Artifact 2f417b7e99b0a809ae92cc5817a98598289cca3ec28262f5e99df8e30f96dcec:
- File
internal/deck/marshal.go
— part of check-in
[e9ec7390c7]
at
2026-10-18 16:44:03
on branch feat/submanifest-reader
— feat(deck): canonical encoding and sub-manifest serializers
draft-fossil-submanifests-00 §6.2 fixes minimal escaping for every filename a tree construction emits: exactly space, newline and backslash are escaped, every other byte raw. CanonicalEncode is that encoding; S cards and a sub-manifest's F cards now marshal with it, while a flat manifest keeps §4.3's, so no existing artifact changes name.
MarshalLeaf and MarshalInterior serialize the two sub-manifest shapes, and LeafCard gives one leaf F card's bytes for the §6.3 cost. A test reproduces the Appendix A.2 leaf byte for byte. (user: [email protected], size: 11009) [annotate] [blame] [check-ins using]
package deck import ( "fmt" "sort" "strings" "time" ) func (d *Deck) Marshal() ([]byte, error) { if d == nil { panic("deck.Marshal: d must not be nil") } // §4.7.16: a tag name that is entirely hexadecimal after the sign is // ambiguous with an artifact hash, so the parser rejects it. Refusing to // emit one keeps the codec symmetric -- without this check a caller can // build and store an artifact this repository can no longer read back. // Canonical fossil never produces one either: it prefixes branch and // user tags with "sym-". for _, tag := range d.T { if isHexToken(tag.Name) { return nil, fmt.Errorf("T-card name %q must not be entirely hexadecimal", tag.Name) } } var b strings.Builder marshalCards(&b, d) body := b.String() zHash := computeZ([]byte(body)) return []byte(fmt.Sprintf("%sZ %s\n", body, zHash)), nil } // marshalCards writes all cards to the builder in strict ASCII order // (A B C D E F G H I J K L M N P Q R S T U W Z), in contiguous groups so each // writer stays small. func marshalCards(b *strings.Builder, d *Deck) { if b == nil { panic("deck.marshalCards: b must not be nil") } if d == nil { panic("deck.marshalCards: d must not be nil") } marshalCardsABCDE(b, d) marshalCardF(b, d) marshalCardsGHIJKL(b, d) marshalCardsMNPQR(b, d) marshalCardS(b, d) marshalCardsTUW(b, d) } // marshalCardsABCDE writes the A, B, C, D and E cards. func marshalCardsABCDE(b *strings.Builder, d *Deck) { if d.A != nil { b.WriteString("A ") b.WriteString(FossilEncode(d.A.Filename)) b.WriteString(" ") // §4.3.1/§4.7.2: the A target is the one escape-decoded hash position, // so it is escape-encoded on the way out (mirror of parseACard). b.WriteString(FossilEncode(d.A.Target)) if d.A.Source != "" { b.WriteString(" ") b.WriteString(d.A.Source) } b.WriteString("\n") } if d.B != "" { fmt.Fprintf(b, "B %s\n", d.B) } // §4.2 erratum: a present-but-empty comment re-emits as "C " + LF, so the // artifact round-trips. A caller that never set a comment leaves C empty // and cPresent false, and no C card is written. if d.C != "" || d.cPresent { fmt.Fprintf(b, "C %s\n", FossilEncode(d.C)) } if !d.D.IsZero() { // §4.6.4: the D card is written to at least millisecond precision (its // writer form), extended to keep whatever finer fraction the parser // read, so a sub-millisecond timestamp round-trips. fmt.Fprintf(b, "D %s\n", marshalTimestamp(d.D, 3)) } if d.E != nil { // The E card's writer form is second precision; a finer fraction the // parser read is kept so it round-trips. fmt.Fprintf(b, "E %s %s\n", marshalTimestamp(d.E.Date, 0), d.E.UUID) } } // marshalCardF writes the F (file) cards, sorted by decoded filename. A // sub-manifest's filenames take the canonical encoding of // draft-fossil-submanifests-00 §6.2; every other artifact's take §4.3's. func marshalCardF(b *strings.Builder, d *Deck) { if len(d.F) == 0 { return } encode := FossilEncode if d.Type == Submanifest { encode = CanonicalEncode } sorted := make([]FileCard, len(d.F)) copy(sorted, d.F) sort.Slice(sorted, func(i, j int) bool { return Compare(sorted[i].Name, sorted[j].Name) < 0 }) for _, f := range sorted { writeFileCard(b, f, encode) } } // writeFileCard writes one F card, its filename and old name through encode. func writeFileCard(b *strings.Builder, f FileCard, encode func(string) string) { b.WriteString("F ") b.WriteString(encode(f.Name)) if f.UUID != "" { b.WriteString(" ") b.WriteString(f.UUID) // Canonical Fossil forces a "w" permission placeholder on a // rename card whose perm would otherwise be empty, so the // prior-name field keeps its 4th positional slot rather than // being misread as the permission (src/checkin.c:1999). perm := f.Perm if perm == "" && f.OldName != "" { perm = "w" } if perm != "" { b.WriteString(" ") b.WriteString(perm) } // The prior name is positional and only meaningful after a // UUID; a deletion card (no UUID) must never carry one, or // the old name would be misread as the hash. if f.OldName != "" { b.WriteString(" ") b.WriteString(encode(f.OldName)) } } b.WriteString("\n") } // marshalCardsGHIJKL writes the G, H, I, J, K and L cards. func marshalCardsGHIJKL(b *strings.Builder, d *Deck) { if d.G != "" { fmt.Fprintf(b, "G %s\n", d.G) } if d.H != "" { fmt.Fprintf(b, "H %s\n", FossilEncode(d.H)) } if d.I != "" { fmt.Fprintf(b, "I %s\n", d.I) } if len(d.J) > 0 { // §4.5.2 requires a strictly ascending J run, which the parser // enforces. Callers build TicketField slices in whatever order // suits them, so wire order is the marshaller's to impose — the // same responsibility it already takes for F, M and T. sorted := make([]TicketField, len(d.J)) copy(sorted, d.J) sort.Slice(sorted, func(i, j int) bool { return Compare(sorted[i].Name, sorted[j].Name) < 0 }) for _, j := range sorted { // §4.7.8 mirror of the parse direction: the name is written // verbatim, the value is escape-encoded. if j.Value != "" { fmt.Fprintf(b, "J %s %s\n", j.Name, FossilEncode(j.Value)) } else { fmt.Fprintf(b, "J %s\n", j.Name) } } } if d.K != "" { // §4.7.9: a parser may accept an uppercase ticket ID, a writer emits // lowercase. fmt.Fprintf(b, "K %s\n", strings.ToLower(d.K)) } if d.L != "" { fmt.Fprintf(b, "L %s\n", FossilEncode(d.L)) } } // marshalCardsMNPQR writes the M, N, P, Q and R cards. func marshalCardsMNPQR(b *strings.Builder, d *Deck) { if len(d.M) > 0 { sorted := make([]string, len(d.M)) copy(sorted, d.M) sort.Slice(sorted, func(i, j int) bool { return Compare(sorted[i], sorted[j]) < 0 }) for _, m := range sorted { fmt.Fprintf(b, "M %s\n", m) } } if d.N != "" || d.nPresent { // §4.3.1: the N mimetype is escape-decoded on parse, so encode it here; // a present-but-empty N re-emits as "N " + LF. fmt.Fprintf(b, "N %s\n", FossilEncode(d.N)) } if len(d.P) > 0 { fmt.Fprintf(b, "P %s\n", strings.Join(d.P, " ")) } else if d.pPresent { // §4.7.13: a bare P card (the initial empty baseline) re-emits as "P" // + LF, with no trailing space. b.WriteString("P\n") } for _, q := range d.Q { prefix := "+" if q.IsBackout { prefix = "-" } if q.Baseline != "" { fmt.Fprintf(b, "Q %s%s %s\n", prefix, q.Target, q.Baseline) } else { fmt.Fprintf(b, "Q %s%s\n", prefix, q.Target) } } if d.R != "" { // §6.3.4: the checksum is always rendered lowercase, though the // parse grammar is case-insensitive. fmt.Fprintf(b, "R %s\n", strings.ToLower(d.R)) } } // marshalCardS writes the S cards of a tree manifest or interior sub-manifest // (draft-fossil-submanifests-00 §3.6), sorted by decoded filename, under the // canonical encoding of §6.2. func marshalCardS(b *strings.Builder, d *Deck) { if len(d.S) == 0 { return } sorted := make([]SubmanifestCard, len(d.S)) copy(sorted, d.S) sort.Slice(sorted, func(i, j int) bool { return Compare(sorted[i].Name, sorted[j].Name) < 0 }) for _, sc := range sorted { fmt.Fprintf(b, "S %s %s\n", CanonicalEncode(sc.Name), sc.UUID) } } // MarshalLeaf serializes a leaf sub-manifest (draft-fossil-submanifests-00 // §6.2 step 2): files, in canonical filename order, as F cards and a Z card. // Every file must carry a hash, since a sub-manifest holds no delete marker // (§4.4). func MarshalLeaf(files []FileCard) []byte { if len(files) == 0 { panic("deck.MarshalLeaf: files must not be empty") } for _, f := range files { if f.UUID == "" { panic("deck.MarshalLeaf: every file must carry a hash") } } return marshalSubmanifest(&Deck{Type: Submanifest, F: files}) } // MarshalInterior serializes an interior sub-manifest (§6.2 step 3): one S // card per member, in key order, and a Z card. func MarshalInterior(refs []SubmanifestCard) []byte { if len(refs) == 0 { panic("deck.MarshalInterior: refs must not be empty") } return marshalSubmanifest(&Deck{Type: Submanifest, S: refs}) } // LeafCard is the F card line a leaf sub-manifest carries for f, its // terminating newline included: what the cost of §6.3 measures a leaf by. func LeafCard(f FileCard) string { var b strings.Builder writeFileCard(&b, f, CanonicalEncode) return b.String() } func marshalSubmanifest(d *Deck) []byte { var b strings.Builder marshalCards(&b, d) body := b.String() return []byte(body + "Z " + computeZ([]byte(body)) + "\n") } // marshalCardsTUW writes the T, U and W cards. func marshalCardsTUW(b *strings.Builder, d *Deck) { if len(d.T) > 0 { sorted := make([]TagCard, len(d.T)) copy(sorted, d.T) // Two-level (decoded name, target hash) per §4.5.2 and §4.7.16, // not a comparison of the concatenated key: concatenating lets the // target's leading characters compete with the tail of a longer // name, which reorders tags whose names are prefixes of one // another into a run the parser rejects. tagOrderKey is the same // key the parser's check uses, so the two cannot drift. sort.Slice(sorted, func(i, j int) bool { if cmp := Compare(tagOrderKey(sorted[i]), tagOrderKey(sorted[j])); cmp != 0 { return cmp < 0 } return Compare(sorted[i].UUID, sorted[j].UUID) < 0 }) for _, tag := range sorted { // §4.7.16: the name and value are escape-encoded (a branch name // may hold a space, and the value carries that same name on a // `T *branch` card); the target token is written verbatim, never // encoded. fmt.Fprintf(b, "T %c%s %s", tag.Type, FossilEncode(tag.Name), tag.UUID) if tag.Value != "" || tag.valuePresent { // A present-but-empty value re-emits as a trailing space. fmt.Fprintf(b, " %s", FossilEncode(tag.Value)) } b.WriteString("\n") } } if d.U != nil && *d.U != "" { fmt.Fprintf(b, "U %s\n", FossilEncode(*d.U)) } // §4.7.18: a W card whose body the parser read -- even a zero-length one -- // is non-nil, so it re-emits as "W 0" + LF + LF; a wholly absent W card is // nil and writes nothing. if d.W != nil { fmt.Fprintf(b, "W %d\n%s\n", len(d.W), d.W) } } // marshalTimestamp renders a D- or E-card time in the writer form // YYYY-MM-DDTHH:MM:SS with a fractional second of at least minFracDigits // digits, extended to keep any finer precision the parser read and with // trailing zeros past minFracDigits removed. minFracDigits is 3 for the D // card and 0 for the E card, matching their writer forms. func marshalTimestamp(t time.Time, minFracDigits int) string { t = t.UTC() base := t.Format("2006-01-02T15:04:05") if frac := fractionDigits(t.Nanosecond(), minFracDigits); frac != "" { return base + "." + frac } return base } // fractionDigits formats nsec as decimal fraction digits, keeping at least // minDigits and trimming trailing zeros beyond that. func fractionDigits(nsec, minDigits int) string { digits := fmt.Sprintf("%09d", nsec) end := len(digits) for end > minDigits && digits[end-1] == '0' { end-- } return digits[:end] }