Files
silo-server/internal/markers/contribute.go
T
0163df3683 [codex] Add IntroDB marker integration and dialogue-aware Chromaprint refinement (#57)
* docs(markers): design + implementation plans for multi-source markers & TheIntroDB contribution

* fix(markers): TheIntroDB read-path correctness (TVDB, real confidence, best candidate)

Honor TVDB ids in /media lookups (previously dropped — anime/TheTVDB-first
libraries got no markers), decode and use the real per-segment confidence and
submission_count instead of a hardcoded 0.9, and pick the most-submitted /
highest-confidence candidate when several are returned. Adds httptest coverage
for the introdb client and provider.

Phase 1 of docs/superpowers/plans/2026-06-06-marker-sources-and-contribution-implementation.md

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(markers): multi-source dispatch, per-provider config, per-segment provenance

Add marker_provider_config (per-provider fetch enable/priority + contribute
gates, contribution off by default) and a cached ProviderConfigStore. Add
Registry.FetchMerged: query all fetch-enabled providers concurrently and keep
the best candidate per segment (submission_count, then confidence, then fetch
priority), stamping each winning marker with its provider/algorithm. Thread
per-segment provenance through MarkerUpdatePayload and scanner.MarkerUpdate
(additive SegmentProvenance overrides) so a merged result writes correct
per-segment provider/confidence/algorithm; the legacy shared columns keep a
summary. The lazy-playback path now uses FetchMerged. With only TheIntroDB
enabled, behavior is unchanged.

Phase 2 of docs/superpowers/plans/2026-06-06-marker-sources-and-contribution-implementation.md

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(markers): TheIntroDB submission client, contribution audit, service engine

Add a markers.Submitter capability and implement it on the introdb provider
(POST /v3/submit, GET /v3/user/stats; key required, usage-limit aware, applies
the null start/end conventions). Add the marker_contributions audit table and a
value-hash-keyed ContributionStore for idempotency. Add ContributionService:
resolves enabled submitter providers, gates eligibility (never re-submit
online-sourced markers; auto runs require contribute_auto_local + scanner-intro
above the per-provider confidence threshold), checks idempotency, submits, and
records. Wired in main.go; no trigger yet (admin API and task follow).

Phase 3 of docs/superpowers/plans/2026-06-06-marker-sources-and-contribution-implementation.md

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(api): admin marker editing, contribution, and provider config endpoints

Add the RequireAdmin marker API: GET/PUT /admin/files/{id}/markers (read with
provenance; manual upsert where a segment object sets and null clears),
DELETE .../markers/{segment}, POST .../contribute and GET .../contributions,
plus GET/PUT /admin/markers/providers[/{provider}] and a
.../validate key-check returning user stats. Manual writes go through the
priority-gated UpsertMarkers (source=manual) and notify live sessions; a new
FileRepository.ClearMarkers nulls a segment's columns. Validation mirrors the
contribution rules.

Phase 4 of docs/superpowers/plans/2026-06-06-marker-sources-and-contribution-implementation.md

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(markers): daily auto-contribution task for local intro markers

Add ContributeMarkersTask (daily 04:00, after local detection): when a provider
has contribute_enabled + contribute_auto_local, page through episode files with
a scanner intro marker at/above the provider's confidence threshold (new
ContributionStore.CandidateLocalIntroFiles keyset query) and run them through
ContributionService with Auto=true. No-op when no provider opts in; idempotent
and resumable across runs.

Phase 5 of docs/superpowers/plans/2026-06-06-marker-sources-and-contribution-implementation.md

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(intromarkers): refine chromaprint starts with dialogue cues

