// Code generated by ent, DO NOT EDIT. package ent import ( "context" "database/sql/driver" "fmt" "math" "code.icod.de/dalu/ka/ent/post" "code.icod.de/dalu/ka/ent/predicate" "code.icod.de/dalu/ka/ent/profile" "entgo.io/ent" "entgo.io/ent/dialect/sql" "entgo.io/ent/dialect/sql/sqlgraph" "entgo.io/ent/schema/field" "github.com/google/uuid" ) // ProfileQuery is the builder for querying Profile entities. type ProfileQuery struct { config ctx *QueryContext order []profile.OrderOption inters []Interceptor predicates []predicate.Profile withPosts *PostQuery modifiers []func(*sql.Selector) loadTotal []func(context.Context, []*Profile) error withNamedPosts map[string]*PostQuery // intermediate query (i.e. traversal path). sql *sql.Selector path func(context.Context) (*sql.Selector, error) } // Where adds a new predicate for the ProfileQuery builder. func (pq *ProfileQuery) Where(ps ...predicate.Profile) *ProfileQuery { pq.predicates = append(pq.predicates, ps...) return pq } // Limit the number of records to be returned by this query. func (pq *ProfileQuery) Limit(limit int) *ProfileQuery { pq.ctx.Limit = &limit return pq } // Offset to start from. func (pq *ProfileQuery) Offset(offset int) *ProfileQuery { pq.ctx.Offset = &offset return pq } // Unique configures the query builder to filter duplicate records on query. // By default, unique is set to true, and can be disabled using this method. func (pq *ProfileQuery) Unique(unique bool) *ProfileQuery { pq.ctx.Unique = &unique return pq } // Order specifies how the records should be ordered. func (pq *ProfileQuery) Order(o ...profile.OrderOption) *ProfileQuery { pq.order = append(pq.order, o...) return pq } // QueryPosts chains the current query on the "posts" edge. func (pq *ProfileQuery) QueryPosts() *PostQuery { query := (&PostClient{config: pq.config}).Query() query.path = func(ctx context.Context) (fromU *sql.Selector, err error) { if err := pq.prepareQuery(ctx); err != nil { return nil, err } selector := pq.sqlQuery(ctx) if err := selector.Err(); err != nil { return nil, err } step := sqlgraph.NewStep( sqlgraph.From(profile.Table, profile.FieldID, selector), sqlgraph.To(post.Table, post.FieldID), sqlgraph.Edge(sqlgraph.O2M, false, profile.PostsTable, profile.PostsColumn), ) fromU = sqlgraph.SetNeighbors(pq.driver.Dialect(), step) return fromU, nil } return query } // First returns the first Profile entity from the query. // Returns a *NotFoundError when no Profile was found. func (pq *ProfileQuery) First(ctx context.Context) (*Profile, error) { nodes, err := pq.Limit(1).All(setContextOp(ctx, pq.ctx, ent.OpQueryFirst)) if err != nil { return nil, err } if len(nodes) == 0 { return nil, &NotFoundError{profile.Label} } return nodes[0], nil } // FirstX is like First, but panics if an error occurs. func (pq *ProfileQuery) FirstX(ctx context.Context) *Profile { node, err := pq.First(ctx) if err != nil && !IsNotFound(err) { panic(err) } return node } // FirstID returns the first Profile ID from the query. // Returns a *NotFoundError when no Profile ID was found. func (pq *ProfileQuery) FirstID(ctx context.Context) (id uuid.UUID, err error) { var ids []uuid.UUID if ids, err = pq.Limit(1).IDs(setContextOp(ctx, pq.ctx, ent.OpQueryFirstID)); err != nil { return } if len(ids) == 0 { err = &NotFoundError{profile.Label} return } return ids[0], nil } // FirstIDX is like FirstID, but panics if an error occurs. func (pq *ProfileQuery) FirstIDX(ctx context.Context) uuid.UUID { id, err := pq.FirstID(ctx) if err != nil && !IsNotFound(err) { panic(err) } return