Files
multica/server/internal/handler/github.go
T
14c2e4e831 MUL-6243 feat(issue-status): board fetches by category; archive retires without migration (#7084)
* feat(issue-status): board fetches by category; archive retires without migration (MUL-6243)

Product decisions from the issue thread:

1. Board fetches by CATEGORY, not per status. PAGINATED_STATUSES becomes
   PAGINATED_CATEGORIES and the cache buckets by category, so the board's
   fan-out stays fixed at 7 parallel requests however many custom statuses a
   workspace defines — instead of one more request per status added. Adds
   status_category / status_categories filters to the issue list endpoints,
   backed by the existing issue_effective_status() SQL function.

3. Archiving retires a status from FUTURE use only. It no longer refuses while
   issues still carry it: those issues keep their status and keep resolving to
   their category's behavior (Effective ignores archived_at), while Resolve
   still rejects the status so nothing new can be assigned to it. The exclusive
   catalog lock stays — it is what makes "no NEW assignment after archive"
   exact rather than approximate.

Also lands the frontend catalog foundation: packages/core/issue-statuses with
useIssueStatuses(wsId) (React Query, keyed by wsId per the state rules) and a
pure buildIssueStatusCatalog for non-React callers, plus IssueStatusCategory as
a distinct type from the status key and status_category on issue payloads.

NOT included, and deliberately: widening IssueStatus to string. That change
forces every Record<IssueStatus, …> presentation lookup in packages/views to
resolve a key to its category first, and it is the first task of the follow-up
PR. Everything here is the groundwork that makes it mechanical.

Co-authored-by: multica-agent <[email protected]>

* fix(issue-status): restore the status index, make category authoritative, keep archived resolvable

Review round 1 on #7084 (MUL-6243):

1. Category filtering wrapped issue.status in issue_effective_status(), which
   makes the (workspace_id, status) index unusable — and the board switched to
   it unconditionally, so every workspace paid for it immediately. Categories
   are now expanded to concrete status keys (issuestatus.ExpandCategories) and
   filtered with an indexed status = ANY(...). Measured on a 100k-row table
   with the production index: seq scan cost 27013 -> index-only scan cost 6.07.
   A workspace with no custom statuses expands each category to exactly its own
   key, reproducing the pre-feature query. Archived keys are included, since
   issues left on them still belong in their category's column.

2. status_category was only filled for built-ins, so a custom-status issue
   arriving over the websocket was silently dropped by addIssueToBuckets and a
   cross-category update was skipped by the cache coordinator, drifting
   off-window totals permanently. List, grouped, get, create, update and batch
   responses now carry an authoritative category via a per-request Resolver
   (archived included), and the client invalidates instead of no-oping when a
   category is unresolvable.

3. The catalog was fetched without archived entries, so after archiving a
   status the issues still on it lost their name and colour and had their
   category guessed as todo — contradicting the decision that archiving leaves
   existing issues untouched. Resolution now includes archived entries;
   activeStatuses / inCategory expose only assignable ones to pickers.

Co-authored-by: multica-agent <[email protected]>

* fix(issue-status): category on every payload exit, one resolver per request, no same-category duplication

Review round 2 on #7084 (MUL-6243):

1. Category fill was incomplete and N+1. The create WEBSOCKET payload was
   unfilled — only the creator's HTTP response was, which is too late for the
   other tabs the event exists to serve. GetIssue, the open_only list, search,
   the duplicate-conflict response, the table rows, the GitHub PR update and
   the onboarding shim were also unfilled. Grouped and batch called the
   single-response helper per row, so each custom-status row hit the catalog
   separately.

   Every exit now goes through newStatusCategoryFiller, a request-scoped
   closure over ONE Resolver: built-in statuses still cost no query, and a page
   of custom-status rows costs one catalog read rather than one per row.

