path: root/skate/zippy.go

// This file contains various "reducers", e.g. working on two data streams and
// applying a function on groups of documents with a shared key.
//
// Note: This is a bit repetitive, but not want to introduce any other
// abstraction for now. Since most of the logic is in the grouper functions, we
// could make them top level and then assemble the zipkey runner on the fly.
//
// The most confusing aspect currently is the variety of schemas hidden within
// the readers (and string groups): release, ref, ref-as-release, open library,
// wikipedia, ...
package skate

import (
	"fmt"
	"io"
	"log"
	"sort"
	"strings"
	"time"

	"git.archive.org/martin/cgraph/skate/set"
	"git.archive.org/martin/cgraph/skate/zipkey"
	json "github.com/segmentio/encoding/json"
)

// groupLogf logs a message alongsize a serialized group for debugging.
func groupLogf(g *zipkey.Group, s string, vs ...interface{}) {
	log.Printf(s, vs...)
	b, _ := json.MarshalIndent(g, "", "    ")
	log.Println(string(b))
}

// ZippyExact takes a release and refs reader (key, doc) and assigns a fixed
// match result.
func ZippyExact(releases, refs io.Reader, matchResult MatchResult, w io.Writer) error {
	var (
		enc     = json.NewEncoder(w)
		keyer   = makeKeyFunc("\t", 1)
		i       = 0
		grouper = func(g *zipkey.Group) error {
			i++
			if i%10000 == 0 {
				log.Printf("processed %v groups", i)
			}
			var (
				target *Release
				ref    *Ref
				err    error
			)
			if len(g.G0) == 0 || len(g.G1) == 0 {
				return nil
			}
			if target, err = stringToRelease(Cut(g.G0[0], 2)); err != nil {
				groupLogf(g, "[skip] failed to parse release: %v", err)
				return nil
			}
			for _, line := range g.G1 {
				if ref, err = stringToRef(Cut(line, 2)); err != nil {
					groupLogf(g, "[skip] failed to parse ref: %v", err)
					continue
				}
				var bref BiblioRef
				bref.SourceReleaseIdent = ref.ReleaseIdent
				bref.SourceWorkIdent = ref.WorkIdent
				bref.SourceReleaseStage = ref.ReleaseStage
				bref.SourceYear = fmt.Sprintf("%d", ref.ReleaseYear)
				bref.RefIndex = ref.Index + 1 // we want 1-index (also helps with omitempty)
				bref.RefKey = ref.Key
				bref.TargetReleaseIdent = target.Ident
				bref.TargetWorkIdent = target.WorkID
				bref.MatchProvenance = ref.RefSource
				bref.MatchStatus = matchResult.Status.Short()
				bref.MatchReason = matchResult.Reason.Short()
				if err := enc.Encode(bref); err != nil {
					return err
				}
			}
			return nil
		}
	)
	zipper := zipkey.New(releases, refs, keyer, grouper)
	return zipper.Run()
}

// ZippyExactReleases takes two release readers (key, doc) and assigns a fixed
// match result.
func ZippyExactReleases(olReader, reReader io.Reader, matchResult MatchResult, w io.Writer) error {
	var (
		enc     = json.NewEncoder(w)
		keyer   = makeKeyFunc("\t", 1)
		i       = 0
		grouper = func(g *zipkey.Group) error {
			i++
			if i%10000 == 0 {
				log.Printf("processed %v groups", i)
			}
			var (
				target, re *Release
				err        error
			)
			if len(g.G0) == 0 || len(g.G1) == 0 {
				return nil
			}
			if target, err = stringToRelease(Cut(g.G0[0], 2)); err != nil {
				groupLogf(g, "[skip] failed to parse release: %v", err)
				return nil
			}
			for _, line := range g.G1 {
				if re, err = stringToRelease(Cut(line, 2)); err != nil {
					groupLogf(g, "[skip] failed to parse release: %v", err)
					continue
				}
				if target.WorkID == "" {
					continue
				}
				var bref BiblioRef
				bref.SourceReleaseIdent = re.Ident
				bref.SourceWorkIdent = re.WorkID
				bref.SourceReleaseStage = re.ReleaseStage
				bref.SourceYear = fmt.Sprintf("%d", re.ReleaseYear())
				bref.RefIndex = re.Extra.Skate.Ref.Index + 1 // we want 1-index (also helps with omitempty)
				bref.RefKey = re.Extra.Skate.Ref.Key
				bref.TargetOpenLibraryWork = target.WorkID
				bref.MatchProvenance = re.Extra.Skate.Ref.Source
				bref.MatchStatus = matchResult.Status.Short()
				bref.MatchReason = matchResult.Reason.Short()
				if err := enc.Encode(bref); err != nil {
					return err
				}
			}
			return nil
		}
	)
	zipper := zipkey.New(olReader, reReader, keyer, grouper)
	return zipper.Run()
}

