/
db_query.go
1102 lines (991 loc) · 27.2 KB
/
db_query.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
Copyright 2020 Google LLC
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
package spannertest
import (
"fmt"
"io"
"sort"
"cloud.google.com/go/spanner/spansql"
)
/*
There's several ways to conceptualise SQL queries. The simplest, and what
we implement here, is a series of pipelines that transform the data, whether
pulling from a table (FROM tbl), filtering (WHERE expr), re-ordering (ORDER BY expr)
or other transformations.
The order of operations among those supported by Cloud Spanner is
FROM + JOIN + set ops [TODO: set ops]
WHERE
GROUP BY
aggregation
HAVING [TODO]
SELECT
DISTINCT
ORDER BY
OFFSET
LIMIT
*/
// rowIter represents some iteration over rows of data.
// It is returned by reads and queries.
type rowIter interface {
// Cols returns the metadata about the returned data.
Cols() []colInfo
// Next returns the next row.
// If done, it returns (nil, io.EOF).
Next() (row, error)
}
// aggSentinel is a synthetic expression that refers to an aggregated value.
// It is transient only; it is never stored and only used during evaluation.
type aggSentinel struct {
spansql.Expr
Type spansql.Type
AggIndex int // Index+1 of SELECT list.
}
// nullIter is a rowIter that returns one empty row only.
// This is used for queries without a table.
type nullIter struct {
done bool
}
func (ni *nullIter) Cols() []colInfo { return nil }
func (ni *nullIter) Next() (row, error) {
if ni.done {
return nil, io.EOF
}
ni.done = true
return nil, nil
}
// tableIter is a rowIter that walks a table.
// It assumes the table is locked for the duration.
type tableIter struct {
t *table
rowIndex int // index of next row to return
alias spansql.ID // if non-empty, "AS <alias>"
}
func (ti *tableIter) Cols() []colInfo {
// Build colInfo in the original column order.
cis := make([]colInfo, len(ti.t.cols))
for _, ci := range ti.t.cols {
if ti.alias != "" {
ci.Alias = spansql.PathExp{ti.alias, ci.Name}
}
cis[ti.t.origIndex[ci.Name]] = ci
}
return cis
}
func (ti *tableIter) Next() (row, error) {
if ti.rowIndex >= len(ti.t.rows) {
return nil, io.EOF
}
r := ti.t.rows[ti.rowIndex]
ti.rowIndex++
// Build output row in the original column order.
res := make(row, len(r))
for i, ci := range ti.t.cols {
res[ti.t.origIndex[ci.Name]] = r[i]
}
return res, nil
}
// rawIter is a rowIter with fixed data.
type rawIter struct {
// cols is the metadata about the returned data.
cols []colInfo
// rows holds the result data itself.
rows []row
}
func (raw *rawIter) Cols() []colInfo { return raw.cols }
func (raw *rawIter) Next() (row, error) {
if len(raw.rows) == 0 {
return nil, io.EOF
}
res := raw.rows[0]
raw.rows = raw.rows[1:]
return res, nil
}
func (raw *rawIter) add(src row, colIndexes []int) {
raw.rows = append(raw.rows, src.copyData(colIndexes))
}
// clone makes a shallow copy.
func (raw *rawIter) clone() *rawIter {
return &rawIter{cols: raw.cols, rows: raw.rows}
}
func toRawIter(ri rowIter) (*rawIter, error) {
if raw, ok := ri.(*rawIter); ok {
return raw, nil
}
raw := &rawIter{cols: ri.Cols()}
for {
row, err := ri.Next()
if err == io.EOF {
break
} else if err != nil {
return nil, err
}
raw.rows = append(raw.rows, row.copyAllData())
}
return raw, nil
}
// whereIter applies a WHERE clause.