2. A status KEY change within one category duplicated the card and inflated the
   total. patchIssueInBuckets compared patch.status (a key) against loc.status
   (a bucket CATEGORY), so in_review -> a custom human_review took the
   cross-bucket path, then deleted from and re-inserted into the same bucket
   from a pre-delete snapshot. It now resolves the destination category first
   and compares category to category; same-category changes replace in place
   and leave the total alone.

   The destination category is resolved from the patch alone — the merged issue
   inherits the previous status_category, which would have kept a genuinely
   moved card in its old column. patchNeedsInvalidation lets the loaded branch
   of the cache coordinator invalidate instead of silently no-oping on a status
   it cannot resolve.

Also drops the unrelated share_link.sql.go churn from an sqlc version
difference, per the review nit.

The four cache cases were verified to fail against the previous logic (card
duplicated, total +1) before the fix was restored.

Co-authored-by: multica-agent <[email protected]>

* fix(issue-status): normalize status_category on every cache write; one filler per create

Review round 3 on #7084 (MUL-6243):

Must-fix — a cached entity kept the STALE status_category after a status
change. The server now sends a category on every issue, so a cached entity is
{status: "todo", status_category: "todo"}; an optimistic patch carries only
{status: "done"}, and a bare {...issue, ...patch} moved the card to the done
bucket while the entity still claimed todo. A single update self-heals from the
server response, but the batch API returns only {updated: n} and does not
refetch bucketed lists, so the inconsistency was permanent — and the next
off-window count then decremented the wrong bucket, recreating total drift.

normalizeStatusPatch now rewrites status_category to match status before any
cache write, applied at applyIssueChange's entry (covering the flat, table and
detail spreads), in patchIssueInBuckets, and in the children mutation. An
unresolvable custom key drops the inherited category rather than keeping it;
patchNeedsInvalidation still routes it to a refetch.

Test fixtures now carry status_category for built-ins, matching what the server
actually sends — without that the regression was invisible.

Non-blocking items also addressed:
- CreateIssue builds one filler and shares it between the broadcast payload and
  the HTTP response, so a custom-status create reads the catalog once.
- Background events via IssueToMap now resolve an authoritative category
  (IssueToMapWithCategory) for the task, issue and autopilot publishers. The
  channel router keeps plain IssueToMap and says why: it always creates with the
  built-in "todo", where key IS category.

The "one resolver serves a whole page" assertion was REMOVED rather than kept:
pg_stat counters lag, so a delta around one request passes even with a per-row
filler. Verified that directly before dropping it. The comment now states what
is and is not pinned instead of implying coverage that does not exist.

Co-authored-by: multica-agent <[email protected]>

* fix(issue-status): address review nits

- Normalize the single-update children optimistic patch too, matching the batch
  path. It self-healed on onSettled, but it broke the entity/bucket invariant
  for the duration of the request.
- Rename TestListEndpointsCarryStatusCategory: the old name claimed it covered
  "without per row queries", which the test explicitly does not assert.
- Drop the unrelated gofmt churn in empty_claim_cache.go.
- Add a background-event assertion: IssueToMap leaves a custom status
  uncategorized, IssueToMapWithCategory resolves it — the reason the publishers
  use the latter.

Co-authored-by: multica-agent <[email protected]>

---------

Co-authored-by: Bohan-J <[email protected]>
Co-authored-by: multica-agent <[email protected]>
2026-08-18 03:00:21 +08:00