// ZippyExactWiki takes a release and wiki reader (key, doc) and assigns a
// fixed match result.
func ZippyExactWiki(releases, wiki io.Reader, mr MatchResult, w io.Writer) error {
	var (
		enc     = json.NewEncoder(w)
		keyer   = makeKeyFunc("\t", 1)
		grouper = func(g *zipkey.Group) error {
			var (
				target *Release
				wiki   *MinimalCitations
				err    error
			)
			if len(g.G0) == 0 || len(g.G1) == 0 {
				return nil
			}
			if target, err = stringToRelease(Cut(g.G0[0], 2)); err != nil {
				return err
			}
			for _, line := range g.G1 {
				if wiki, err = stringToWiki(Cut(line, 2)); err != nil {
					return err
				}
				var bref BiblioRef
				bref.Key = fmt.Sprintf("%s_%s", slugifyString(wiki.PageTitle), target.Ident) // XXX: what should we use?
				bref.SourceWikipediaArticle = wiki.PageTitle
				bref.TargetReleaseIdent = target.Ident
				bref.TargetWorkIdent = target.WorkID
				bref.MatchProvenance = "wikipedia"
				bref.MatchStatus = mr.Status.Short()
				bref.MatchReason = mr.Reason.Short()
				if err := enc.Encode(bref); err != nil {
					return err
				}
			}
			return nil
		}
	)
	zipper := zipkey.New(releases, wiki, keyer, grouper)
	return zipper.Run()
}

// ZippyVerifyRefs takes a release and refs reader (key, doc), run fuzzy
// verification and will emit a biblioref document, if exact or strong match.
func ZippyVerifyRefs(releases, refs io.Reader, w io.Writer) error {
	var (
		enc     = json.NewEncoder(w)
		keyer   = makeKeyFunc("\t", 1)
		grouper = func(g *zipkey.Group) error {
			var (
				re, pivot *Release
				err       error
			)
			if len(g.G0) == 0 || len(g.G1) == 0 {
				return nil
			}
			if pivot, err = stringToRelease(Cut(g.G0[0], 2)); err != nil {
				return err
			}
			for _, line := range g.G1 {
				if re, err = stringToRelease(Cut(line, 2)); err != nil {
					return err
				}
				result := Verify(pivot, re)
				switch result.Status {
				case StatusExact, StatusStrong:
					if result.Reason == ReasonDOI {
						continue
					}
					br := generateBiblioRef(re, pivot, result, "fuzzy")
					if err := enc.Encode(br); err != nil {
						return err
					}
				default:
					// XXX: We want to add unmatched pieces as well; here?  We
					// probably want to do a single final pass to complete the
					// dataset.
				}
			}
			return nil
		}
	)
	zipper := zipkey.New(releases, refs, keyer, grouper)
	return zipper.Run()
}

// ZippyVerifyRefsOpenLibraryTable takes OL editions (as release) and refs (as
// release) and emits a match table for manual inspection.
func ZippyVerifyRefsOpenLibraryTable(olr, refs io.Reader, w io.Writer) error {
	var (
		keyer   = makeKeyFunc("\t", 1)
		grouper = func(g *zipkey.Group) error {
			var (
				re, pivot *Release
				err       error
			)
			if len(g.G0) == 0 || len(g.G1) == 0 {
				return nil
			}
			// We take a single edition from OL.
			if pivot, err = stringToRelease(Cut(g.G0[0], 2)); err != nil {
				return err
			}
			for _, line := range g.G1 {
				if re, err = stringToRelease(Cut(line, 2)); err != nil {
					return err
				}
				// The refs have a container name, but not a title, but here we
				// compare against titles from open library.
				re.Title = re.ContainerName
				result := Verify(pivot, re)
				fmt.Printf("%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n",
					result.Status.Short(),
					result.Reason.Short(),
					pivot.Extra.OpenLibrary.WorkID,
					FindByPrefix(pivot.Extra.OpenLibrary.SourceRecords, "ia:"),
					re.Ident,
					CutSep(g.G0[0], "\t", 1),
					pivot.Title,
					re.Title)
			}
			return nil
		}
	)
	zipper := zipkey.New(olr, refs, keyer, grouper)
	return zipper.Run()
}

