importccl: compute expected number of data columns only once

This commit moves the computation of finding the expected number of data
columns to the creation of the input converter, rather than
re-calculating it on every row. The memory footprint of loading all of
the visible columns for every row was noticeable.

Release note (performance improvement): CSV imports should now be
slightly faster.

pkg/ccl/importccl/import_processor.go

@@ -156,9 +156,7 @@ func makeInputConverter(
 		if len(singleTableTargetCols) == 0 && !formatHasNamedColumns(spec.Format.Format) {
 			for _, col := range singleTable.VisibleColumns() {
 				if col.IsComputed() {
-					// TODO(before merge): Replace this issue number with one that
-					// summarizes the above comment in a ticket.
-					return nil, unimplemented.NewWithIssueDetail(42846, "import.computed",
+					return nil, unimplemented.NewWithIssueDetail(56002, "import.computed",
 						"to use computed columns, use IMPORT INTO")
 				}
 			}

pkg/ccl/importccl/read_import_csv.go

@@ -26,7 +26,9 @@ import (
 
 type csvInputReader struct {
 	importCtx *parallelImportContext
-	opts      roachpb.CSVOptions
+	// The number of columns that we expect in the CSV data file.
+	numExpectedDataCols int
+	opts                roachpb.CSVOptions
 }
 
 var _ inputConverter = &csvInputReader{}
@@ -40,6 +42,10 @@ func newCSVInputReader(
 	targetCols tree.NameList,
 	evalCtx *tree.EvalContext,
 ) *csvInputReader {
+	numExpectedDataCols := len(targetCols)
+	if numExpectedDataCols == 0 {
+		numExpectedDataCols = len(tableDesc.VisibleColumns())
+	}
 	return &csvInputReader{
 		importCtx: &parallelImportContext{
 			walltime:   walltime,
@@ -49,7 +55,8 @@ func newCSVInputReader(
 			targetCols: targetCols,
 			kvCh:       kvCh,
 		},
-		opts: opts,
+		numExpectedDataCols: numExpectedDataCols,
+		opts:                opts,
 	}
 }
 
@@ -86,14 +93,14 @@ func (c *csvInputReader) readFile(
 }
 
 type csvRowProducer struct {
-	importCtx       *parallelImportContext
-	opts            *roachpb.CSVOptions
-	csv             *csv.Reader
-	rowNum          int64
-	err             error
-	record          []string
-	progress        func() float32
-	expectedColumns tree.NameList
+	importCtx          *parallelImportContext
+	opts               *roachpb.CSVOptions
+	csv                *csv.Reader
+	rowNum             int64
+	err                error
+	record             []string
+	progress           func() float32
+	numExpectedColumns int
 }
 
 var _ importRowProducer = &csvRowProducer{}
@@ -132,10 +139,7 @@ func strRecord(record []string, sep rune) string {
 // Row() implements importRowProducer interface.
 func (p *csvRowProducer) Row() (interface{}, error) {
 	p.rowNum++
-	expectedColsLen := len(p.expectedColumns)
-	if expectedColsLen == 0 {
-		expectedColsLen = len(p.importCtx.tableDesc.VisibleColumns())
-	}
+	expectedColsLen := p.numExpectedColumns
 
 	if len(p.record) == expectedColsLen {
 		// Expected number of columns.
@@ -207,11 +211,11 @@ func newCSVPipeline(c *csvInputReader, input *fileReader) (*csvRowProducer, *csv
 	cr.Comment = c.opts.Comment
 
 	producer := &csvRowProducer{
-		importCtx:       c.importCtx,
-		opts:            &c.opts,
-		csv:             cr,
-		progress:        func() float32 { return input.ReadFraction() },
-		expectedColumns: c.importCtx.targetCols,
+		importCtx:          c.importCtx,
+		opts:               &c.opts,
+		csv:                cr,
+		progress:           func() float32 { return input.ReadFraction() },
+		numExpectedColumns: c.numExpectedDataCols,
 	}
 	consumer := &csvRowConsumer{
 		importCtx: c.importCtx,