1952 lines
75 KiB
Go

package handler
import (
"context"
"crypto/hmac"
"crypto/rand"
"crypto/sha256"
"encoding/hex"
"encoding/json"
"errors"
"fmt"
"io"
"log/slog"
"net/http"
"net/url"
"os"
"regexp"
"sort"
"strconv"
"strings"
"time"
"github.com/go-chi/chi/v5"
"github.com/golang-jwt/jwt/v5"
"github.com/jackc/pgx/v5"
"github.com/jackc/pgx/v5/pgtype"
"github.com/multica-ai/multica/server/internal/issuestatus"
"github.com/multica-ai/multica/server/internal/middleware"
db "github.com/multica-ai/multica/server/pkg/db/generated"
"github.com/multica-ai/multica/server/pkg/protocol"
)
// githubAPIBase is the base URL for GitHub's REST API. Mutable so tests can
// point App-authenticated calls at an httptest server without touching GitHub.
var githubAPIBase = "https://api.github.com"
const (
githubReturnToGitHub = "github"
githubReturnToRepositories = "repositories"
githubAPIResponseLimit = 4 << 20
)
// ── Response shapes ─────────────────────────────────────────────────────────
// GitHubInstallationResponse is the JSON shape returned by the installation
// list endpoint and broadcast on installation-related WS events.
//
// InstallationID is admin-only: the numeric GitHub installation_id is the
// management handle used by the Connect/Disconnect flows, so non-admin
// members receive responses with the field omitted. The list handler gates
// it by role; realtime broadcasts always omit it because the WS fanout has
// no per-recipient view (admins re-query the list endpoint on invalidation
// to recover the management handle).
type GitHubInstallationResponse struct {
ID string `json:"id"`
WorkspaceID string `json:"workspace_id"`
InstallationID *int64 `json:"installation_id,omitempty"`
AccountLogin string `json:"account_login"`
AccountType string `json:"account_type"`
AccountAvatarURL *string `json:"account_avatar_url"`
CreatedAt string `json:"created_at"`
}
type GitHubPullRequestResponse struct {
ID string `json:"id"`
// Provider is the Git provider this PR was mirrored from: "github", "forgejo",
// "gitea", or "gitlab". The frontend uses it to pick the host icon and
// label (e.g. GitLab "merge request").
Provider string `json:"provider"`
WorkspaceID string `json:"workspace_id"`
RepoOwner string `json:"repo_owner"`
RepoName string `json:"repo_name"`
Number int32 `json:"number"`
Title string `json:"title"`
State string `json:"state"`
HtmlURL string `json:"html_url"`
Branch *string `json:"branch"`
AuthorLogin *string `json:"author_login"`
AuthorAvatarURL *string `json:"author_avatar_url"`
MergedAt *string `json:"merged_at"`
ClosedAt *string `json:"closed_at"`
PRCreatedAt string `json:"pr_created_at"`
PRUpdatedAt string `json:"pr_updated_at"`
// Mergeable state mirrors GitHub's REST `mergeable_state` field, retained
// for compatibility. The card now reads the richer GraphQL fields below.
MergeableState *string `json:"mergeable_state"`
// ── GitHub API snapshot (MUL-5265, Plan C) ──────────────────────────────
// These come from an authenticated GraphQL query, the single source of
// truth. All are null / empty / 0 when no snapshot has landed (or the
// GitHub App private key is unconfigured), so the card hides the CI / merge
// region and degrades cleanly.
//
// Mergeable answers ONLY "is there a conflict": "mergeable" | "conflicting"
// | "unknown" | null. A false "conflicting" must never be reported as
// "not mergeable" — that verdict is MergeStateStatus's job.
Mergeable *string `json:"mergeable"`
// MergeStateStatus is GitHub's merge-state verdict, lowercased: "clean" |