// ZippyVerifyRefsOpenLibrary takes OL editions (as release) and refs (as
// release) and emits a match table for manual inspection.
func ZippyVerifyRefsOpenLibrary(olr, refs io.Reader, w io.Writer) error {
	var (
		enc     = json.NewEncoder(w)
		keyer   = makeKeyFunc("\t", 1)
		grouper = func(g *zipkey.Group) error {
			var (
				ref, pivot *Release // ref (reference), pivot (open library)
				err        error
			)
			if len(g.G0) == 0 || len(g.G1) == 0 {
				return nil
			}
			// We take a single edition from OL.
			if pivot, err = stringToRelease(Cut(g.G0[0], 2)); err != nil {
				return err
			}
			for _, line := range g.G1 {
				if ref, err = stringToRelease(Cut(line, 2)); err != nil {
					return err
				}
				// The refs have a container name, but not a title, but here we
				// compare against titles from open library.
				ref.Title = ref.ContainerName
				result := Verify(pivot, ref)
				switch result.Status {
				case StatusExact, StatusStrong:
					var bref BiblioRef
					bref.SourceReleaseIdent = ref.Ident
					bref.SourceWorkIdent = ref.WorkID
					bref.SourceReleaseStage = ref.ReleaseStage
					bref.SourceYear = fmt.Sprintf("%d", ref.ReleaseYear())
					bref.RefIndex = ref.Extra.Skate.Ref.Index + 1 // we want 1-index (also helps with omitempty)
					bref.RefKey = ref.Extra.Skate.Ref.Key
					bref.TargetOpenLibraryWork = pivot.WorkID
					bref.MatchProvenance = ref.Extra.Skate.Ref.Source
					bref.MatchStatus = result.Status.Short()
					bref.MatchReason = result.Reason.Short()
					if err := enc.Encode(bref); err != nil {
						return err
					}
				default:
				}
			}
			return nil
		}
	)
	zipper := zipkey.New(olr, refs, keyer, grouper)
	return zipper.Run()
}

// ZippyBrefAugment takes all matched docs from bref and adds docs from raw
// refs, which have not been matched. It also gets rid of duplicate matches.
//
// We can identify, which docs have been matched by checking the source ident,
// ref index and key.
func ZippyBrefAugment(bref, raw io.Reader, w io.Writer) error {
	var (
		stats   = statsAugment{}
		enc     = json.NewEncoder(w)
		keyer   = makeKeyFunc("\t", 1)
		grouper = func(g *zipkey.Group) error {
			// g.G0 contains matched docs for a given work id, g.G1 all raw
			// refs, with the same work id.

			// First, iterate over all matches and sort out duplicates, e.g.
			// docs that have the same source and target id.
			log.Printf("group K=%s, G0=%d, G1=%d", g.Key, len(g.G0), len(g.G1))
			matched, err := uniqueMatches(CutBatch(g.G0, 2), &stats)
			if err != nil {
				return err
			}
			var refs = make([]*Ref, len(g.G1))
			for i := 0; i < len(refs); i++ {
				var (
					data []byte = []byte(Cut(g.G1[i], 2))
					ref  Ref
				)
				if err := json.Unmarshal(data, &ref); err != nil {
					return err
				}
				refs[i] = &ref
			}
			matched = matchedRefsExtend(matched, refs, &stats)
			for _, bref := range matched {
				stats.total++
				if err := enc.Encode(bref); err != nil {
					return err
				}
			}
			return nil
		}
	)
	zipper := zipkey.New(bref, raw, keyer, grouper)
	err := zipper.Run()
	log.Println(stats)
	return err
}

// matchedRefsExtend takes a set of (unique) biblioref docs and will emit that
// set of biblioref docs (unchanged) plus raw references as biblioref, which
// did not result in a match (determined by e.g. ref key and index).  XXX: We
// may have duplicate refs as well - how to distinguish them?
func matchedRefsExtend(matched []*BiblioRef, refs []*Ref, stats *statsAugment) []*BiblioRef {
	s := set.New() // store key + index of matched items
	for _, m := range matched {
		s.Add(m.RefKey + fmt.Sprintf("%d", m.RefIndex))
	}
	for _, r := range refs {
		if s.Contains(r.Key + fmt.Sprintf("%d", r.Index)) {
			stats.skipMatchedRef++
			log.Printf("skip-matched-ref [%d]: from %d matches; ident=%v, title=%s, key=%v, index=%v",
				stats.skipMatchedRef, len(matched), r.ReleaseIdent, r.Biblio.Title, r.Key, fmt.Sprintf("%d", r.Index))
			continue
		}
		var bref BiblioRef
		bref.IndexedTs = time.Now().UTC().Format(time.RFC3339)
		bref.Key = fmt.Sprintf("%s_%d", r.ReleaseIdent, r.Index)
		bref.RefIndex = r.Index
		bref.RefKey = r.Key
		bref.SourceReleaseIdent = r.ReleaseIdent
		bref.SourceReleaseStage = r.ReleaseStage
		bref.SourceWorkIdent = r.WorkIdent
		bref.SourceYear = fmt.Sprintf("%d", r.ReleaseYear)
		bref.TargetUnstructured = r.Biblio.Unstructured
		// Reuse fields for debugging, for now.
		bref.MatchStatus = StatusUnmatched.Short()
		bref.MatchReason = ReasonUnknown.Short()
		// bref.Extra.Ref = *r
		matched = append(matched, &bref)
	}
	return matched
}

