diff options
Diffstat (limited to 'internal/bible/validate.go')
| -rw-r--r-- | internal/bible/validate.go | 38 |
1 files changed, 35 insertions, 3 deletions
diff --git a/internal/bible/validate.go b/internal/bible/validate.go index 27ec519..9c99436 100644 --- a/internal/bible/validate.go +++ b/internal/bible/validate.go @@ -3,6 +3,8 @@ package bible import ( "fmt" "sort" + "strings" + "unicode/utf8" ) // CorpusReport is the outcome of validating one corpus's sidecar + text (see @@ -34,7 +36,10 @@ var validPsalmSystems = map[string]bool{"vulgate": true, "hebrew": true, "drb": // chapter, or a repeated verse number within a chapter, is a warning only -- // real source texts legitimately do this (e.g. the LXX's lettered doublet // verses in 3 Kingdoms, or a scanned translation's occasional merged verse), -// and it must never turn a corpus that otherwise loads fine into a failure. +// and it must never turn a corpus that otherwise loads fine into a failure. A +// verse that IS present but whose text is a stub next to the aligned Vulgate +// verse (a likely truncation or lost line -- the one defect the count-based +// checks cannot see) is likewise a warning. func CheckCorpus(code string) CorpusReport { r := CorpusReport{Code: code} @@ -62,6 +67,7 @@ func CheckCorpus(code string) CorpusReport { ref := load("vul") for book, chaps := range c.books { + refBook := ref.books[book] // nil if vul lacks the book (e.g. an EF-only name) for ch, verses := range chaps { seen := map[int]bool{} maxV := 0 @@ -77,9 +83,35 @@ func CheckCorpus(code string) CorpusReport { if maxV > len(seen) { r.Warnings = append(r.Warnings, fmt.Sprintf("%s %d: verse gap (have %d verse(s), highest numbered %d)", book, ch, len(seen), maxV)) } + + // Suspiciously-short verse: present, but a stub next to the SAME verse + // in the reference Vulgate -- a likely truncation or lost line (an + // opening dropped in a scrape, a merge artifact) that the + // presence-only checks cannot see, since the verse IS there. The + // comparison is cross-corpus, not within the chapter, precisely so a + // genuinely terse verse ("Non occides", "Jesus wept") is NOT flagged: + // the Vulgate's own copy is short too, so the guard below fails. It + // fires only when the corpus verse is a stub AND the aligned Vulgate + // verse is substantial. Skipped on the Psalms when the corpus + // renumbers them (psalm_system != vulgate), where the per-verse + // alignment with the Vulgate does not hold. + if refBook != nil && !(book == "Psalms" && m.PsalmSystem != "vulgate") { + if refCh, ok := refBook[ch]; ok { + refLen := make(map[int]int, len(refCh)) + for _, rv := range refCh { + refLen[rv.Verse] = utf8.RuneCountInString(strings.TrimSpace(rv.Text)) + } + for _, v := range verses { + n := utf8.RuneCountInString(strings.TrimSpace(v.Text)) + if rl := refLen[v.Verse]; rl >= 40 && n < 12 && n*4 < rl { + r.Warnings = append(r.Warnings, fmt.Sprintf("%s %d:%d: text suspiciously short (%d chars vs vul %d)", book, ch, v.Verse, n, rl)) + } + } + } + } } - if refChaps, ok := ref.books[book]; ok { - for ch := range refChaps { + if refBook != nil { + for ch := range refBook { if _, have := chaps[ch]; !have { r.Warnings = append(r.Warnings, fmt.Sprintf("%s: missing chapter %d (present in vul)", book, ch)) } |