miller/pkg/transformers/join.go

package transformers

import (
	"fmt"
	"os"
	"strings"

	"github.com/johnkerl/miller/v6/pkg/cli"
	"github.com/johnkerl/miller/v6/pkg/input"
	"github.com/johnkerl/miller/v6/pkg/lib"
	"github.com/johnkerl/miller/v6/pkg/mlrval"
	"github.com/johnkerl/miller/v6/pkg/transformers/utils"
	"github.com/johnkerl/miller/v6/pkg/types"
)

const verbNameJoin = "join"

var joinOptions = []OptionSpec{
	{Flag: "-f", Arg: "{left file name}", Type: "filename", Desc: "Left file name for join."},
	{Flag: "-j", Arg: "{a,b,c}", Type: "csv-list", Desc: "Comma-separated join-field names for output."},
	{Flag: "-l", Arg: "{a,b,c}", Type: "csv-list", Desc: "Comma-separated join-field names for left input file; defaults to -j values if omitted."},
	{Flag: "-r", Arg: "{a,b,c}", Type: "csv-list", Desc: "Comma-separated join-field names for right input file(s); defaults to -j values if omitted."},
	{Flag: "--lk", Aliases: []string{"--left-keep-field-names"}, Arg: "{a,b,c}", Type: "csv-list", Desc: "If supplied, this means keep only the specified field names from the left file. Automatically includes the join-field name(s). Helpful for when you only want a limited subset of information from the left file. Tip: you can use --lk \"\": this means the left file becomes solely a row-selector for the input files."},
	{Flag: "--lp", Arg: "{text}", Type: "string", Desc: "Additional prefix for non-join output field names from the left file. Applies to paired and unpaired output records."},
	{Flag: "--rp", Arg: "{text}", Type: "string", Desc: "Additional prefix for non-join output field names from the right file(s). Applies to paired and unpaired output records."},
	{Flag: "--np", Type: "bool", Desc: "Do not emit paired records."},
	{Flag: "--ul", Type: "bool", Desc: "Emit unpaired records from the left file."},
	{Flag: "--ur", Type: "bool", Desc: "Emit unpaired records from the right file(s)."},
	{Flag: "-s", Aliases: []string{"--sorted-input"}, Type: "bool", Desc: "Require sorted input: records must be sorted lexically by their join-field names, else not all records will be paired. The only likely use case for this is with a left file which is too big to fit into system memory otherwise."},
	{Flag: "-u", Type: "bool", Desc: "Enable unsorted input. (This is the default even without -u.) In this case, the entire left file will be loaded into memory."},
	{Flag: "--prepipe", Arg: "{command}", Type: "string", Desc: "Shell command to prepipe the left-file input through. As in main input options; see mlr --help for details. If you wish to use a prepipe command for the main input as well as here, it must be specified there as well as here."},
	{Flag: "--prepipex", Arg: "{command}", Type: "string", Desc: "Shell command to prepipe the left-file input through (no shell quoting). As in main input options; see mlr --help for details."},
}

var JoinSetup = TransformerSetup{
	Verb:         verbNameJoin,
	UsageFunc:    transformerJoinUsage,
	ParseCLIFunc: transformerJoinParseCLI,
	IgnoresInput: false,
	Options:      joinOptions,
}

// Most transformers have option-variables as individual locals within the
// transformerXYZParseCLI function, which are passed as individual arguments to
// the NewTransformerXYZ function. For join, things are a bit more complex
// and we bag up the option-variables into this data structure.

type tJoinOptions struct {
	leftPrefix  string
	rightPrefix string

	outputJoinFieldNames []string
	leftKeepFieldNames   []string
	leftJoinFieldNames   []string
	rightJoinFieldNames  []string

	allowUnsortedInput   bool
	emitPairables        bool
	emitLeftUnpairables  bool
	emitRightUnpairables bool

	leftFileName string
	prepipe      string
	prepipeIsRaw bool

	// These allow the joiner to have its own different format/delimiter for the left-file:
	joinFlagOptions cli.TOptions
}

func newJoinOptions() *tJoinOptions {
	return &tJoinOptions{
		leftPrefix:  "",
		rightPrefix: "",