type whereIter struct {
ri rowIter
ec evalContext
where spansql.BoolExpr
}
func (wi whereIter) Cols() []colInfo { return wi.ri.Cols() }
func (wi whereIter) Next() (row, error) {
for {
row, err := wi.ri.Next()
if err != nil {
return nil, err
}
wi.ec.row = row
b, err := wi.ec.evalBoolExpr(wi.where)
if err != nil {
return nil, err
}
if b != nil && *b {
return row, nil
}
}
}
// selIter applies a SELECT list.
type selIter struct {
ri rowIter
ec evalContext
cis []colInfo
list []spansql.Expr
distinct bool // whether this is a SELECT DISTINCT
seen []row
}
func (si *selIter) Cols() []colInfo { return si.cis }
func (si *selIter) Next() (row, error) {
for {
r, err := si.next()
if err != nil {
return nil, err
}
if si.distinct && !si.keep(r) {
continue
}
return r, nil
}
}
// next retrieves the next row for the SELECT and evaluates its expression list.
func (si *selIter) next() (row, error) {
r, err := si.ri.Next()
if err != nil {
return nil, err
}
si.ec.row = r
var out row
for _, e := range si.list {
if e == spansql.Star {
out = append(out, r...)
} else {
v, err := si.ec.evalExpr(e)
if err != nil {
return nil, err
}
out = append(out, v)
}
}
return out, nil
}
func (si *selIter) keep(r row) bool {
// This is hilariously inefficient; O(N^2) in the number of returned rows.
// Some sort of hashing could be done to deduplicate instead.
// This also breaks on array/struct types.
for _, prev := range si.seen {
if rowEqual(prev, r) {
return false
}
}
si.seen = append(si.seen, r)
return true
}
// offsetIter applies an OFFSET clause.
type offsetIter struct {
ri rowIter
skip int64
}
func (oi *offsetIter) Cols() []colInfo { return oi.ri.Cols() }
func (oi *offsetIter) Next() (row, error) {
for oi.skip > 0 {
_, err := oi.ri.Next()
if err != nil {
return nil, err
}
oi.skip--
}
row, err := oi.ri.Next()
if err != nil {
return nil, err
}
return row, nil
}
// limitIter applies a LIMIT clause.
type limitIter struct {
ri rowIter
rem int64
}
func (li *limitIter) Cols() []colInfo { return li.ri.Cols() }
func (li *limitIter) Next() (row, error) {
if li.rem <= 0 {
return nil, io.EOF
}
row, err := li.ri.Next()
if err != nil {
return nil, err
}
li.rem--
return row, nil
}
type queryParam struct {
Value interface{} // internal representation
Type spansql.Type
}
type queryParams map[string]queryParam // TODO: change key to spansql.Param?
type queryContext struct {
params queryParams
tables []*table // sorted by name
tableIndex map[spansql.ID]*table
locks int
}
func (qc *queryContext) Lock() {
// Take locks in name order.
for _, t := range qc.tables {
t.mu.Lock()
qc.locks++
}
}
func (qc *queryContext) Unlock() {
for _, t := range qc.tables {
t.mu.Unlock()
qc.locks--
}
}
func (d *database) Query(q spansql.Query, params queryParams) (ri rowIter, err error) {
// Figure out the context of the query and take any required locks.
qc, err := d.queryContext(q, params)
if err != nil {
return nil, err
}
qc.Lock()
// On the way out, if there were locks taken, flatten the output
// and release the locks.
if qc.locks > 0 {
defer func() {
if err == nil {
ri, err = toRawIter(ri)
}
qc.Unlock()
}()
}
// Prepare auxiliary expressions to evaluate for ORDER BY.
var aux []spansql.Expr
var desc []bool
for _, o := range q.Order {
aux = append(aux, o.Expr)
desc = append(desc, o.Desc)
}
si, err := d.evalSelect(q.Select, qc)
if err != nil {
return nil, err
}
ri = si
// Apply ORDER BY.
if len(q.Order) > 0 {
// Evaluate the selIter completely, and sort the rows by the auxiliary expressions.
rows, keys, err := evalSelectOrder(si, aux)
if err != nil {
return nil, err
}
sort.Sort(externalRowSorter{rows: rows, keys: keys, desc: desc})
ri = &rawIter{cols: si.cis, rows: rows}
}
// Apply LIMIT, OFFSET.