// "dirty" | "blocked" | "behind" | "unstable" | "draft" | "has_hooks" |
// "unknown" | null. "Ready to merge" is derived ONLY from "clean".
MergeStateStatus *string `json:"merge_state_status"`
// SnapshotAvailable distinguishes a current API snapshot from both
// "feature disabled / not fetched yet" and "a current snapshot whose
// statusCheckRollup is null". Only the last case may render "no checks".
// It is omitted for non-GitHub providers, which keep their webhook-derived
// checks_conclusion compatibility path.
SnapshotAvailable *bool `json:"snapshot_available,omitempty"`
// ChecksRollup is GitHub's overall CI verdict, lowercased: "success" |
// "failure" | "pending" | "error" | "expected" | null. null means
// statusCheckRollup was null (no checks yet) and must NEVER render as
// passed.
ChecksRollup *string `json:"checks_rollup"`
// ChecksConclusion is a coarse compat alias derived from the snapshot:
// "passed" | "failed" | "pending" | null.
ChecksConclusion *string `json:"checks_conclusion"`
// Run-level counts for the PR's snapshot head. ChecksPending mirrors
// ChecksRunning for older clients that still read the old key.
ChecksTotal int64 `json:"checks_total"`
ChecksPassed int64 `json:"checks_passed"`
ChecksFailed int64 `json:"checks_failed"`
ChecksRunning int64 `json:"checks_running"`
ChecksPending int64 `json:"checks_pending"`
// FailedCheckNames names the failing checks so the card can point at them
// (e.g. "✗ 2/7 · backend, e2e").
FailedCheckNames []string `json:"failed_check_names"`
// SnapshotStale is true when an open PR's last successful fetch is older
// than the stale threshold (GitHub outage / revoked key): the card shows
// last-known data greyed out rather than blank.
SnapshotStale bool `json:"snapshot_stale"`
// SnapshotFetchedAt is when the snapshot was last fetched (RFC3339), or null.
SnapshotFetchedAt *string `json:"snapshot_fetched_at"`
// Diff stats (lines added/removed and file count) sourced from the
// `pull_request` webhook payload. Legacy rows that pre-date this
// field default to 0; the frontend treats total == 0 as "unknown"
// and hides the stats row.
Additions int32 `json:"additions"`
Deletions int32 `json:"deletions"`
ChangedFiles int32 `json:"changed_files"`
}
type GitHubConnectResponse struct {
URL string `json:"url"`
Configured bool `json:"configured"`
}
type GitHubRepositoryResponse struct {
ID int64 `json:"id"`
FullName string `json:"full_name"`
HTMLURL string `json:"html_url"`
CloneURL string `json:"clone_url"`
Description *string `json:"description"`
Private bool `json:"private"`
Archived bool `json:"archived"`
DefaultBranch string `json:"default_branch"`
}
type GitHubRepositoriesResponse struct {
Repositories []GitHubRepositoryResponse `json:"repositories"`
TotalCount int64 `json:"total_count"`
NextPage *int `json:"next_page"`
}
func githubInstallationToResponse(i db.GithubInstallation) GitHubInstallationResponse {
instID := i.InstallationID
return GitHubInstallationResponse{
ID: uuidToString(i.ID),
WorkspaceID: uuidToString(i.WorkspaceID),
InstallationID: &instID,
AccountLogin: i.AccountLogin,
AccountType: i.AccountType,
AccountAvatarURL: textToPtr(i.AccountAvatarUrl),
CreatedAt: timestampToString(i.CreatedAt),
}
}
// githubInstallationToBroadcast returns the same shape as the list endpoint's
// per-role response with the numeric `installation_id` stripped. Realtime
// events fan out to every WS client subscribed to the workspace, so the
// payload must match the weakest-role view — admin/owner clients re-query
// the list endpoint to recover the management handle. The frontend uses
