mesh-delivery/modules/mesh-delivery-checker/internal/check/builds.go

package check

import (
	"context"
	"fmt"
	"maps"
	"slices"
	"sort"
	"strings"
	"time"

	"git.novox.be/novox/mesh-delivery/contracts/go/clients/meshdeliverycheckerclient"
	"git.novox.be/novox/mesh-delivery/contracts/go/seats/meshdeliverychecker"
	"git.novox.be/novox/mesh-delivery/modules/mesh-delivery-builder/builds"
)

// The states of a build the status line says (ADR 0484 clause 7).
const (
	BuildBuilt   = "built"
	BuildFailed  = "failed"
	BuildWaiting = "waiting"
)

// BuildStates reads the builder's build of a module at a fingerprint of the new form (builds.<module>.<fingerprint>): the
// value and whether the key exists.
type BuildStates interface {
	Build(ctx context.Context, module, fp2 string) (builds.BuilderBuild, bool, error)
}

// NoteOf is the state of a build as the line says it: done is built, failed is failed, and one queued, building or not
// begun is waiting.
func NoteOf(b builds.BuilderBuild, found bool) string {
	switch {
	case !found:
		return BuildWaiting
	case b.State == builds.BuildDone:
		return BuildBuilt
	case b.State == builds.BuildFailed:
		return BuildFailed
	}
	return BuildWaiting
}

// buildNotes are the state of each build the case's modules carry a fingerprint of, by module, sorted. A module two pull
// requests give is said once.
func (j *Examiner) buildNotes(ctx context.Context, c *Case) []BuildNote {
	if j.Builds == nil {
		return nil
	}
	fps := map[string]string{}
	for _, p := range c.Pulls {
		for module, cm := range p.Fps {
			if cm.Fp2 != "" {
				fps[module] = cm.Fp2
			}
		}
	}
	modules := make([]string, 0, len(fps))
	for m := range fps {
		modules = append(modules, m)
	}
	sort.Strings(modules)
	var notes []BuildNote
	for _, m := range modules {
		b, found, err := j.Builds.Build(ctx, m, fps[m])
		if err != nil {
			j.logf("the build of %s at %s cannot be read: %v", m, fps[m], err)
			found = false
		}
		notes = append(notes, BuildNote{Module: m, State: NoteOf(b, found)})
	}
	return notes
}

// DefaultBuildWait is how long a check waits for the candidate's builds to end before it composes (the setting
// build-wait-bound, novox/hq ADR 0603 §10): past it the check composes without the builds not ended and names them.
const DefaultBuildWait = 30 * time.Minute

// BuildEnds waits on the builder's build keys (builds.<module>.<fingerprint>), hearing each key's changes on the bus as the
// builder writes them: the builder owns the fact, and nothing is polled.
type BuildEnds interface {
	// Ended waits until the build of every module at its fingerprint (module to fingerprint) is done or failed, or ctx ends,
	// and answers the builds that ended, by module. An error is only builds that cannot be heard at all.
	Ended(ctx context.Context, builds map[string]string) (map[string]builds.BuilderBuild, error)
	// Done waits until the build of every module at its fingerprint is done (a failed one built again, builder_build_again),
	// and says whether they all are; false when ctx ends first or they cannot be heard.
	Done(ctx context.Context, builds map[string]string) bool
}

// BuildFailedTest is what a failure of the check names a failed build of the candidate by: the test `build: <module>`.
// A verdict failed on builds is asked again by the checker itself once each is built again (Checker.Rebuilt).
const BuildFailedTest = "build: "

// failedBuilds are the candidate's builds a failed verdict names as failed (BuildFailedTest), by module, at the fingerprint
// the case carries for it; none for a verdict that names no failed build.
func failedBuilds(v meshdeliverychecker.VerdictsByGroupHeadsValue, cs *Case) map[string]string {
	if v.State != meshdeliverycheckerclient.VerdictFail {
		return nil
	}
	out := map[string]string{}
	for _, f := range append([]FailedFirst{failedOf(v)}, failedList(v.Failures)...) {
		module, named := strings.CutPrefix(f.Test, BuildFailedTest)
		if !named {
			continue
		}
		for _, p := range cs.Pulls {
			if cm, ok := p.Fps[module]; ok && cm.Fp2 != "" {
				out[module] = cm.Fp2
			}
		}
	}
	if len(out) == 0 {
		return nil
	}
	return out
}

func (j *Examiner) buildWait() time.Duration {
	if j.BuildWait > 0 {
		return j.BuildWait
	}
	return DefaultBuildWait
}

// awaitBuilds waits for the candidate's builds before the check composes (ADR 0603 §10): the check builds what the builder
// builds of the candidate and composes with it. A failed build fails the check, naming the module and the builder's words;
// a build not ended within the bound is named, and the check composes without it. It answers the words of the composition's
// warnings: what the check composes without waiting on its build, and why.
func (st *state) awaitBuilds(ctx context.Context, v *meshdeliverychecker.VerdictsByGroupHeadsValue) []string {
	wait, words := st.buildsOfTheCheck(ctx)
	if len(wait) == 0 {
		return words
	}
	modules := slices.Sorted(maps.Keys(wait))
	if v.State == meshdeliverycheckerclient.VerdictFail {
		// The check has failed already: its red status is never held back by a build (the lead, 2026-11-03).
		return append(words, fmt.Sprintf("the check failed before its builds were waited on, so it composes at once, without waiting on the builds of %s (ADR 0603 §10)", strings.Join(modules, ", ")))
	}
	if st.j.BuildEnds == nil {
		return append(words, fmt.Sprintf("the builds of %s cannot be heard, so the check composes without waiting on them (ADR 0603 §10)", strings.Join(modules, ", ")))
	}
	bound := st.j.buildWait()
	waits := fmt.Sprintf("waits on the builder's builds of %s before it composes, at most %s (ADR 0603 §10)", strings.Join(modules, ", "), bound)
	st.c.note(waits)
	st.j.logf("%s %s", st.c.Key, waits)
	at, cancel := context.WithTimeout(ctx, bound)
	ended, err := st.j.BuildEnds.Ended(at, wait)
	cancel()
	if err != nil {
		st.j.logf("%s: the builds of %s cannot be heard: %v", st.c.Key, strings.Join(modules, ", "), err)
		return append(words, fmt.Sprintf("the builds of %s cannot be heard (%v), so the check composes without waiting on them (ADR 0603 §10)", strings.Join(modules, ", "), err))
	}
	var late []string
	for _, module := range modules {
		b, done := ended[module]
		switch {
		case !done:
			late = append(late, module)
		case b.State == builds.BuildFailed:
			f := FailedFirst{Test: BuildFailedTest + module, Words: fmt.Sprintf("the candidate's build at %s failed (rebuild with builder_build_again; the check asks itself again when the build is done): %s", wait[module], b.Error)}
			if v.State != meshdeliverycheckerclient.VerdictFail {
				v.State, v.Failed = meshdeliverycheckerclient.VerdictFail, &f
				continue
			}
			addFailure(v, f)
		}
	}
	if len(late) > 0 && ctx.Err() == nil {
		words = append(words, fmt.Sprintf("the builds of %s did not end within the build-wait-bound of %s, so the check composes without them (ADR 0603 §10)", strings.Join(late, ", "), bound))
	}
	return words
}