if q.Limit != nil {
if q.Offset != nil {
off, err := evalLiteralOrParam(q.Offset, params)
if err != nil {
return nil, err
}
ri = &offsetIter{ri: ri, skip: off}
}
lim, err := evalLiteralOrParam(q.Limit, params)
if err != nil {
return nil, err
}
ri = &limitIter{ri: ri, rem: lim}
}
return ri, nil
}
func (d *database) queryContext(q spansql.Query, params queryParams) (*queryContext, error) {
qc := &queryContext{
params: params,
}
// Look for any mentioned tables and add them to qc.tableIndex.
addTable := func(name spansql.ID) error {
if _, ok := qc.tableIndex[name]; ok {
return nil // Already found this table.
}
t, err := d.table(name)
if err != nil {
return err
}
if qc.tableIndex == nil {
qc.tableIndex = make(map[spansql.ID]*table)
}
qc.tableIndex[name] = t
return nil
}
var findTables func(sf spansql.SelectFrom) error
findTables = func(sf spansql.SelectFrom) error {
switch sf := sf.(type) {
default:
return fmt.Errorf("can't prepare query context for SelectFrom of type %T", sf)
case spansql.SelectFromTable:
return addTable(sf.Table)
case spansql.SelectFromJoin:
if err := findTables(sf.LHS); err != nil {
return err
}
return findTables(sf.RHS)
case spansql.SelectFromUnnest:
// TODO: if array paths get supported, this will need more work.
return nil
}
}
for _, sf := range q.Select.From {
if err := findTables(sf); err != nil {
return nil, err
}
}
// Build qc.tables in name order so we can take locks in a well-defined order.
var names []spansql.ID
for name := range qc.tableIndex {
names = append(names, name)
}
sort.Slice(names, func(i, j int) bool { return names[i] < names[j] })
for _, name := range names {
qc.tables = append(qc.tables, qc.tableIndex[name])
}
return qc, nil
}
func (d *database) evalSelect(sel spansql.Select, qc *queryContext) (si *selIter, evalErr error) {
var ri rowIter = &nullIter{}
ec := evalContext{
params: qc.params,
}
// First stage is to identify the data source.
// If there's a FROM then that names a table to use.
if len(sel.From) > 1 {
return nil, fmt.Errorf("selecting with more than one FROM clause not yet supported")
}
if len(sel.From) == 1 {
var err error
ec, ri, err = d.evalSelectFrom(qc, ec, sel.From[0])
if err != nil {
return nil, err
}
}
// Apply WHERE.
if sel.Where != nil {
ri = whereIter{
ri: ri,
ec: ec,
where: sel.Where,
}
}
// Load aliases visible to any future iterators,
// including GROUP BY and ORDER BY. These are not visible to the WHERE clause.
ec.aliases = make(map[spansql.ID]spansql.Expr)
for i, alias := range sel.ListAliases {
ec.aliases[alias] = sel.List[i]
}
// TODO: Add aliases for "1", "2", etc.
// Apply GROUP BY.
// This only reorders rows to group rows together;
// aggregation happens next.
var rowGroups [][2]int // Sequence of half-open intervals of row numbers.
if len(sel.GroupBy) > 0 {
raw, err := toRawIter(ri)
if err != nil {
return nil, err
}
keys := make([][]interface{}, 0, len(raw.rows))
for _, row := range raw.rows {
// Evaluate sort key for this row.
ec.row = row
key, err := ec.evalExprList(sel.GroupBy)
if err != nil {
return nil, err
}
keys = append(keys, key)
}
// Reorder rows base on the evaluated keys.
ers := externalRowSorter{rows: raw.rows, keys: keys}
sort.Sort(ers)
raw.rows = ers.rows
// Record groups as a sequence of row intervals.