		outputJoinFieldNames: nil,
		leftKeepFieldNames:   nil,
		leftJoinFieldNames:   nil,
		rightJoinFieldNames:  nil,

		allowUnsortedInput:   true,
		emitPairables:        true,
		emitLeftUnpairables:  false,
		emitRightUnpairables: false,

		leftFileName: "",
		prepipe:      "",
		prepipeIsRaw: false,
	}
}

func transformerJoinUsage(
	o *os.File,
) {
	fmt.Fprintf(o, "Usage: %s %s [options]\n", "mlr", verbNameJoin)
	fmt.Fprintf(o, "Joins records from specified left file name with records from all file names\n")
	fmt.Fprintf(o, "at the end of the Miller argument list.\n")
	fmt.Fprintf(o, "Functionality is essentially the same as the system \"join\" command, but for\n")
	fmt.Fprintf(o, "record streams.\n")
	WriteVerbOptions(o, joinOptions)
	fmt.Fprintf(o, "File-format options default to those for the right file names on the Miller\n")
	fmt.Fprintf(o, "argument list, but may be overridden for the left file as follows. Please see\n")
	fmt.Fprintf(o, "the main \"%s --help\" for more information on syntax for these arguments:\n", "mlr")
	fmt.Fprintf(o, "  -i {format name} for the left-file format, e.g. 'csv' or 'json'. Or, flags\n")
	fmt.Fprintf(o, "     like --icsv, --ijson, etc.: '--icsv' is the same as '-i csv', and so on.\n")
	fmt.Fprintf(o, "  --irs {record-separator character}\n")
	fmt.Fprintf(o, "  --ifs {field-separator character}\n")
	fmt.Fprintf(o, "  --ips {pair-separator character}\n")
	fmt.Fprintf(o, "  --repifs\n")
	fmt.Fprintf(o, "  --implicit-csv-header\n")
	fmt.Fprintf(o, "  --implicit-tsv-header\n")
	fmt.Fprintf(o, "  --no-implicit-csv-header\n")
	fmt.Fprintf(o, "  --no-implicit-tsv-header\n")
	fmt.Fprintf(o, "For example, if you have 'mlr --csv ... join -l foo ... ' then the left-file format will\n")
	fmt.Fprintf(o, "be specified CSV as well unless you override with 'mlr --csv ... join --ijson -l foo' etc.\n")
	fmt.Fprintf(o, "Likewise, if you have 'mlr --csv --implicit-csv-header ...' then the join-in file will be\n")
	fmt.Fprintf(o, "expected to be headerless as well unless you put '--no-implicit-csv-header' after 'join'.\n")
	fmt.Fprintf(o, "Please use \"%s --usage-separator-options\" for information on specifying separators.\n",
		"mlr")
	fmt.Fprintf(o, "Please see https://miller.readthedocs.io/en/latest/reference-verbs#join for more information\n")
	fmt.Fprintf(o, "including examples.\n")
}

func transformerJoinParseCLI(
	pargi *int,
	argc int,
	args []string,
	mainOptions *cli.TOptions, // Options for the right-files
	doConstruct bool, // false for first pass of CLI-parse, true for second pass
) (RecordTransformer, error) {

	// Skip the verb name from the current spot in the mlr command line
	argi := *pargi
	verb := args[argi]
	argi++

	// Parse local flags
	opts := newJoinOptions()

	if mainOptions != nil { // for 'mlr --usage-all-verbs', it's nil
		// TODO: make sure this is a full nested-struct copy.
		opts.joinFlagOptions = *mainOptions // struct copy
	}

	var err error
	for argi < argc /* variable increment: 1 or 2 depending on flag */ {
		opt := args[argi]
		if !strings.HasPrefix(opt, "-") {
			break // No more flag options to process
		}
		if args[argi] == "--" {
			break // All transformers must do this so main-flags can follow verb-flags
		}
		argi++

		switch opt {
		case "-h", "--help":
			transformerJoinUsage(os.Stdout)
			return nil, cli.ErrHelpRequested

		case "--prepipe":
			opts.prepipe, err = cli.VerbGetStringArg(verb, opt, args, &argi, argc)
			if err != nil {
				return nil, err
			}
			opts.prepipeIsRaw = false