* feat(markers): finish marker management backend

* feat(web): add marker editing UI

* feat(markers): use plugin marker providers

* fix(markers): address PR review feedback

* feat(player): show marker labels on seek hover

* fix(markers): type nullable marker mutation params

* feat(markers): audit marker edits and add permission

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-06 22:29:26 -04:00

300 lines
9.0 KiB
Go

package markers
import (
"context"
"log/slog"
"strconv"
"strings"
"time"
"github.com/Silo-Server/silo-server/internal/models"
)
// OutcomeStatus values for a contribution attempt, in addition to the provider
// SubmissionStatus* values.
const (
OutcomeStatusError = "error"
OutcomeStatusRateLimited = "rate_limited"
OutcomeStatusSkipped = "skipped"
)
// ContributeOptions scopes a contribution run.
type ContributeOptions struct {
// Provider restricts the run to a single provider id ("" = all enabled).
Provider string
// Segments restricts the run to specific kinds (nil = every present segment).
Segments []MarkerKind
// Auto marks a background run: a segment is only contributed when the
// provider has contribute_auto_local enabled and the segment is a local
// (scanner) intro at or above the provider's confidence threshold.
Auto bool
}
// ContributionOutcome is the per (provider, segment) result of a run.
type ContributionOutcome struct {
Provider string
Segment MarkerKind
Status string // pending | accepted | rejected | error | rate_limited | skipped
SubmissionID string
Reason string // skip reason or error message
RetryAfter time.Duration
}
// providerConfigReader is the read surface ContributionService needs from the
// provider config store (satisfied by *ProviderConfigStore).
type providerConfigReader interface {
Get(provider string) (ProviderConfig, bool)
}
// contributionRecorder is the audit surface ContributionService needs
// (satisfied by *ContributionStore).
type contributionRecorder interface {
AlreadySubmitted(ctx context.Context, fileID int, provider, segmentKind, contentHash string) (bool, error)
Record(ctx context.Context, row ContributionRow) error
}
// ContributionService submits a file's eligible markers to enabled submitter
// providers, idempotently and audited. Both the on-demand admin path and the
// daily auto-submission task funnel through ContributeFile.
type ContributionService struct {
registry *Registry
resolver ExternalIDResolver
config providerConfigReader
store contributionRecorder
logger *slog.Logger
}
// NewContributionService constructs the service.
func NewContributionService(registry *Registry, resolver ExternalIDResolver, config providerConfigReader, store contributionRecorder, logger *slog.Logger) *ContributionService {
if logger == nil {
logger = slog.Default()
}
return &ContributionService{registry: registry, resolver: resolver, config: config, store: store, logger: logger}
}
type segmentData struct {
kind MarkerKind
name string
start *float64
end *float64
source *string
confidence *float64
}
func fileSegments(file *models.MediaFile) []segmentData {
return []segmentData{
{MarkerKindIntro, "intro", file.IntroStart, file.IntroEnd, file.IntroMarkersSource, file.IntroMarkersConfidence},
{MarkerKindCredits, "credits", file.CreditsStart, file.CreditsEnd, file.CreditsMarkersSource, file.CreditsMarkersConfidence},
{MarkerKindRecap, "recap", file.RecapStart, file.RecapEnd, file.RecapMarkersSource, file.RecapMarkersConfidence},
{MarkerKindPreview, "preview", file.PreviewStart, file.PreviewEnd, file.PreviewMarkersSource, file.PreviewMarkersConfidence},
}
}
func (s *ContributionService) submitters() []Submitter {
if s == nil || s.registry == nil {
return nil
}
var out []Submitter
for _, p := range s.registry.Providers() {
if sub, ok := p.(Submitter); ok {
out = append(out, sub)
}
}
return out
}
// ContributeFile submits the file's eligible markers and returns one outcome
// per (provider, segment) attempted (including explicit skips).
func (s *ContributionService) ContributeFile(ctx context.Context, file *models.MediaFile, opts ContributeOptions) ([]ContributionOutcome, error) {
if s == nil || file == nil || s.registry == nil || s.resolver == nil || s.config == nil || s.store == nil {
return nil, nil
}
ids, err := s.resolver.ResolveForFile(ctx, file)
if err != nil {
return nil, err
}
if !ids.HasAnyID() {
return nil, nil
}
want := func(k MarkerKind) bool {
if len(opts.Segments) == 0 {
return true
}
for _, kind := range opts.Segments {
if kind == k {
return true
}
}
return false
}
var outcomes []ContributionOutcome