// Each group is a run of the same keys.
start := 0
for i := 1; i < len(keys); i++ {
if compareValLists(keys[i-1], keys[i], nil) == 0 {
continue
}
rowGroups = append(rowGroups, [2]int{start, i})
start = i
}
if len(keys) > 0 {
rowGroups = append(rowGroups, [2]int{start, len(keys)})
}
// Clear aliases, since they aren't visible elsewhere.
ec.aliases = nil
ri = raw
}
// Handle aggregation.
// TODO: Support more than one aggregation function; does Spanner support that?
aggI := -1
for i, e := range sel.List {
// Supported aggregate funcs have exactly one arg.
f, ok := e.(spansql.Func)
if !ok || len(f.Args) != 1 {
continue
}
_, ok = aggregateFuncs[f.Name]
if !ok {
continue
}
if aggI > -1 {
return nil, fmt.Errorf("only one aggregate function is supported")
}
aggI = i
}
if aggI > -1 {
raw, err := toRawIter(ri)
if err != nil {
return nil, err
}
if len(sel.GroupBy) == 0 {
// No grouping, so aggregation applies to the entire table (e.g. COUNT(*)).
// This may result in a [0,0) entry for empty inputs.
rowGroups = [][2]int{{0, len(raw.rows)}}
}
fexpr := sel.List[aggI].(spansql.Func)
fn := aggregateFuncs[fexpr.Name]
starArg := fexpr.Args[0] == spansql.Star
if starArg && !fn.AcceptStar {
return nil, fmt.Errorf("aggregate function %s does not accept * as an argument", fexpr.Name)
}
var argType spansql.Type
if !starArg {
ci, err := ec.colInfo(fexpr.Args[0])
if err != nil {
return nil, fmt.Errorf("evaluating aggregate function %s arg type: %v", fexpr.Name, err)
}
argType = ci.Type
}
// Prepare output.
rawOut := &rawIter{
// Same as input columns, but also the aggregate value.
// Add the colInfo for the aggregate at the end
// so we know the type.
// Make a copy for safety.
cols: append([]colInfo(nil), raw.cols...),
}
var aggType spansql.Type
for _, rg := range rowGroups {
// Compute aggregate value across this group.
var values []interface{}
for i := rg[0]; i < rg[1]; i++ {
ec.row = raw.rows[i]
if starArg {
// A non-NULL placeholder is sufficient for aggregation.
values = append(values, 1)
} else {
x, err := ec.evalExpr(fexpr.Args[0])
if err != nil {
return nil, err
}
values = append(values, x)
}
}
x, typ, err := fn.Eval(values, argType)
if err != nil {
return nil, err
}
aggType = typ
var outRow row
// Output for the row group is the first row of the group (arbitrary,
// but it should be representative), and the aggregate value.
// TODO: Should this exclude the aggregated expressions so they can't be selected?
// If the row group is empty then only the aggregation value is used;
// this covers things like COUNT(*) with no matching rows.
if rg[0] < len(raw.rows) {
repRow := raw.rows[rg[0]]
for i := range repRow {
outRow = append(outRow, repRow.copyDataElem(i))
}
} else {
// Fill with NULLs to keep the rows and colInfo aligned.
for i := 0; i < len(rawOut.cols); i++ {
outRow = append(outRow, nil)
}
}
outRow = append(outRow, x)
rawOut.rows = append(rawOut.rows, outRow)
}
if aggType == (spansql.Type{}) {
// Fallback; there might not be any groups.
// TODO: Should this be in aggregateFunc?
aggType = int64Type
}
rawOut.cols = append(raw.cols, colInfo{
Name: spansql.ID(fexpr.SQL()), // TODO: this is a bit hokey, but it is output only
Type: aggType,
AggIndex: aggI + 1,
})
ri = rawOut
ec.cols = rawOut.cols
sel.List[aggI] = aggSentinel{ // Mutate query so evalExpr in selIter picks out the new value.
Type: aggType,
AggIndex: aggI + 1,
}
}
// TODO: Support table sampling.
// Apply SELECT list.
var colInfos []colInfo
for i, e := range sel.List {
if e == spansql.Star {
colInfos = append(colInfos, ec.cols...)