// these events only to invalidate the installations query, so it does not
// read `installation_id` off the broadcast.
func githubInstallationToBroadcast(i db.GithubInstallation) GitHubInstallationResponse {
resp := githubInstallationToResponse(i)
resp.InstallationID = nil
return resp
}
func githubPullRequestToResponse(p db.GithubPullRequest, snapshotEnabled bool) GitHubPullRequestResponse {
snapshotAvailable := currentGitHubSnapshotAvailable(
snapshotEnabled, p.HeadSha, p.SnapshotHeadSha, p.SnapshotFetchedAt,
)
return GitHubPullRequestResponse{
ID: uuidToString(p.ID),
Provider: "github",
WorkspaceID: uuidToString(p.WorkspaceID),
RepoOwner: p.RepoOwner,
RepoName: p.RepoName,
Number: p.PrNumber,
Title: p.Title,
State: p.State,
HtmlURL: p.HtmlUrl,
Branch: textToPtr(p.Branch),
AuthorLogin: textToPtr(p.AuthorLogin),
AuthorAvatarURL: textToPtr(p.AuthorAvatarUrl),
MergedAt: timestampToPtr(p.MergedAt),
ClosedAt: timestampToPtr(p.ClosedAt),
PRCreatedAt: timestampToString(p.PrCreatedAt),
PRUpdatedAt: timestampToString(p.PrUpdatedAt),
MergeableState: textToPtr(p.MergeableState),
SnapshotAvailable: &snapshotAvailable,
// A bare PR row has no aggregated check counts — webhook broadcasts of a
// single PR fall through here and the frontend re-queries the list for
// the full snapshot (mergeable / rollup / counts).
ChecksConclusion: nil,
FailedCheckNames: []string{},
Additions: p.Additions,
Deletions: p.Deletions,
ChangedFiles: p.ChangedFiles,
}
}
// prSnapshotStaleThreshold is how old an open PR's last successful fetch may be
// before the card greys it out as stale. Healthy pipelines refresh open PRs at
// least every sweep interval (~10m), so crossing 30m means refreshes are not
// landing (GitHub outage, revoked key) and the shown data is last-known.
const prSnapshotStaleThreshold = 30 * time.Minute
func issuePullRequestRowToResponse(p db.ListPullRequestsByIssueRow, snapshotEnabled bool) GitHubPullRequestResponse {
snapshotAvailable := currentGitHubSnapshotAvailable(
snapshotEnabled, p.HeadSha, p.SnapshotHeadSha, p.SnapshotFetchedAt,
)
stale := false
if snapshotAvailable && (p.State == "open" || p.State == "draft") {
stale = time.Since(p.SnapshotFetchedAt.Time) > prSnapshotStaleThreshold
}
failedNames := p.FailedCheckNames
if failedNames == nil {
failedNames = []string{}
}
resp := GitHubPullRequestResponse{
ID: uuidToString(p.ID),
Provider: "github",
WorkspaceID: uuidToString(p.WorkspaceID),
RepoOwner: p.RepoOwner,
RepoName: p.RepoName,
Number: p.PrNumber,
Title: p.Title,
State: p.State,
HtmlURL: p.HtmlUrl,
Branch: textToPtr(p.Branch),
AuthorLogin: textToPtr(p.AuthorLogin),
AuthorAvatarURL: textToPtr(p.AuthorAvatarUrl),
MergedAt: timestampToPtr(p.MergedAt),
ClosedAt: timestampToPtr(p.ClosedAt),
PRCreatedAt: timestampToString(p.PrCreatedAt),
PRUpdatedAt: timestampToString(p.PrUpdatedAt),
MergeableState: textToPtr(p.MergeableState),
SnapshotAvailable: &snapshotAvailable,
FailedCheckNames: []string{},
SnapshotStale: stale,
Additions: p.Additions,
Deletions: p.Deletions,
ChangedFiles: p.ChangedFiles,
}
if snapshotAvailable {
resp.Mergeable = lowerTextPtr(p.ApiMergeable)
resp.MergeStateStatus = lowerTextPtr(p.ApiMergeStateStatus)
resp.ChecksRollup = lowerTextPtr(p.ChecksRollupState)
resp.ChecksConclusion = rollupToConclusion(p.ChecksRollupState, p.ChecksFailed, p.ChecksRunning, p.ChecksPassed)
resp.ChecksTotal = p.ChecksTotal
resp.ChecksPassed = p.ChecksPassed
resp.ChecksFailed = p.ChecksFailed
resp.ChecksRunning = p.ChecksRunning
resp.ChecksPending = p.ChecksRunning
resp.FailedCheckNames = failedNames