for _, sub := range s.submitters() {
providerID := sub.ID()
if opts.Provider != "" && opts.Provider != providerID {
continue
}
cfg, ok := s.config.Get(providerID)
if !ok || !cfg.ContributeEnabled {
continue
}
for _, seg := range fileSegments(file) {
if !want(seg.kind) {
continue
}
if outcome, attempted := s.contributeSegment(ctx, sub, providerID, cfg, file, ids, seg, opts.Auto); attempted {
outcomes = append(outcomes, outcome)
if outcome.Status == OutcomeStatusRateLimited {
return outcomes, nil
}
}
}
}
return outcomes, nil
}
func (s *ContributionService) contributeSegment(
ctx context.Context,
sub Submitter,
providerID string,
cfg ProviderConfig,
file *models.MediaFile,
ids ExternalIDs,
seg segmentData,
auto bool,
) (ContributionOutcome, bool) {
if seg.start == nil || seg.end == nil {
return ContributionOutcome{}, false
}
source := ""
if seg.source != nil {
source = *seg.source
}
// Only ever contribute our own data; never round-trip online markers.
if source != models.MarkerSourceScanner && source != models.MarkerSourceManual {
return ContributionOutcome{}, false
}
if auto {
if !cfg.ContributeAutoLocal || source != models.MarkerSourceScanner || seg.kind != MarkerKindIntro {
return ContributionOutcome{}, false
}
confidence := 0.0
if seg.confidence != nil {
confidence = *seg.confidence
}
if confidence < cfg.ContributeMinConfidence {
return ContributionOutcome{}, false
}
}
if missing := missingRequiredExternalIDs(sub, ids); len(missing) > 0 {
return ContributionOutcome{
Provider: providerID,
Segment: seg.kind,
Status: OutcomeStatusSkipped,
Reason: strings.Join(missing, ",") + " id required",
}, true
}
startMs := int64(*seg.start * 1000)
endMs := int64(*seg.end * 1000)
durMs := int64(file.Duration) * 1000
hash := ContentHash(seg.name, &startMs, &endMs, &durMs, contributionTargetParts(ids)...)
already, err := s.store.AlreadySubmitted(ctx, file.ID, providerID, seg.name, hash)
if err != nil {
return ContributionOutcome{Provider: providerID, Segment: seg.kind, Status: OutcomeStatusError, Reason: err.Error()}, true
}
if already {
return ContributionOutcome{Provider: providerID, Segment: seg.kind, Status: OutcomeStatusSkipped, Reason: "already submitted"}, true
}
startDur := time.Duration(*seg.start * float64(time.Second))
endDur := time.Duration(*seg.end * float64(time.Second))
req := SubmissionRequest{
Kind: ids.Kind,
ExternalIDs: ids.AsRequestMap(),
SeasonNumber: ids.SeasonNumber,
EpisodeNumber: ids.EpisodeNumber,
Segment: seg.kind,
Start: &startDur,
End: &endDur,
Duration: time.Duration(file.Duration) * time.Second,
}
row := ContributionRow{
MediaFileID: file.ID,
Provider: providerID,
SegmentKind: seg.name,
Source: source,
SubmittedStartMs: &startMs,
SubmittedEndMs: &endMs,
VideoDurationMs: &durMs,
ContentHash: hash,
}
result, err := sub.SubmitMarker(ctx, req)
if err != nil {
msg := err.Error()
row.Status = OutcomeStatusError
row.Error = &msg
if recErr := s.store.Record(ctx, row); recErr != nil {
s.logger.Warn("record contribution error failed", "file_id", file.ID, "provider", providerID, "segment", seg.name, "error", recErr)
}
if after, ok := RetryAfter(err); ok {
return ContributionOutcome{Provider: providerID, Segment: seg.kind, Status: OutcomeStatusRateLimited, Reason: msg, RetryAfter: after}, true
}
return ContributionOutcome{Provider: providerID, Segment: seg.kind, Status: OutcomeStatusError, Reason: msg}, true
}
row.Status = result.Status
if row.Status == "" {
row.Status = SubmissionStatusPending
}
if result.ID != "" {
id := result.ID
row.SubmissionID = &id
}
if err := s.store.Record(ctx, row); err != nil {
s.logger.Warn("record contribution failed", "file_id", file.ID, "provider", providerID, "segment", seg.name, "error", err)
}
return ContributionOutcome{Provider: providerID, Segment: seg.kind, Status: row.Status, SubmissionID: result.ID}, true
}
func contributionTargetParts(ids ExternalIDs) []string {
return []string{
itemTypeName(ids.Kind),
ids.TmdbID,
ids.ImdbID,
ids.TvdbID,
intPart(ids.SeasonNumber),
intPart(ids.EpisodeNumber),
}
}
func intPart(v int) string {
if v == 0 {
return ""
}
return strconv.Itoa(v)
}
func missingRequiredExternalIDs(sub Submitter, ids ExternalIDs) []string {
reqProvider, ok := sub.(SubmissionRequirementProvider)
if !ok {
return nil
}
reqs := reqProvider.SubmissionRequirements()
if len(reqs.RequiredExternalIDs) == 0 {
return nil
}
values := ids.AsRequestMap()
var missing []string
for _, key := range reqs.RequiredExternalIDs {
if strings.TrimSpace(values[key]) == "" {
missing = append(missing, key)
}
}
return missing
}