1
0
mirror of https://github.com/pocketbase/pocketbase.git synced 2024-11-28 10:03:42 +02:00
pocketbase/daos/record_expand.go
2022-07-07 00:19:05 +03:00

156 lines
4.1 KiB
Go

package daos
import (
"errors"
"fmt"
"strings"
"github.com/pocketbase/pocketbase/models"
"github.com/pocketbase/pocketbase/models/schema"
"github.com/pocketbase/pocketbase/tools/list"
)
// MaxExpandDepth specifies the max allowed nested expand depth path.
const MaxExpandDepth = 6
// ExpandFetchFunc defines the function that is used to fetch the expanded relation records.
type ExpandFetchFunc func(relCollection *models.Collection, relIds []string) ([]*models.Record, error)
// ExpandRecord expands the relations of a single Record model.
func (dao *Dao) ExpandRecord(record *models.Record, expands []string, fetchFunc ExpandFetchFunc) error {
return dao.ExpandRecords([]*models.Record{record}, expands, fetchFunc)
}
// ExpandRecords expands the relations of the provided Record models list.
func (dao *Dao) ExpandRecords(records []*models.Record, expands []string, fetchFunc ExpandFetchFunc) error {
normalized := normalizeExpands(expands)
for _, expand := range normalized {
if err := dao.expandRecords(records, expand, fetchFunc, 1); err != nil {
return err
}
}
return nil
}
// notes:
// - fetchFunc must be non-nil func
// - all records are expected to be from the same collection
// - if MaxExpandDepth is reached, the function returns nil ignoring the remaining expand path
func (dao *Dao) expandRecords(records []*models.Record, expandPath string, fetchFunc ExpandFetchFunc, recursionLevel int) error {
if fetchFunc == nil {
return errors.New("Relation records fetchFunc is not set.")
}
if expandPath == "" || recursionLevel > MaxExpandDepth || len(records) == 0 {
return nil
}
parts := strings.SplitN(expandPath, ".", 2)
// extract the relation field (if exist)
mainCollection := records[0].Collection()
relField := mainCollection.Schema.GetFieldByName(parts[0])
if relField == nil {
return fmt.Errorf("Couldn't find field %q in collection %q.", parts[0], mainCollection.Name)
}
relField.InitOptions()
relFieldOptions, _ := relField.Options.(*schema.RelationOptions)
relCollection, err := dao.FindCollectionByNameOrId(relFieldOptions.CollectionId)
if err != nil {
return fmt.Errorf("Couldn't find collection %q.", relFieldOptions.CollectionId)
}
// extract the id of the relations to expand
relIds := []string{}
for _, record := range records {
relIds = append(relIds, record.GetStringSliceDataValue(relField.Name)...)
}
// fetch rels
rels, relsErr := fetchFunc(relCollection, relIds)
if relsErr != nil {
return relsErr
}
// expand nested fields
if len(parts) > 1 {
err := dao.expandRecords(rels, parts[1], fetchFunc, recursionLevel+1)
if err != nil {
return err
}
}
// reindex with the rel id
indexedRels := map[string]*models.Record{}
for _, rel := range rels {
indexedRels[rel.GetId()] = rel
}
for _, model := range records {
relIds := model.GetStringSliceDataValue(relField.Name)
validRels := []*models.Record{}
for _, id := range relIds {
if rel, ok := indexedRels[id]; ok {
validRels = append(validRels, rel)
}
}
if len(validRels) == 0 {
continue // no valid relations
}
expandData := model.GetExpand()
// normalize and set the expanded relations
if relFieldOptions.MaxSelect == 1 {
expandData[relField.Name] = validRels[0]
} else {
expandData[relField.Name] = validRels
}
model.SetExpand(expandData)
}
return nil
}
// normalizeExpands normalizes expand strings and merges self containing paths
// (eg. ["a.b.c", "a.b", " test ", " ", "test"] -> ["a.b.c", "test"]).
func normalizeExpands(paths []string) []string {
result := []string{}
// normalize paths
normalized := []string{}
for _, p := range paths {
p := strings.ReplaceAll(p, " ", "") // replace spaces
p = strings.Trim(p, ".") // trim incomplete paths
if p == "" {
continue
}
normalized = append(normalized, p)
}
// merge containing paths
for i, p1 := range normalized {
var skip bool
for j, p2 := range normalized {
if i == j {
continue
}
if strings.HasPrefix(p2, p1+".") {
// skip because there is more detailed expand path
skip = true
break
}
}
if !skip {
result = append(result, p1)
}
}
return list.ToUniqueStringSlice(result)
}