} else {
ci, err := ec.colInfo(e)
if err != nil {
return nil, err
}
if len(sel.ListAliases) > 0 {
alias := sel.ListAliases[i]
if alias != "" {
ci.Name = alias
}
}
// TODO: deal with ci.Name == ""?
colInfos = append(colInfos, ci)
}
}
return &selIter{
ri: ri,
ec: ec,
cis: colInfos,
list: sel.List,
distinct: sel.Distinct, // Apply DISTINCT.
}, nil
}
func (d *database) evalSelectFrom(qc *queryContext, ec evalContext, sf spansql.SelectFrom) (evalContext, rowIter, error) {
switch sf := sf.(type) {
default:
return ec, nil, fmt.Errorf("selecting with FROM clause of type %T not yet supported", sf)
case spansql.SelectFromTable:
t, ok := qc.tableIndex[sf.Table]
if !ok {
// This shouldn't be possible; the queryContext should have discovered missing tables already.
return ec, nil, fmt.Errorf("unknown table %q", sf.Table)
}
ti := &tableIter{t: t}
if sf.Alias != "" {
ti.alias = sf.Alias
} else {
// There is an implicit alias using the table name.
// https://cloud.google.com/spanner/docs/query-syntax#implicit_aliases
ti.alias = sf.Table
}
ec.cols = ti.Cols()
return ec, ti, nil
case spansql.SelectFromJoin:
// TODO: Avoid the toRawIter calls here by doing the RHS recursive evalSelectFrom in joinIter.Next on demand.
lhsEC, lhs, err := d.evalSelectFrom(qc, ec, sf.LHS)
if err != nil {
return ec, nil, err
}
lhsRaw, err := toRawIter(lhs)
if err != nil {
return ec, nil, err
}
rhsEC, rhs, err := d.evalSelectFrom(qc, ec, sf.RHS)
if err != nil {
return ec, nil, err
}
rhsRaw, err := toRawIter(rhs)
if err != nil {
return ec, nil, err
}
ji, ec, err := newJoinIter(lhsRaw, rhsRaw, lhsEC, rhsEC, sf)
if err != nil {
return ec, nil, err
}
return ec, ji, nil
case spansql.SelectFromUnnest:
// TODO: Do all relevant types flow through here? Path expressions might be tricky here.
col, err := ec.colInfo(sf.Expr)
if err != nil {
return ec, nil, fmt.Errorf("evaluating type of UNNEST arg: %v", err)
}
if !col.Type.Array {
return ec, nil, fmt.Errorf("type of UNNEST arg is non-array %s", col.Type.SQL())
}
// The output of this UNNEST is the non-array version.
col.Name = sf.Alias // may be empty
col.Type.Array = false
// Evaluate the expression, and yield a virtual table with one column.
e, err := ec.evalExpr(sf.Expr)
if err != nil {
return ec, nil, fmt.Errorf("evaluating UNNEST arg: %v", err)
}
arr, ok := e.([]interface{})
if !ok {
return ec, nil, fmt.Errorf("evaluating UNNEST arg gave %t, want array", e)
}
var rows []row
for _, v := range arr {
rows = append(rows, row{v})
}
ri := &rawIter{
cols: []colInfo{col},
rows: rows,
}
ec.cols = ri.cols
return ec, ri, nil
}
}
func newJoinIter(lhs, rhs *rawIter, lhsEC, rhsEC evalContext, sfj spansql.SelectFromJoin) (*joinIter, evalContext, error) {
if sfj.On != nil && len(sfj.Using) > 0 {
return nil, evalContext{}, fmt.Errorf("JOIN may not have both ON and USING clauses")
}
if sfj.On == nil && len(sfj.Using) == 0 && sfj.Type != spansql.CrossJoin {
// TODO: This isn't correct for joining against a non-table.
return nil, evalContext{}, fmt.Errorf("non-CROSS JOIN must have ON or USING clause")
}
// Start with the context from the LHS (aliases and params should be the same on both sides).
