package cache import ( "encoding/gob" "fmt" "strings" "github.com/MichaelMure/git-bug/entity" ) // Package initialisation used to register the type for (de)serialization func init() { gob.Register(IdentityExcerpt{}) } var _ Excerpt = &IdentityExcerpt{} // IdentityExcerpt hold a subset of the identity values to be able to sort and // filter identities efficiently without having to read and compile each raw // identity. type IdentityExcerpt struct { id entity.Id Name string Login string ImmutableMetadata map[string]string } func NewIdentityExcerpt(i *IdentityCache) *IdentityExcerpt { return &IdentityExcerpt{ id: i.Id(), Name: i.Name(), Login: i.Login(), ImmutableMetadata: i.ImmutableMetadata(), } } func (i *IdentityExcerpt) setId(id entity.Id) { i.id = id } func (i *IdentityExcerpt) Id() entity.Id { return i.id } // DisplayName return a non-empty string to display, representing the // identity, based on the non-empty values. func (i *IdentityExcerpt) DisplayName() string { switch { case i.Name == "" && i.Login != "": return i.Login case i.Name != "" && i.Login == "": return i.Name case i.Name != "" && i.Login != "": return fmt.Sprintf("%s (%s)", i.Name, i.Login) } panic("invalid person data") } // Match matches a query with the identity name, login and ID prefixes func (i *IdentityExcerpt) Match(query string) bool { return i.id.HasPrefix(query) || strings.Contains(strings.ToLower(i.Name), query) || strings.Contains(strings.ToLower(i.Login), query) } /* * Sorting */ type IdentityById []*IdentityExcerpt func (b IdentityById) Len() int { return len(b) } func (b IdentityById) Less(i, j int) bool { return b[i].id < b[j].id } func (b IdentityById) Swap(i, j int) { b[i], b[j] = b[j], b[i] }