id } // Only returns a single Profile entity found by the query, ensuring it only returns one. // Returns a *NotSingularError when more than one Profile entity is found. // Returns a *NotFoundError when no Profile entities are found. func (pq *ProfileQuery) Only(ctx context.Context) (*Profile, error) { nodes, err := pq.Limit(2).All(setContextOp(ctx, pq.ctx, ent.OpQueryOnly)) if err != nil { return nil, err } switch len(nodes) { case 1: return nodes[0], nil case 0: return nil, &NotFoundError{profile.Label} default: return nil, &NotSingularError{profile.Label} } } // OnlyX is like Only, but panics if an error occurs. func (pq *ProfileQuery) OnlyX(ctx context.Context) *Profile { node, err := pq.Only(ctx) if err != nil { panic(err) } return node } // OnlyID is like Only, but returns the only Profile ID in the query. // Returns a *NotSingularError when more than one Profile ID is found. // Returns a *NotFoundError when no entities are found. func (pq *ProfileQuery) OnlyID(ctx context.Context) (id uuid.UUID, err error) { var ids []uuid.UUID if ids, err = pq.Limit(2).IDs(setContextOp(ctx, pq.ctx, ent.OpQueryOnlyID)); err != nil { return } switch len(ids) { case 1: id = ids[0] case 0: err = &NotFoundError{profile.Label} default: err = &NotSingularError{profile.Label} } return } // OnlyIDX is like OnlyID, but panics if an error occurs. func (pq *ProfileQuery) OnlyIDX(ctx context.Context) uuid.UUID { id, err := pq.OnlyID(ctx) if err != nil { panic(err) } return id } // All executes the query and returns a list of Profiles. func (pq *ProfileQuery) All(ctx context.Context) ([]*Profile, error) { ctx = setContextOp(ctx, pq.ctx, ent.OpQueryAll) if err := pq.prepareQuery(ctx); err != nil { return nil, err } qr := querierAll[[]*Profile, *ProfileQuery]() return withInterceptors[[]*Profile](ctx, pq, qr, pq.inters) } // AllX is like All, but panics if an error occurs. func (pq *ProfileQuery) AllX(ctx context.Context) []*Profile { nodes, err := pq.All(ctx) if err != nil { panic(err) } return nodes } // IDs executes the query and returns a list of Profile IDs. func (pq *ProfileQuery) IDs(ctx context.Context) (ids []uuid.UUID, err error) { if pq.ctx.Unique == nil && pq.path != nil { pq.Unique(true) } ctx = setContextOp(ctx, pq.ctx, ent.OpQueryIDs) if err = pq.Select(profile.FieldID).Scan(ctx, &ids); err != nil { return nil, err } return ids, nil } // IDsX is like IDs, but panics if an error occurs. func (pq *ProfileQuery) IDsX(ctx context.Context) []uuid.UUID { ids, err := pq.IDs(ctx) if err != nil { panic(err) } return ids } // Count returns the count of the given query. func (pq *ProfileQuery) Count(ctx context.Context) (int, error) { ctx = setContextOp(ctx, pq.ctx, ent.OpQueryCount) if err := pq.prepareQuery(ctx); err != nil { return 0, err } return withInterceptors[int](ctx, pq, querierCount[*ProfileQuery](), pq.inters) } // CountX is like Count, but panics if an error occurs. func (pq *ProfileQuery) CountX(ctx context.Context) int { count, err := pq.Count(ctx) if err != nil { panic(err) } return count } // Exist returns true if the query has elements in the graph. func (pq *ProfileQuery) Exist(ctx context.Context) (bool, error) { ctx = setContextOp(ctx, pq.ctx, ent.OpQueryExist) switch _, err := pq.FirstID(ctx); { case IsNotFound(err): return false, nil case err != nil: return false, fmt.Errorf("ent: check existence: %w", err) default: return true, nil } } // ExistX is like Exist, but panics if an error occurs. func (pq *ProfileQuery) ExistX(ctx context.Context) bool { exist, err := pq.Exist(ctx) if err != nil { panic(err) } return exist } // Clone returns a duplicate of the ProfileQuery builder, including all associated steps. It can be // used to prepare common query builders and use them differently after the clone is made. func (pq *ProfileQuery) Clone() *ProfileQuery { if pq == nil { return nil } return &ProfileQuery{ config: pq.config, ctx: pq.ctx.Clone(), order: append([]profile.OrderOption{}, pq.order...), inters: append([]Interceptor{}, pq.inters...), predicates: append([]predicate.Profile{}, pq.predicates...), withPosts: pq.withPosts.Clone(), // clone intermediate query. sql: pq.sql.Clone(), path: pq.path, } } // WithPosts tells the query-builder to eager-load the nodes that are connected to // the "posts" edge. The optional arguments are used to configure the query builder of the edge. func (pq *ProfileQuery) WithPosts(opts ...func(*PostQuery)) *ProfileQuery { query := (&PostClient{config: pq.config}).Query() for _, opt := range opts { opt(query) } pq.withPosts = query return pq } // GroupBy is used to group vertices by one or more fields/columns. // It is often used with aggregate functions, like: count, max, mean, min, sum. // // Example: // // var v []struct { // CreatedAt time.Time `json:"created_at,omitempty"` // Count int `json:"count,omitempty"` // } // // client.Profile.Query(). // GroupBy(profile.FieldCreatedAt). // Aggregate(ent.Count()). // Scan(ctx, &v) func (pq *ProfileQuery) GroupBy(field string, fields ...string) *ProfileGroupBy { pq.ctx.Fields = append([]string{field}, fields...) grbuild := &ProfileGroupBy{build: pq} grbuild.flds = &pq.ctx.Fields grbuild.label = profile.Label grbuild.scan = grbuild.Scan return grbuild } // Select allows the selection one or more fields/columns for the given query, // instead of selecting all fields in the entity. // // Example: // // var v []struct { // CreatedAt time.Time `json:"created_at,omitempty"` // } // // client.Profile.Query(). // Select(profile.FieldCreatedAt). // Scan(ctx, &v) func (pq *ProfileQuery) Select(fields ...string) *ProfileSelect { pq.ctx.Fields = append(pq.ctx.Fields, fields...) sbuild := &ProfileSelect{ProfileQuery: pq} sbuild.label = profile.Label sbuild.flds, sbuild.scan = &pq.ctx.Fields, sbuild.Scan return sbuild } // Aggregate returns a ProfileSelect configured with the given aggregations. func (pq *ProfileQuery) Aggregate(fns ...AggregateFunc) *ProfileSelect { return pq.Select().Aggregate(fns...) } func (pq *ProfileQuery) prepareQuery(ctx context.Context) error { for _, inter := range pq.inters { if inter == nil { return fmt.Errorf("ent: uninitialized interceptor (forgotten import ent/runtime?)") } if trv, ok := inter.(Traverser); ok { if err := trv.Traverse(ctx, pq); err != nil { return err } } } for _, f := range pq.ctx.Fields { if !profile.ValidColumn(f) { return &ValidationError{Name: f, err: fmt.Errorf("ent: invalid field %q for query", f)} } } if pq.path != nil { prev, err := pq.path(ctx) if err != nil { return err } pq.sql = prev } return nil } func (pq *ProfileQuery) sqlAll(ctx context.Context, hooks ...queryHook) ([]*Profile, error) { var ( nodes = []*Profile{} _spec = pq.querySpec() loadedTypes = [1]bool{ pq.withPosts != nil, } ) _spec.ScanValues = func(columns []string) ([]any, error) { return (*Profile).scanValues(nil, columns) } _spec.Assign = func(columns []string, values []any) error { node := &Profile{config: pq.config} nodes = append(nodes, node) node.Edges.loadedTypes = loadedTypes return node.assignValues(columns, values) } if len(pq.modifiers) > 