ji := &joinIter{
jt: sfj.Type,
ec: lhsEC,
primary: lhs,
secondaryOrig: rhs,
primaryOffset: 0,
secondaryOffset: len(lhsEC.cols),
}
switch ji.jt {
case spansql.LeftJoin:
ji.nullPad = true
case spansql.RightJoin:
ji.nullPad = true
// Primary is RHS.
ji.ec = rhsEC
ji.primary, ji.secondaryOrig = rhs, lhs
ji.primaryOffset, ji.secondaryOffset = len(rhsEC.cols), 0
case spansql.FullJoin:
// FULL JOIN is implemented as a LEFT JOIN with tracking for which rows of the RHS
// have been used. Then, at the end of the iteration, the unused RHS rows are emitted.
ji.nullPad = true
ji.used = make([]bool, 0, 10) // arbitrary preallocation
}
ji.ec.cols, ji.ec.row = nil, nil
// Construct a merged evalContext, and prepare the join condition evaluation.
// TODO: Remove ambiguous names here? Or catch them when evaluated?
// TODO: aliases might need work?
if len(sfj.Using) == 0 {
ji.prepNonUsing(sfj.On, lhsEC, rhsEC)
} else {
if err := ji.prepUsing(sfj.Using, lhsEC, rhsEC, ji.jt == spansql.RightJoin); err != nil {
return nil, evalContext{}, err
}
}
return ji, ji.ec, nil
}
// prepNonUsing configures the joinIter to evaluate with an ON clause or no join clause.
// The arg is nil in the latter case.
func (ji *joinIter) prepNonUsing(on spansql.BoolExpr, lhsEC, rhsEC evalContext) {
// Having ON or no clause results in the full set of columns from both sides.
// Force a copy.
ji.ec.cols = append(ji.ec.cols, lhsEC.cols...)
ji.ec.cols = append(ji.ec.cols, rhsEC.cols...)
ji.ec.row = make(row, len(ji.ec.cols))
ji.cond = func(primary, secondary row) (bool, error) {
copy(ji.ec.row[ji.primaryOffset:], primary)
copy(ji.ec.row[ji.secondaryOffset:], secondary)
if on == nil {
// No condition; all rows match.
return true, nil
}
b, err := ji.ec.evalBoolExpr(on)
if err != nil {
return false, err
}
return b != nil && *b, nil
}
ji.zero = func(primary, secondary row) {
for i := range ji.ec.row {
ji.ec.row[i] = nil
}
copy(ji.ec.row[ji.primaryOffset:], primary)
copy(ji.ec.row[ji.secondaryOffset:], secondary)
}
}
func (ji *joinIter) prepUsing(using []spansql.ID, lhsEC, rhsEC evalContext, flipped bool) error {
// Having a USING clause results in the set of named columns once,
// followed by the unnamed columns from both sides.
// lhsUsing is the column indexes in the LHS that the USING clause references.
// rhsUsing is similar.
// lhsNotUsing/rhsNotUsing are the complement.
var lhsUsing, rhsUsing []int
var lhsNotUsing, rhsNotUsing []int
// lhsUsed, rhsUsed are the set of column indexes in lhsUsing/rhsUsing.
lhsUsed, rhsUsed := make(map[int]bool), make(map[int]bool)
for _, id := range using {
lhsi, err := lhsEC.resolveColumnIndex(id)
if err != nil {
return err
}
lhsUsing = append(lhsUsing, lhsi)
lhsUsed[lhsi] = true
rhsi, err := rhsEC.resolveColumnIndex(id)
if err != nil {
return err
}
rhsUsing = append(rhsUsing, rhsi)
rhsUsed[rhsi] = true
// TODO: Should this hide or merge column aliases?