		case "--prepipex":
			opts.prepipe, err = cli.VerbGetStringArg(verb, opt, args, &argi, argc)
			if err != nil {
				return nil, err
			}
			opts.prepipeIsRaw = true

		case "-f":
			opts.leftFileName, err = cli.VerbGetStringArg(verb, opt, args, &argi, argc)
			if err != nil {
				return nil, err
			}

		case "-j":
			opts.outputJoinFieldNames, err = cli.VerbGetStringArrayArg(verb, opt, args, &argi, argc)
			if err != nil {
				return nil, err
			}

		case "-l":
			opts.leftJoinFieldNames, err = cli.VerbGetStringArrayArg(verb, opt, args, &argi, argc)
			if err != nil {
				return nil, err
			}

		case "--lk", "--left-keep-field-names":
			opts.leftKeepFieldNames, err = cli.VerbGetStringArrayArg(verb, opt, args, &argi, argc)
			if err != nil {
				return nil, err
			}

		case "-r":
			opts.rightJoinFieldNames, err = cli.VerbGetStringArrayArg(verb, opt, args, &argi, argc)
			if err != nil {
				return nil, err
			}

		case "--lp":
			opts.leftPrefix, err = cli.VerbGetStringArg(verb, opt, args, &argi, argc)
			if err != nil {
				return nil, err
			}

		case "--rp":
			opts.rightPrefix, err = cli.VerbGetStringArg(verb, opt, args, &argi, argc)
			if err != nil {
				return nil, err
			}

		case "--np":
			opts.emitPairables = false

		case "--ul":
			opts.emitLeftUnpairables = true

		case "--ur":
			opts.emitRightUnpairables = true

		case "-u":
			opts.allowUnsortedInput = true

		case "--sorted-input", "-s":
			opts.allowUnsortedInput = false

		default:
			// This is inelegant. For error-proofing we advance argi already in our
			// loop (so individual if-statements don't need to). However,
			// cli.Parse expects it unadvanced.
			largi := argi - 1
			if cli.FLAG_TABLE.Parse(args, argc, &largi, &opts.joinFlagOptions) {
				// This lets mlr main and mlr join have different input formats.
				// Nothing else to handle here.
				argi = largi
			} else {
				return nil, cli.VerbErrorf(verb, "input format option not recognized")
			}
		}
	}

	if err := cli.FinalizeReaderOptions(&opts.joinFlagOptions.ReaderOptions); err != nil {
		return nil, cli.VerbErrorf(verb, "%v", err)
	}

	if opts.leftFileName == "" {
		return nil, cli.VerbErrorf(verb, "need left file name")
	}

	if !opts.emitPairables && !opts.emitLeftUnpairables && !opts.emitRightUnpairables {
		return nil, cli.VerbErrorf(verb, "all emit flags are unset; no output is possible")
	}

	if opts.outputJoinFieldNames == nil {
		return nil, cli.VerbErrorf(verb, "need output field names")
	}

	if opts.leftJoinFieldNames == nil {
		opts.leftJoinFieldNames = opts.outputJoinFieldNames // array copy
	}
	if opts.rightJoinFieldNames == nil {
		opts.rightJoinFieldNames = opts.outputJoinFieldNames // array copy
	}

	llen := len(opts.leftJoinFieldNames)
	rlen := len(opts.rightJoinFieldNames)
	olen := len(opts.outputJoinFieldNames)
	if llen != rlen || llen != olen {
		return nil, cli.VerbErrorf(verb, "left, right, and output field lists must have same length")
	}

	*pargi = argi
	if !doConstruct { // All transformers must do this for main command-line parsing
		return nil, nil
	}

	transformer, err := NewTransformerJoin(opts)
	if err != nil {
		return nil, err
	}

	return transformer, nil
}

type TransformerJoin struct {
	opts *tJoinOptions

	leftFieldNameSet     map[string]bool
	rightFieldNameSet    map[string]bool
	leftKeepFieldNameSet map[string]bool

	// For unsorted/half-streaming input
	ingested                         bool
	leftBucketsByJoinFieldValues     *lib.OrderedMap[*utils.JoinBucket]
	leftUnpairableRecordsAndContexts []*types.RecordAndContext