// uniqueMatches takes a list of bref docs (unserialized) and will return a
// list of deserialized bref docs, containing unique matches only (e.g. filter
// out things duplicate matches, e.g. from exact and fuzzy). We are including
// "skate-bref-id" post-processing here as well (but there is surely a better
// place for that).
func uniqueMatches(docs []string, stats *statsAugment) (result []*BiblioRef, err error) {
	var brefs []*BiblioRef
	for _, doc := range docs {
		var bref BiblioRef
		if err := json.Unmarshal([]byte(doc), &bref); err != nil {
			return nil, err
		}
		// On-the-fly add elasticsearch "_id" and indexed timestamp, if not already set.
		if bref.Key == "" && bref.SourceReleaseIdent != "" {
			bref.Key = fmt.Sprintf("%s_%d", bref.SourceReleaseIdent, bref.RefIndex)
			bref.IndexedTs = time.Now().UTC().Format(time.RFC3339)
		}
		brefs = append(brefs, &bref)
	}
	// Make sure exact matches come first.
	sort.Slice(brefs, func(i, j int) bool {
		return brefs[i].MatchStatus == StatusExact.Short()
	})
	seen := set.New()
	for _, doc := range brefs {
		h := doc.LinkHash()
		if seen.Contains(h) {
			stats.skipDuplicatedBref++
			log.Printf("skip-dup-bref [%d]: hash=%v source=%v status=%v reason=%v",
				stats.skipDuplicatedBref, h, doc.SourceReleaseIdent, doc.MatchStatus, doc.MatchReason)
			continue
		}
		seen.Add(h)
		result = append(result, doc)
	}
	return result, nil
}

type statsAugment struct {
	skipDuplicatedBref int64
	skipMatchedRef     int64
	total              int64
}

func (s statsAugment) String() string {
	return fmt.Sprintf("total=%d, skipMatchedRef=%d, skipDuplicatedBref=%d",
		s.total, s.skipMatchedRef, s.skipDuplicatedBref)
}

// CutBatch runs Cut over a list of lines.
func CutBatch(lines []string, column int) (result []string) {
	for _, line := range lines {
		result = append(result, Cut(line, column))
	}
	return result
}

// Cut returns a specific column (1-indexed) from a line, returns empty string
// if column is invalid.
func Cut(line string, column int) string {
	return CutSep(line, "\t", column)
}

// CutSep allows to specify a separator, column is 1-indexed.
func CutSep(line, sep string, column int) string {
	parts := strings.Split(strings.TrimSpace(line), sep)
	if len(parts) < column {
		return ""
	} else {
		return parts[column-1]
	}
}

// FindByPrefix return the first element for a slice of strings, which matches a prefix.
func FindByPrefix(ss []string, prefix string) string {
	for _, s := range ss {
		if strings.HasPrefix(s, prefix) {
			return s
		}
	}
	return ""
}

// makeKeyFunc creates a function that can be used as keyFunc, selecting a
// column from fields separated by sep; column is 1-indexed.
func makeKeyFunc(sep string, column int) func(string) (string, error) {
	return func(s string) (string, error) {
		if k := CutSep(s, sep, column); k == "" {
			return k, fmt.Errorf("cannot get key from column %d in line (len=%d): %s", column, len(s), s)
		} else {
			return k, nil
		}
	}
}

func stringToRelease(s string) (r *Release, err error) {
	err = json.Unmarshal([]byte(s), &r)
	return
}

func stringToRef(s string) (r *Ref, err error) {
	err = json.Unmarshal([]byte(s), &r)
	return
}

func stringToWiki(s string) (r *MinimalCitations, err error) {
	err = json.Unmarshal([]byte(s), &r)
	return
}

func stringToBiblioref(s string) (r *BiblioRef, err error) {
	err = json.Unmarshal([]byte(s), &r)
	return
}