ji.ec.cols = append(ji.ec.cols, lhsEC.cols[lhsi])
}
for i, col := range lhsEC.cols {
if !lhsUsed[i] {
ji.ec.cols = append(ji.ec.cols, col)
lhsNotUsing = append(lhsNotUsing, i)
}
}
for i, col := range rhsEC.cols {
if !rhsUsed[i] {
ji.ec.cols = append(ji.ec.cols, col)
rhsNotUsing = append(rhsNotUsing, i)
}
}
ji.ec.row = make(row, len(ji.ec.cols))
primaryUsing, secondaryUsing := lhsUsing, rhsUsing
if flipped {
primaryUsing, secondaryUsing = secondaryUsing, primaryUsing
}
orNil := func(r row, i int) interface{} {
if r == nil {
return nil
}
return r[i]
}
// populate writes the data to ji.ec.row in the correct positions.
populate := func(primary, secondary row) { // either may be nil
j := 0
if primary != nil {
for _, pi := range primaryUsing {
ji.ec.row[j] = primary[pi]
j++
}
} else {
for _, si := range secondaryUsing {
ji.ec.row[j] = secondary[si]
j++
}
}
lhs, rhs := primary, secondary
if flipped {
rhs, lhs = lhs, rhs
}
for _, i := range lhsNotUsing {
ji.ec.row[j] = orNil(lhs, i)
j++
}
for _, i := range rhsNotUsing {
ji.ec.row[j] = orNil(rhs, i)
j++
}
}
ji.cond = func(primary, secondary row) (bool, error) {
for i, pi := range primaryUsing {
si := secondaryUsing[i]
if compareVals(primary[pi], secondary[si]) != 0 {
return false, nil
}
}
populate(primary, secondary)
return true, nil
}
ji.zero = func(primary, secondary row) {
populate(primary, secondary)
}
return nil
}
type joinIter struct {
jt spansql.JoinType
ec evalContext // combined context
// The "primary" is scanned (consumed), but the secondary is cloned for each primary row.
// Most join types have primary==LHS; a RIGHT JOIN is the exception.
primary, secondaryOrig *rawIter
// The offsets into ec.row that the primary/secondary rows should appear
// in the final output. Not used when there's a USING clause.
primaryOffset, secondaryOffset int
// nullPad is whether primary rows without matching secondary rows
// should be yielded with null padding (e.g. OUTER JOINs).
nullPad bool
primaryRow row // current row from primary, or nil if it is time to advance
secondary *rawIter // current clone of secondary
secondaryRead int // number of rows already read from secondary
any bool // true if any secondary rows have matched primaryRow
// cond reports whether the primary and secondary rows "join" (e.g. the ON clause is true).
// It populates ec.row with the output.
cond func(primary, secondary row) (bool, error)
// zero populates ec.row with the primary or secondary row data (either of which may be nil),
// and sets the remainder to NULL.
// This is used when nullPad is true and a primary or secondary row doesn't match.
zero func(primary, secondary row)
// For FULL JOIN, this tracks the secondary rows that have been used.
// It is non-nil when being used.
used []bool
zeroUnused bool // set when emitting unused secondary rows
unusedIndex int // next index of used to check
}
func (ji *joinIter) Cols() []colInfo { return ji.ec.cols }
func (ji *joinIter) nextPrimary() error {
var err error
ji.primaryRow, err = ji.primary.Next()
if err != nil {
return err
}
ji.secondary = ji.secondaryOrig.clone()
ji.secondaryRead = 0
ji.any = false
return nil
}
func (ji *joinIter) Next() (row, error) {
if ji.primaryRow == nil && !ji.zeroUnused {
err := ji.nextPrimary()
if err == io.EOF && ji.used != nil {
// Drop down to emitting unused secondary rows.
ji.zeroUnused = true
ji.secondary = nil
goto scanJiUsed
}
if err != nil {
return nil, err
}
}
scanJiUsed:
if ji.zeroUnused {
if ji.secondary == nil {
ji.secondary = ji.secondaryOrig.clone()
ji.secondaryRead = 0
}
for ji.unusedIndex < len(ji.used) && ji.used[ji.unusedIndex] {
ji.unusedIndex++
}
if ji.unusedIndex >= len(ji.used) || ji.secondaryRead >= len(ji.used) {