0 { _spec.Modifiers = pq.modifiers } for i := range hooks { hooks[i](ctx, _spec) } if err := sqlgraph.QueryNodes(ctx, pq.driver, _spec); err != nil { return nil, err } if len(nodes) == 0 { return nodes, nil } if query := pq.withPosts; query != nil { if err := pq.loadPosts(ctx, query, nodes, func(n *Profile) { n.Edges.Posts = []*Post{} }, func(n *Profile, e *Post) { n.Edges.Posts = append(n.Edges.Posts, e) }); err != nil { return nil, err } } for name, query := range pq.withNamedPosts { if err := pq.loadPosts(ctx, query, nodes, func(n *Profile) { n.appendNamedPosts(name) }, func(n *Profile, e *Post) { n.appendNamedPosts(name, e) }); err != nil { return nil, err } } for i := range pq.loadTotal { if err := pq.loadTotal[i](ctx, nodes); err != nil { return nil, err } } return nodes, nil } func (pq *ProfileQuery) loadPosts(ctx context.Context, query *PostQuery, nodes []*Profile, init func(*Profile), assign func(*Profile, *Post)) error { fks := make([]driver.Value, 0, len(nodes)) nodeids := make(map[uuid.UUID]*Profile) for i := range nodes { fks = append(fks, nodes[i].ID) nodeids[nodes[i].ID] = nodes[i] if init != nil { init(nodes[i]) } } query.withFKs = true query.Where(predicate.Post(func(s *sql.Selector) { s.Where(sql.InValues(s.C(profile.PostsColumn), fks...)) })) neighbors, err := query.All(ctx) if err != nil { return err } for _, n := range neighbors { fk := n.profile_posts if fk == nil { return fmt.Errorf(`foreign-key "profile_posts" is nil for node %v`, n.ID) } node, ok := nodeids[*fk] if !ok { return fmt.Errorf(`unexpected referenced foreign-key "profile_posts" returned %v for node %v`, *fk, n.ID) } assign(node, n) } return nil } func (pq *ProfileQuery) sqlCount(ctx context.Context) (int, error) { _spec := pq.querySpec() if len(pq.modifiers) > 0 { _spec.Modifiers = pq.modifiers } _spec.Node.Columns = pq.ctx.Fields if len(pq.ctx.Fields) > 0 { _spec.Unique = pq.ctx.Unique != nil && *pq.ctx.Unique } return sqlgraph.CountNodes(ctx, pq.driver, _spec) } func (pq *ProfileQuery) querySpec() *sqlgraph.QuerySpec { _spec := sqlgraph.NewQuerySpec(profile.Table, profile.Columns, sqlgraph.NewFieldSpec(profile.FieldID, field.TypeUUID)) _spec.From = pq.sql if unique := pq.ctx.Unique; unique != nil { _spec.Unique = *unique } else if pq.path != nil { _spec.Unique = true } if fields := pq.ctx.Fields; len(fields) > 0 { _spec.Node.Columns = make([]string, 0, len(fields)) _spec.Node.Columns = append(_spec.Node.Columns, profile.FieldID) for i := range fields { if fields[i] != profile.FieldID { _spec.Node.Columns = append(_spec.Node.Columns, fields[i]) } } } if ps := pq.predicates; len(ps) > 0 { _spec.Predicate = func(selector *sql.Selector) { for i := range ps { ps[i](selector) } } } if limit := pq.ctx.Limit; limit != nil { _spec.Limit = *limit } if offset := pq.ctx.Offset; offset != nil { _spec.Offset = *offset } if ps := pq.order; len(ps) > 0 { _spec.Order = func(selector *sql.Selector) { for i := range ps { ps[i](selector) } } } return _spec } func (pq *ProfileQuery) sqlQuery(ctx context.Context) *sql.Selector { builder := sql.Dialect(pq.driver.Dialect()) t1 := builder.Table(profile.Table) columns := pq.ctx.Fields if len(columns) == 0 { columns = profile.Columns } selector := builder.Select(t1.Columns(columns...)...).From(t1) if pq.sql != nil { selector = pq.sql selector.Select(selector.Columns(columns...)