	// For sorted/doubly-streaming input
	joinBucketKeeper *utils.JoinBucketKeeper

	recordTransformerFunc RecordTransformerFunc
}

func NewTransformerJoin(
	opts *tJoinOptions,
) (*TransformerJoin, error) {

	tr := &TransformerJoin{
		opts: opts,

		leftFieldNameSet:     lib.StringListToSet(opts.leftJoinFieldNames),
		rightFieldNameSet:    lib.StringListToSet(opts.rightJoinFieldNames),
		leftKeepFieldNameSet: lib.StringListToSet(opts.leftKeepFieldNames),

		ingested:                         false,
		leftBucketsByJoinFieldValues:     nil,
		leftUnpairableRecordsAndContexts: nil,
		joinBucketKeeper:                 nil,
	}
	// Suppose left file has "id,foo,bar" and right has "id,baz,quux" and the join field name is
	// "id".  If they ask for --lk id,foo we should keep only id,foo from the left file. But if
	// they ask for --lk foo we should keep id *and* foo fromn the left file.
	if tr.leftKeepFieldNameSet != nil {
		for _, name := range opts.leftJoinFieldNames {
			tr.leftKeepFieldNameSet[name] = true
		}
	}

	if opts.allowUnsortedInput {
		// Half-streaming (default) case: ingest entire left file first.

		tr.leftUnpairableRecordsAndContexts = []*types.RecordAndContext{}
		tr.leftBucketsByJoinFieldValues = lib.NewOrderedMap[*utils.JoinBucket]()
		tr.recordTransformerFunc = tr.transformHalfStreaming

	} else {
		// Doubly-streaming (non-default) case: step left/right files forward.
		// Requires both files be sorted on their join keys in order to not
		// miss anything. This lets people do joins that would otherwise take
		// too much RAM.

		tr.joinBucketKeeper = utils.NewJoinBucketKeeper(
			// opts.prepipe,
			opts.leftFileName,
			&opts.joinFlagOptions.ReaderOptions,
			opts.leftJoinFieldNames,
			tr.leftKeepFieldNameSet,
		)

		tr.recordTransformerFunc = tr.transformDoublyStreaming
	}

	return tr, nil
}

func (tr *TransformerJoin) Transform(
	inrecAndContext *types.RecordAndContext,
	outputRecordsAndContexts *[]*types.RecordAndContext, // list of *types.RecordAndContext
	inputDownstreamDoneChannel <-chan bool,
	outputDownstreamDoneChannel chan<- bool,
) {
	HandleDefaultDownstreamDone(inputDownstreamDoneChannel, outputDownstreamDoneChannel)
	tr.recordTransformerFunc(inrecAndContext, outputRecordsAndContexts,
		inputDownstreamDoneChannel, outputDownstreamDoneChannel)
}

// This is for the half-streaming case. We ingest the entire left file,
// matching each right record against those.
func (tr *TransformerJoin) transformHalfStreaming(
	inrecAndContext *types.RecordAndContext,
	outputRecordsAndContexts *[]*types.RecordAndContext, // list of *types.RecordAndContext
	inputDownstreamDoneChannel <-chan bool,
	outputDownstreamDoneChannel chan<- bool,
) {
	// This can't be done in the CLI-parser since it requires information which
	// isn't known until after the CLI-parser is called.
	//
	// TODO: check if this is still true for the Go port, once everything else
	// is done.
	if !tr.ingested { // First call
		tr.ingestLeftFile()
		tr.ingested = true
	}

	if !inrecAndContext.EndOfStream {
		inrec := inrecAndContext.Record
		groupingKey, hasAllJoinKeys := inrec.GetSelectedValuesJoined(
			tr.opts.rightJoinFieldNames,
		)
		if hasAllJoinKeys {
			leftBucket := tr.leftBucketsByJoinFieldValues.Get(groupingKey)
			if leftBucket == nil {
				if tr.opts.emitRightUnpairables {
					*outputRecordsAndContexts = append(*outputRecordsAndContexts,
						tr.transformRightUnpairedRecord(inrecAndContext))
				}
			} else {
				leftBucket.WasPaired = true
				if tr.opts.emitPairables {
					tr.formAndEmitPairs(
						leftBucket.RecordsAndContexts,
						inrecAndContext,
						outputRecordsAndContexts,
					)
				}
			}
		} else if tr.opts.emitRightUnpairables {
			*outputRecordsAndContexts = append(*outputRecordsAndContexts,
				tr.transformRightUnpairedRecord(inrecAndContext))
		}