resp.SnapshotFetchedAt = timestampToPtr(p.SnapshotFetchedAt)
}
return resp
}
func currentGitHubSnapshotAvailable(
enabled bool,
headSHA string,
snapshotHeadSHA string,
fetchedAt pgtype.Timestamptz,
) bool {
return enabled && fetchedAt.Valid && snapshotHeadSHA != "" && snapshotHeadSHA == headSHA
}
// aggregateChecksConclusion collapses per-PR commit-status counts into a
// single coarse status. Still used by the self-hosted VCS provider path
// (Forgejo / Gitea / GitLab), which mirrors commit statuses via webhook rather
// than fetching a GitHub-style API snapshot:
// - any failed status wins ("failed");
// - any not-yet-completed status makes the PR "pending";
// - all completed and passed is "passed";
// - no observed status at all is nil (rendered as "no checks" / hidden).
func aggregateChecksConclusion(failed, passed, pending, total int64) *string {
if total == 0 {
return nil
}
var v string
switch {
case failed > 0:
v = "failed"
case pending > 0:
v = "pending"
case passed > 0:
v = "passed"
default:
return nil
}
return &v
}
// lowerTextPtr returns a lowercased *string for a non-empty pgtype.Text, else
// nil. Used to expose GraphQL enums (MERGEABLE / CLEAN / SUCCESS …) to the API
// in the project's lowercase convention.
func lowerTextPtr(t pgtype.Text) *string {
if !t.Valid || t.String == "" {
return nil
}
v := strings.ToLower(t.String)
return &v
}
// rollupToConclusion derives the coarse compat "checks_conclusion" from the
// GraphQL rollup, falling back to the run counts when the rollup enum is
// unfamiliar. A null/empty rollup means "no checks yet" → nil (never "passed").
func rollupToConclusion(rollup pgtype.Text, failed, running, passed int64) *string {
if !rollup.Valid || rollup.String == "" {
return nil
}
var v string
switch strings.ToUpper(rollup.String) {
case "FAILURE", "ERROR":
v = "failed"
case "PENDING", "EXPECTED":
v = "pending"
case "SUCCESS":
v = "passed"
default:
switch {
case failed > 0:
v = "failed"
case running > 0:
v = "pending"
case passed > 0:
v = "passed"
default:
return nil
}
}
return &v
}
// ── Connect / state token ───────────────────────────────────────────────────
// githubAppSlug returns the GitHub App slug used to build the install URL.
// Empty when the integration is not configured for this deployment.
func githubAppSlug() string { return strings.TrimSpace(os.Getenv("GITHUB_APP_SLUG")) }
// githubWebhookSecret is shared by webhook verification and state-token signing.
// We reuse the webhook secret as the state HMAC key so operators only need to
// configure one value.
func githubWebhookSecret() string { return strings.TrimSpace(os.Getenv("GITHUB_WEBHOOK_SECRET")) }
// isGitHubConfigured returns true only when BOTH the install slug and the
// webhook secret are set. The Connect button uses this single flag, so the
// frontend never offers a flow that the backend would reject.
func isGitHubConfigured() bool { return githubAppSlug() != "" && githubWebhookSecret() != "" }
// isGitHubRepositoryBrowseConfigured is deliberately separate from the
// install-flow flag. The App slug + webhook secret are enough to connect an
// installation, but browsing its repositories also requires App JWT
// credentials so the server can mint a short-lived installation token.
func isGitHubRepositoryBrowseConfigured() bool {
return strings.TrimSpace(os.Getenv("GITHUB_APP_ID")) != "" &&
strings.TrimSpace(os.Getenv("GITHUB_APP_PRIVATE_KEY")) != ""
}
// signState produces an opaque token that binds a workspace ID to the
// install flow so the setup callback can recover the workspace without
