Kristian WindandClaude Opus 5
40e2cad346
Initial data model for verdande: users, sessions, invites, projects, sections, tasks, labels, reminders, comments, attachments, activity and filters, plus personal API tokens. Search indexes a generated `fold` column alongside the original text. FTS5's remove_diacritics does not touch ø and æ — Unicode treats them as letters rather than accented forms — so without it a Danish user cannot find "grøn" by typing "gron". MatchExpr quotes every term so free text containing FTS5 operators cannot reach the query planner as syntax. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
80 lines
2.5 KiB
Go
80 lines
2.5 KiB
Go
package store
|
|
|
|
import (
|
|
"strings"
|
|
"unicode"
|
|
)
|
|
|
|
// FoldDanish transliterates the Danish letters that Unicode does not treat as
|
|
// accented forms of anything, so they survive into a form FTS5 can fold further.
|
|
//
|
|
// This has to be applied to the query with exactly the same mapping the `fold`
|
|
// column on `tasks` uses. The two are a pair: change one and searching quietly
|
|
// stops finding half the words in the user's own language.
|
|
func FoldDanish(s string) string {
|
|
r := strings.NewReplacer(
|
|
"ø", "o", "Ø", "o",
|
|
"æ", "ae", "Æ", "ae",
|
|
"å", "aa", "Å", "aa",
|
|
)
|
|
return r.Replace(s)
|
|
}
|
|
|
|
// MatchExpr turns what someone typed into the search box into an FTS5 MATCH
|
|
// expression.
|
|
//
|
|
// User input is never passed to FTS5 as-is. The MATCH grammar has its own operators
|
|
// (`AND`, `NEAR`, `*`, `^`, `:`, `-`), so a perfectly ordinary task title like
|
|
// "budget: q3 - final" parses as a query and fails with a syntax error rather than
|
|
// finding anything. Every term is quoted, which makes it a literal.
|
|
//
|
|
// Each term must match somewhere in the row, and may match either the text as
|
|
// written or its folded form — so "gron" finds "grøn" and "grøn" still finds
|
|
// itself. Terms are prefix-matched, which is what makes the Cmd+K box feel like it
|
|
// is searching as you type.
|
|
//
|
|
// Returns "" when there is nothing searchable in the input; callers should treat
|
|
// that as "no query" rather than as "match everything".
|
|
func MatchExpr(query string) string {
|
|
terms := tokenize(query)
|
|
if len(terms) == 0 {
|
|
return ""
|
|
}
|
|
|
|
var b strings.Builder
|
|
for i, term := range terms {
|
|
if i > 0 {
|
|
b.WriteString(" AND ")
|
|
}
|
|
folded := FoldDanish(term)
|
|
b.WriteString(`({content description} : `)
|
|
b.WriteString(quote(term))
|
|
b.WriteString(`* OR fold : `)
|
|
b.WriteString(quote(folded))
|
|
b.WriteString(`*)`)
|
|
}
|
|
return b.String()
|
|
}
|
|
|
|
// tokenize splits on everything that is not a letter or a digit. Splitting on
|
|
// punctuation rather than whitespace alone means "q3/2026" searches for "q3" and
|
|
// "2026" instead of for a term FTS5's own tokenizer would never have indexed.
|
|
func tokenize(query string) []string {
|
|
fields := strings.FieldsFunc(query, func(r rune) bool {
|
|
return !unicode.IsLetter(r) && !unicode.IsDigit(r)
|
|
})
|
|
terms := make([]string, 0, len(fields))
|
|
for _, f := range fields {
|
|
if f != "" {
|
|
terms = append(terms, f)
|
|
}
|
|
}
|
|
return terms
|
|
}
|
|
|
|
// quote wraps a term as an FTS5 string literal. Inside one, the only character
|
|
// with meaning is the double quote, escaped by doubling it.
|
|
func quote(term string) string {
|
|
return `"` + strings.ReplaceAll(term, `"`, `""`) + `"`
|
|
}
|