	} else { // end of record stream
		if tr.opts.emitLeftUnpairables {
			tr.emitLeftUnpairedBuckets(outputRecordsAndContexts)
			tr.emitLeftUnpairables(outputRecordsAndContexts)
		}
		*outputRecordsAndContexts = append(*outputRecordsAndContexts, inrecAndContext) // emit end-of-stream marker
	}
}

func (tr *TransformerJoin) transformDoublyStreaming(
	rightRecAndContext *types.RecordAndContext,
	outputRecordsAndContexts *[]*types.RecordAndContext, // list of *types.RecordAndContext
	inputDownstreamDoneChannel <-chan bool,
	outputDownstreamDoneChannel chan<- bool,
) {
	keeper := tr.joinBucketKeeper // keystroke-saver

	if !rightRecAndContext.EndOfStream {
		rightRec := rightRecAndContext.Record
		isPaired := false

		rightFieldValues, hasAllJoinKeys := rightRec.ReferenceSelectedValues(
			tr.opts.rightJoinFieldNames,
		)
		if hasAllJoinKeys {
			isPaired = keeper.FindJoinBucket(rightFieldValues)
		}
		if tr.opts.emitLeftUnpairables {
			tr.outputLeftUnpaireds(keeper, outputRecordsAndContexts)
		} else {
			keeper.ReleaseLeftUnpaireds(outputRecordsAndContexts)
		}

		lefts := keeper.JoinBucket.RecordsAndContexts // keystroke-saver

		if !isPaired && tr.opts.emitRightUnpairables {
			*outputRecordsAndContexts = append(*outputRecordsAndContexts,
				tr.transformRightUnpairedRecord(rightRecAndContext))
		}

		if isPaired && tr.opts.emitPairables && lefts != nil {
			tr.formAndEmitPairs(lefts, rightRecAndContext, outputRecordsAndContexts)
		}

	} else { // end of record stream
		keeper.FindJoinBucket(nil)

		if tr.opts.emitLeftUnpairables {
			tr.outputLeftUnpaireds(keeper, outputRecordsAndContexts)
		}

		*outputRecordsAndContexts = append(*outputRecordsAndContexts, rightRecAndContext) // emit end-of-stream marker
	}
}

func (tr *TransformerJoin) outputLeftUnpaireds(
	keeper *utils.JoinBucketKeeper,
	outputRecordsAndContexts *[]*types.RecordAndContext, // list of *types.RecordAndContext
) {
	startIdx := len(*outputRecordsAndContexts)
	keeper.OutputAndReleaseLeftUnpaireds(outputRecordsAndContexts)
	for i := startIdx; i < len(*outputRecordsAndContexts); i++ {
		(*outputRecordsAndContexts)[i] = tr.transformLeftUnpairedRecord((*outputRecordsAndContexts)[i])
	}
}

// This is for the half-streaming case. We ingest the entire left file,
// matching each right record against those.
//
// Note: this logic is very similar to that in stream.go, which is what
// processes the main/right files.

func (tr *TransformerJoin) ingestLeftFile() {
	readerOpts := &tr.opts.joinFlagOptions.ReaderOptions

	// Instantiate the record-reader
	// TODO: perhaps increase recordsPerBatch, and/or refactor
	recordReader, err := input.Create(readerOpts, 1)
	if err != nil {
		// TODO: propagate error to caller
		return
	}

	// Set the initial context for the left-file.
	//
	// Since Go is concurrent, the context struct needs to be duplicated and
	// passed through the channels along with each record.
	initialContext := types.NewNilContext()
	initialContext.UpdateForStartOfFile(tr.opts.leftFileName)