// trusting query params alone. Format: "<workspaceID>.<nonce>.<sigHex>".
func signState(workspaceID string) (string, error) {
return signStateForReturn(workspaceID, githubReturnToGitHub)
}
func signStateForReturn(workspaceID, returnTo string) (string, error) {
secret := githubWebhookSecret()
if secret == "" {
return "", errors.New("github integration is not configured")
}
if !isAllowedGitHubReturnTo(returnTo) {
return "", errors.New("invalid github return target")
}
nonceBytes := make([]byte, 12)
if _, err := rand.Read(nonceBytes); err != nil {
return "", err
}
nonce := hex.EncodeToString(nonceBytes)
payload := workspaceID + "." + nonce
if returnTo != githubReturnToGitHub {
payload = workspaceID + "." + returnTo + "." + nonce
}
mac := hmac.New(sha256.New, []byte(secret))
mac.Write([]byte(payload))
sig := hex.EncodeToString(mac.Sum(nil))
return payload + "." + sig, nil
}
func verifyState(token string) (string, bool) {
workspaceID, _, ok := verifyStateWithReturn(token)
return workspaceID, ok
}
func verifyStateWithReturn(token string) (workspaceID, returnTo string, ok bool) {
secret := githubWebhookSecret()
if secret == "" {
return "", "", false
}
parts := strings.Split(token, ".")
if len(parts) != 3 && len(parts) != 4 {
return "", "", false
}
workspaceID = parts[0]
returnTo = githubReturnToGitHub
nonceIndex := 1
if len(parts) == 4 {
returnTo = parts[1]
nonceIndex = 2
if !isAllowedGitHubReturnTo(returnTo) {
return "", "", false
}
}
sig := parts[nonceIndex+1]
payload := strings.Join(parts[:nonceIndex+1], ".")
mac := hmac.New(sha256.New, []byte(secret))
mac.Write([]byte(payload))
expected := hex.EncodeToString(mac.Sum(nil))
if !hmac.Equal([]byte(expected), []byte(sig)) {
return "", "", false
}
return workspaceID, returnTo, true
}
func isAllowedGitHubReturnTo(returnTo string) bool {
return returnTo == githubReturnToGitHub || returnTo == githubReturnToRepositories
}
func githubSettingsURL(frontend, returnTo string) string {
if !isAllowedGitHubReturnTo(returnTo) {
returnTo = githubReturnToGitHub
}
return strings.TrimRight(frontend, "/") + "/settings?tab=" + url.QueryEscape(returnTo)
}
// GitHubConnect (GET /api/workspaces/{id}/github/connect) returns the URL the
// browser should open to install the Multica GitHub App against the caller's
// repos. The state token binds the resulting setup callback to this workspace.
func (h *Handler) GitHubConnect(w http.ResponseWriter, r *http.Request) {
workspaceID := chi.URLParam(r, "id")
if _, ok := parseUUIDOrBadRequest(w, workspaceID, "workspace id"); !ok {
return
}
if !isGitHubConfigured() {
writeJSON(w, http.StatusOK, GitHubConnectResponse{Configured: false})
return
}
returnTo := strings.TrimSpace(r.URL.Query().Get("return_to"))
if returnTo == "" {
returnTo = githubReturnToGitHub
}
if !isAllowedGitHubReturnTo(returnTo) {
writeError(w, http.StatusBadRequest, "invalid return target")
return
}
slug := githubAppSlug()
state, err := signStateForReturn(workspaceID, returnTo)
if err != nil {
writeError(w, http.StatusInternalServerError, "failed to sign state")
return
}
installURL := fmt.Sprintf(
"https://github.com/apps/%s/installations/new?state=%s",
url.PathEscape(slug),
url.QueryEscape(state),
)
writeJSON(w, http.StatusOK, GitHubConnectResponse{URL: installURL, Configured: true})
}
// GitHubSetupCallback (GET /api/github/setup) handles the redirect GitHub
// sends after a user installs (or re-authorizes) the App. We expect
// ?installation_id=<id>&state=<signed token>. We persist the installation
// row (workspace ↔ installation_id mapping), then bounce the user back to
