| 1 |
|
package search |
| 2 |
|
|
| 3 |
|
import ( |
| 4 |
|
"fmt" |
| 5 |
|
"strings" |
| 6 |
|
"sync" |
| 7 |
|
|
| 8 |
|
"sourcecraft.dev/bigbes/sr-ht-spec/doc" |
| 9 |
|
"sourcecraft.dev/bigbes/sr-ht-spec/gitx" |
| 10 |
|
) |
| 11 |
|
|
| 12 |
|
// renderer is shared: doc.Renderer is documented as reusable and |
| 13 |
|
// concurrency-safe, and building one per extraction would rebuild the whole |
| 14 |
|
// goldmark pipeline for every space. |
| 15 |
|
var renderer = sync.OnceValue(doc.NewRenderer) |
| 16 |
|
|
| 17 |
|
// Bodies keys a revision's documents by tree path, which is the shape Extract |
| 18 |
|
// wants them in. It pairs with doc.FromDocuments: the same []gitx.Document |
| 19 |
|
// builds the Archive and supplies the text. |
| 20 |
51 |
func Bodies(docs []gitx.Document) map[string][]byte { |
| 21 |
51 |
m := make(map[string][]byte, len(docs)) |
| 22 |
1161 |
for _, d := range docs { |
| 23 |
1161 |
m[d.Path] = d.Data |
| 24 |
1161 |
} |
| 25 |
51 |
return m |
| 26 |
|
} |
| 27 |
|
|
| 28 |
|
// Extract projects one space at one revision into the documents the index |
| 29 |
|
// stores. bodies holds each page's raw markdown — frontmatter included, exactly |
| 30 |
|
// gitx.Document.Data — keyed by doc.Page.Path. |
| 31 |
|
// |
| 32 |
|
// A page in the archive with no body in bodies is an error, not a page indexed |
| 33 |
|
// with an empty body. The two are indistinguishable once indexed, and the |
| 34 |
|
// second is how a document silently stops being findable. |
| 35 |
|
// |
| 36 |
|
// Three document shapes come out, matching what doc/ models: |
| 37 |
|
// |
| 38 |
|
// - an ordinary document, indexed as its frontmatter projected to "key: |
| 39 |
|
// value" lines followed by its rendered plain text. The frontmatter is in |
| 40 |
|
// there because tags, owners and summaries render as chips rather than |
| 41 |
|
// prose, and a search for one of them should still find the document. |
| 42 |
|
// - a catalog (`type: catalog`, or index.md), indexed by title and section |
| 43 |
|
// only. A catalog is a page of one-line descriptions of other documents; |
| 44 |
|
// indexed whole, a query lands on the description instead of on the |
| 45 |
|
// document that owns it. |
| 46 |
|
// - an activity log (`type: log`, or log.md), which contributes its own |
| 47 |
|
// title-only document plus one document per dated entry. A hit anywhere in |
| 48 |
|
// a log otherwise resolves to the whole log; split, each entry is the size |
| 49 |
|
// of the thing it describes and carries an anchor into it. |
| 50 |
52 |
func Extract(arc *doc.Archive, bodies map[string][]byte) ([]Document, error) { |
| 51 |
52 |
if arc == nil { |
| 52 |
1 |
return nil, fmt.Errorf("search: Extract needs an archive") |
| 53 |
1 |
} |
| 54 |
51 |
pages := arc.All() |
| 55 |
51 |
out := make([]Document, 0, len(pages)) |
| 56 |
51 |
r := renderer() |
| 57 |
51 |
|
| 58 |
1160 |
for _, p := range pages { |
| 59 |
1160 |
src, ok := bodies[p.Path] |
| 60 |
1160 |
if !ok { |
| 61 |
1 |
return nil, fmt.Errorf("search: no body supplied for %s in %s", p.Path, arc.Space) |
| 62 |
1 |
} |
| 63 |
1159 |
front, body := doc.ParseFront(src) |
| 64 |
1159 |
res := r.Render(body, doc.DirOf(p.Path), arc) |
| 65 |
1159 |
|
| 66 |
1159 |
d := Document{ |
| 67 |
1159 |
Space: arc.Space, |
| 68 |
1159 |
ID: p.ID, |
| 69 |
1159 |
Rev: arc.Rev, |
| 70 |
1159 |
Path: p.Path, |
| 71 |
1159 |
Section: p.Section, |
| 72 |
1159 |
Title: p.Title, |
| 73 |
1159 |
Text: front.SearchText() + res.PlainText, |
| 74 |
1159 |
} |
| 75 |
1159 |
switch p.Kind { |
| 76 |
1 |
case doc.KindCatalog: |
| 77 |
1 |
d.Text = "" |
| 78 |
4 |
case doc.KindLog: |
| 79 |
4 |
d.Section = doc.LogSection |
| 80 |
4 |
d.Text = "" |
| 81 |
4 |
out = append(out, d) |
| 82 |
4 |
out = append(out, logEntries(arc, p, body)...) |
| 83 |
4 |
continue |
| 84 |
|
} |
| 85 |
1155 |
out = append(out, d) |
| 86 |
|
} |
| 87 |
50 |
return out, nil |
| 88 |
|
} |
| 89 |
|
|
| 90 |
|
// logEntries splits an activity log into one indexable document per dated |
| 91 |
|
// entry. |
| 92 |
|
// |
| 93 |
|
// The entry ids doc.SplitLog produces are "log#<date>-<n>", named after |
| 94 |
|
// warren's single vault-wide log. In a space that is not unique: a second |
| 95 |
|
// document marked `type: log` — or simply a second file named log.md in another |
| 96 |
|
// directory — produces the same ids, and in one index the same ids are the same |
| 97 |
|
// documents, so one log would silently overwrite the other. The owning page's |
| 98 |
|
// id is therefore substituted for the "log" prefix, which is a no-op for a log |
| 99 |
|
// whose page id is in fact "log" and disambiguates every other case. The result |
| 100 |
|
// also resolves better: "notes/dev-log#2026-05-31-1" names the document the |
| 101 |
|
// entry is in. |
| 102 |
4 |
func logEntries(arc *doc.Archive, p *doc.Page, body []byte) []Document { |
| 103 |
4 |
entries := doc.SplitLog(body) |
| 104 |
4 |
out := make([]Document, 0, len(entries)) |
| 105 |
5 |
for _, e := range entries { |
| 106 |
5 |
suffix := strings.TrimPrefix(e.ID, "log#") |
| 107 |
5 |
out = append(out, Document{ |
| 108 |
5 |
Space: arc.Space, |
| 109 |
5 |
ID: p.ID + "#" + suffix, |
| 110 |
5 |
Rev: arc.Rev, |
| 111 |
5 |
Path: p.Path, |
| 112 |
5 |
Anchor: e.Anchor, |
| 113 |
5 |
Section: doc.LogSection, |
| 114 |
5 |
Title: e.Date + " " + e.Title, |
| 115 |
5 |
Text: e.SearchText(), |
| 116 |
5 |
}) |
| 117 |
5 |
} |
| 118 |
4 |
return out |
| 119 |
|
} |