,OAAOqB,KAAKrB;QACd;QACA,MAAMkC,cAAc7C,qBAAqBsC,qBAAAA,sBAAAA,WAAY,IAAIC;QACzD,0FAA0F;QAC1F,yFAAyF;QACzF,0EAA0E;QAC1E,IAAIM,YAAYC,IAAI,KAAK,KAAKD,YAAYE,GAAG,CAAC,aAAa;YACzDN,QAAQC,KAAK,CAAC;YACdhC,GAAGM,IAAI,CAAC;YACRwB,sBAAsB;QACxB,OAAO;YACL,MAAMQ,UAAU7C,cAAcmC,qBAAAA,sBAAAA,WAAY,IAAIC;YAC9CE,QAAQC,KAAK,CAAC,CAAC,sBAAsB,EAAEM,QAAQ,oBAAoB,CAAC;YACpEtC,GAAGiC,KAAK;YACRC,WAAWjC;YACX,OAAOqB,KAAKrB;QACd;IACF;IAEA,yFAAyF;IACzF,6FAA6F;IAC7F,wFAAwF;IACxF,8CAA8C;IAC9C,MAAMsC,SAAS9B,eAAeT;IAC9B,IAAIuC,WAAW,QAAQA,WAAWtB,UAAU;QAC1Cc,QAAQC,KAAK,CAAC;QACdhC,GAAGiC,KAAK;QACRC,WAAWjC;QACX,OAAOqB,KAAKrB;IACd;IAEAe,aAAahB,IAAIC,KAAKgB;IAEtB,IAAIuB,kBAA4B,EAAE;IAClC,IAAIV,qBAAqB;QACvBU,kBAAkBjD,oBAAoBS,IAAIC,KAAKA,IAAIuB,OAAO;QAC1D7B,QAAQK,IAAI,YAAY6B;IAC1B;IAEA,wFAAwF;IACxF,2FAA2F;IAC3F,MAAMY,gBAAgBC,QAAOhD,WAAAA,QAAQM,IAAI,iCAAZN,sBAAAA,WAAmC;IAChEM,GAAGM,IAAI,CAAC,CAAC,sBAAsB,EAAEqC,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,OAAO,IAAIJ,gBAAgB,SAAU;IAExF,MAAM,EAAEK,MAAM,EAAEC,QAAQ,EAAE,GAAGvD,UAAUQ,IAAIC,KAAKA,IAAIuB,OAAO;IAE3D,OAAO;QAAExB;QAAIC;QAAKyB;QAAQoB;QAAQC,UAAU;eAAIP;eAAoBO;SAAS;IAAC;AAChF;AAEA,+FAA+F;AAC/F,gGAAgG;AAChG,gGAAgG;AAChG,sDAAsD;AACtD,OAAO,SAASb,WAAWjC,GAAmB;IAC5CpB,OAAOC,KAAKmB,IAAIuB,OAAO,EAAEtC,YAAY;QAAEuC,WAAW;QAAMuB,OAAO;IAAK;AACtE"}
|
|
@@ -0,0 +1,9 @@
|
|
|
1
|
+
import type { DatabaseSync } from 'node:sqlite';
|
|
2
|
+
import type { Config } from '../config/index.js';
|
|
3
|
+
export declare function reconcile(db: DatabaseSync, cfg: Config, baseDir: string): {
|
|
4
|
+
parsed: number;
|
|
5
|
+
warnings: string[];
|
|
6
|
+
};
|
|
7
|
+
export declare function changedSignatureKeys(before: string, after: string): Set<string>;
|
|
8
|
+
export declare function signatureDiff(before: string, after: string): string;
|
|
9
|
+
export declare function rebuildContentTable(db: DatabaseSync, cfg: Config, baseDir: string): string[];
|