	// Set up channels for the record-reader.
	readerChannel := make(chan []*types.RecordAndContext, 2) // list of *types.RecordAndContext
	errorChannel := make(chan error, 1)
	downstreamDoneChannel := make(chan bool, 1)

	// Start the record reader.
	// TODO: prepipe
	leftFileNameArray := [1]string{tr.opts.leftFileName}
	go recordReader.Read(leftFileNameArray[:], *initialContext, readerChannel, errorChannel, downstreamDoneChannel)

	// Ingest parsed records and bucket them by their join-field values.  E.g.
	// if the join-field is "id" then put all records with id=1 in one bucket,
	// all those with id=2 in another bucket, etc. And any records lacking an
	// "id" field go into the unpairable list.
	done := false
	for !done {
		select {

		case err := <-errorChannel:
			// TODO: propagate error to caller
			_ = err
			return

		case leftrecsAndContexts := <-readerChannel:
			// TODO: temp for batch-reader refactor
			lib.InternalCodingErrorIf(len(leftrecsAndContexts) != 1)
			leftrecAndContext := leftrecsAndContexts[0]
			leftrecAndContext.Record = utils.KeepLeftFieldNames(leftrecAndContext.Record, tr.leftKeepFieldNameSet)

			if leftrecAndContext.EndOfStream {
				done = true
				break // breaks the switch, not the for, in Golang
			}
			leftrec := leftrecAndContext.Record
			if leftrec == nil {
				// E.g. the only payload is OutputString or EndOfStream
				continue
			}

			groupingKey, leftFieldValues, ok := leftrec.GetSelectedValuesAndJoined(
				tr.opts.leftJoinFieldNames,
			)
			if ok {
				bucket := tr.leftBucketsByJoinFieldValues.Get(groupingKey)
				if bucket == nil { // New key-field-value: new bucket and hash-map entry
					bucket := utils.NewJoinBucket(leftFieldValues)
					bucket.RecordsAndContexts = append(bucket.RecordsAndContexts, leftrecAndContext)
					tr.leftBucketsByJoinFieldValues.Put(groupingKey, bucket)
				} else { // Previously seen key-field-value: append record to bucket
					bucket.RecordsAndContexts = append(bucket.RecordsAndContexts, leftrecAndContext)
				}
			} else {
				tr.leftUnpairableRecordsAndContexts = append(tr.leftUnpairableRecordsAndContexts, leftrecAndContext)
			}
		}
	}
}

// This helper method is used by the half-streaming/unsorted join, as well as
// the doubly-streaming/sorted join.

func (tr *TransformerJoin) formAndEmitPairs(
	leftRecordsAndContexts []*types.RecordAndContext,
	rightRecordAndContext *types.RecordAndContext,
	outputRecordsAndContexts *[]*types.RecordAndContext, // list of *types.RecordAndContext
) {
	////fmt.Println("-- pairs start") // VERBOSE
	// Loop over each to-be-paired-with record from the left file.
	for _, leftRecordAndContext := range leftRecordsAndContexts {
		////fmt.Println("-- pairs pe") // VERBOSE
		leftrec := leftRecordAndContext.Record
		rightrec := rightRecordAndContext.Record

		// Allocate a new output record which is the join of the left and right records.
		outrec := mlrval.NewMlrmapAsRecord()

		// Add the joined-on fields to the new output record
		n := len(tr.opts.leftJoinFieldNames)
		for i := range n {
			// These arrays are already guaranteed same-length by CLI parser
			leftJoinFieldName := tr.opts.leftJoinFieldNames[i]
			outputJoinFieldName := tr.opts.outputJoinFieldNames[i]
			value := leftrec.Get(leftJoinFieldName)
			if value != nil {
				outrec.PutCopy(outputJoinFieldName, value)
			}
		}

		// Add the left-record fields not already added
		for pl := leftrec.Head; pl != nil; pl = pl.Next {
			_, ok := tr.leftFieldNameSet[pl.Key]
			if !ok {
				key := tr.opts.leftPrefix + pl.Key
				outrec.PutCopy(key, pl.Value)
			}
		}