...) } if pq.ctx.Unique != nil && *pq.ctx.Unique { selector.Distinct() } for _, p := range pq.predicates { p(selector) } for _, p := range pq.order { p(selector) } if offset := pq.ctx.Offset; offset != nil { // limit is mandatory for offset clause. We start // with default value, and override it below if needed. selector.Offset(*offset).Limit(math.MaxInt32) } if limit := pq.ctx.Limit; limit != nil { selector.Limit(*limit) } return selector } // WithNamedPosts tells the query-builder to eager-load the nodes that are connected to the "posts" // edge with the given name. The optional arguments are used to configure the query builder of the edge. func (pq *ProfileQuery) WithNamedPosts(name string, opts ...func(*PostQuery)) *ProfileQuery { query := (&PostClient{config: pq.config}).Query() for _, opt := range opts { opt(query) } if pq.withNamedPosts == nil { pq.withNamedPosts = make(map[string]*PostQuery) } pq.withNamedPosts[name] = query return pq } // ProfileGroupBy is the group-by builder for Profile entities. type ProfileGroupBy struct { selector build *ProfileQuery } // Aggregate adds the given aggregation functions to the group-by query. func (pgb *ProfileGroupBy) Aggregate(fns ...AggregateFunc) *ProfileGroupBy { pgb.fns = append(pgb.fns, fns...) return pgb } // Scan applies the selector query and scans the result into the given value. func (pgb *ProfileGroupBy) Scan(ctx context.Context, v any) error { ctx = setContextOp(ctx, pgb.build.ctx, ent.OpQueryGroupBy) if err := pgb.build.prepareQuery(ctx); err != nil { return err } return scanWithInterceptors[*ProfileQuery, *ProfileGroupBy](ctx, pgb.build, pgb, pgb.build.inters, v) } func (pgb *ProfileGroupBy) sqlScan(ctx context.Context, root *ProfileQuery, v any) error { selector := root.sqlQuery(ctx).Select() aggregation := make([]string, 0, len(pgb.fns)) for _, fn := range pgb.fns { aggregation = append(aggregation, fn(selector)) } if len(selector.SelectedColumns()) == 0 { columns := make([]string, 0, len(*pgb.flds)+len(pgb.fns)) for _, f := range *pgb.flds { columns = append(columns, selector.C(f)) } columns = append(columns, aggregation...) selector.Select(columns...) } selector.GroupBy(selector.Columns(*pgb.flds...)...) if err := selector.Err(); err != nil { return err } rows := &sql.Rows{} query, args := selector.Query() if err := pgb.build.driver.Query(ctx, query, args, rows); err != nil { return err } defer rows.Close() return sql.ScanSlice(rows, v) } // ProfileSelect is the builder for selecting fields of Profile entities. type ProfileSelect struct { *ProfileQuery selector } // Aggregate adds the given aggregation functions to the selector query. func (ps *ProfileSelect) Aggregate(fns ...AggregateFunc) *ProfileSelect { ps.fns = append(ps.fns, fns...) return ps } // Scan applies the selector query and scans the result into the given value. func (ps *ProfileSelect) Scan(ctx context.Context, v any) error { ctx = setContextOp(ctx, ps.ctx, ent.OpQuerySelect) if err := ps.prepareQuery(ctx); err != nil { return err } return scanWithInterceptors[*ProfileQuery, *ProfileSelect](ctx, ps.ProfileQuery, ps, ps.inters, v) } func (ps *ProfileSelect) sqlScan(ctx context.Context, root *ProfileQuery, v any) error { selector := root.sqlQuery(ctx) aggregation := make([]string, 0, len(ps.fns)) for _, fn := range ps.fns { aggregation = append(aggregation, fn(selector)) } switch n := len(*ps.selector.flds); { case n == 0 && len(aggregation) > 0: selector.Select(aggregation...) case n != 0 && len(aggregation) > 0: selector.AppendSelect(aggregation...) } rows := &sql.Rows{} query, args := selector.Query() if err := ps.driver.Query(ctx, query, args, rows); err != nil { return err } defer rows.Close() return sql.ScanSlice(rows, v) }