		// Add the right-record fields not already added
		for pr := rightrec.Head; pr != nil; pr = pr.Next {
			_, ok := tr.rightFieldNameSet[pr.Key]
			if !ok {
				key := tr.opts.rightPrefix + pr.Key
				outrec.PutCopy(key, pr.Value)
			}
		}
		////fmt.Println("-- pairs outrec") // VERBOSE
		////outrec.Print() // VERBOSE

		// Clone the right record's context (NR, FILENAME, etc) to use for the new output record
		context := rightRecordAndContext.Context // struct copy
		outrecAndContext := types.NewRecordAndContext(outrec, &context)

		// Emit the new joined record on the downstream channel
		*outputRecordsAndContexts = append(*outputRecordsAndContexts, outrecAndContext)
	}
	////fmt.Println("-- pairs end") // VERBOSE
}

// There are two kinds of left non-pair records: (a) those lacking the
// specified join-keys -- can't possibly pair with anything on the right; (b)
// those having the join-keys but not matching with a record on the right.
//
// Example: join on "id" field. Records lacking an "id" field are in the first
// category.  Now suppose there's a left record with id=0, but there were three
// right-file records with id-field values 1,2,3. Then the id=0 left records is
// in the second category.

func (tr *TransformerJoin) emitLeftUnpairables(
	outputRecordsAndContexts *[]*types.RecordAndContext, // list of *types.RecordAndContext
) {
	for _, recordAndContext := range tr.leftUnpairableRecordsAndContexts {
		*outputRecordsAndContexts = append(*outputRecordsAndContexts,
			tr.transformLeftUnpairedRecord(recordAndContext))
	}
}

func (tr *TransformerJoin) emitLeftUnpairedBuckets(
	outputRecordsAndContexts *[]*types.RecordAndContext, // list of *types.RecordAndContext
) {
	for pe := tr.leftBucketsByJoinFieldValues.Head; pe != nil; pe = pe.Next {
		bucket := pe.Value
		if !bucket.WasPaired {
			for _, recordAndContext := range bucket.RecordsAndContexts {
				*outputRecordsAndContexts = append(*outputRecordsAndContexts,
					tr.transformLeftUnpairedRecord(recordAndContext))
			}
		}
	}
}

// transformLeftUnpairedRecord and transformRightUnpairedRecord rename the
// join-field names to the output-join-field names and apply the side's prefix
// to all non-join field names. This keeps unpaired-record column names
// consistent with those produced by paired joins, so downstream operations
// like `unsparsify` align columns across paired and unpaired output.
func (tr *TransformerJoin) transformLeftUnpairedRecord(
	inrecAndContext *types.RecordAndContext,
) *types.RecordAndContext {
	return tr.transformUnpairedRecord(
		inrecAndContext, tr.opts.leftJoinFieldNames, tr.opts.leftPrefix,
	)
}

func (tr *TransformerJoin) transformRightUnpairedRecord(
	inrecAndContext *types.RecordAndContext,
) *types.RecordAndContext {
	return tr.transformUnpairedRecord(
		inrecAndContext, tr.opts.rightJoinFieldNames, tr.opts.rightPrefix,
	)
}

func (tr *TransformerJoin) transformUnpairedRecord(
	inrecAndContext *types.RecordAndContext,
	joinFieldNames []string,
	prefix string,
) *types.RecordAndContext {
	inrec := inrecAndContext.Record
	if inrec == nil {
		return inrecAndContext
	}

	needsRename := false
	for i, name := range joinFieldNames {
		if name != tr.opts.outputJoinFieldNames[i] {
			needsRename = true
			break
		}
	}
	if !needsRename && prefix == "" {
		return inrecAndContext
	}

	renameMap := make(map[string]string, len(joinFieldNames))
	for i, name := range joinFieldNames {
		renameMap[name] = tr.opts.outputJoinFieldNames[i]
	}

	outrec := mlrval.NewMlrmapAsRecord()
	for pe := inrec.Head; pe != nil; pe = pe.Next {
		if outName, ok := renameMap[pe.Key]; ok {
			outrec.PutCopy(outName, pe.Value)
		} else {
			outrec.PutCopy(prefix+pe.Key, pe.Value)
		}
	}

	context := inrecAndContext.Context // struct copy
	return types.NewRecordAndContext(outrec, &context)
}