// the new Settings → GitHub tab in the web app (RFC MUL-2414 §4.1). The
// previous destination was the catch-all Settings page, which after the
// GitHub-tab split would land users on the default profile tab instead of
// the place that shows the connection they just completed.
func (h *Handler) GitHubSetupCallback(w http.ResponseWriter, r *http.Request) {
q := r.URL.Query()
installationIDStr := q.Get("installation_id")
state := q.Get("state")
frontend := strings.TrimSpace(os.Getenv("FRONTEND_ORIGIN"))
if frontend == "" {
frontend = "http://localhost:3000"
}
settingsURL := githubSettingsURL(frontend, githubReturnToGitHub)
if state == "" {
http.Redirect(w, r, settingsURL+"&github_error=missing_params", http.StatusFound)
return
}
workspaceID, returnTo, ok := verifyStateWithReturn(state)
if !ok {
http.Redirect(w, r, settingsURL+"&github_error=invalid_state", http.StatusFound)
return
}
settingsURL = githubSettingsURL(frontend, returnTo)
if installationIDStr == "" {
http.Redirect(w, r, settingsURL+"&github_error=missing_params", http.StatusFound)
return
}
installationID, err := strconv.ParseInt(installationIDStr, 10, 64)
if err != nil {
http.Redirect(w, r, settingsURL+"&github_error=bad_installation_id", http.StatusFound)
return
}
wsUUID, err := parseStrictUUID(workspaceID)
if err != nil {
http.Redirect(w, r, settingsURL+"&github_error=bad_workspace", http.StatusFound)
return
}
// Resolve the installation against GitHub's API to capture display info.
// If the App auth is not configured we still create the row with the
// minimum we know; webhook events will refresh it as soon as one fires.
login, accountType, avatar := fetchInstallationAccount(r.Context(), installationID)
// Best-effort capture of the connecting user (may be nil if the public
// callback was hit without a session — e.g. user wasn't logged in to
// Multica when they finished the GitHub install). Either way we save
// the row so the workspace owner sees the connection on next reload.
connectedBy := pgtype.UUID{}
if userID := requestUserID(r); userID != "" {
if u, err := parseStrictUUID(userID); err == nil {
connectedBy = u
}
}
inst, err := h.Queries.CreateGitHubInstallation(r.Context(), db.CreateGitHubInstallationParams{
WorkspaceID: wsUUID,
InstallationID: installationID,
AccountLogin: login,
AccountType: accountType,
AccountAvatarUrl: ptrToText(avatar),
ConnectedByID: connectedBy,
})
if err != nil {
slog.Error("github: failed to persist installation", "err", err, "installation_id", installationID)
http.Redirect(w, r, settingsURL+"&github_error=persist_failed", http.StatusFound)
return
}
inst, err = h.consumePendingGitHubInstallation(r.Context(), inst)
if err != nil {
slog.Error("github: failed to apply pending installation metadata", "err", err, "installation_id", installationID)
http.Redirect(w, r, settingsURL+"&github_error=persist_failed", http.StatusFound)
return
}
h.publish(protocol.EventGitHubInstallationCreated, workspaceID, "system", "", map[string]any{
"installation": githubInstallationToBroadcast(inst),
})
http.Redirect(w, r, settingsURL+"&github_connected=1", http.StatusFound)
}
func (h *Handler) consumePendingGitHubInstallation(ctx context.Context, inst db.GithubInstallation) (db.GithubInstallation, error) {
pending, err := h.Queries.GetPendingGitHubInstallation(ctx, inst.InstallationID)
if err != nil {
if errors.Is(err, pgx.ErrNoRows) {
return inst, nil
}
return inst, err
}
refreshed, err := h.Queries.CreateGitHubInstallation(ctx, db.CreateGitHubInstallationParams{
WorkspaceID: inst.WorkspaceID,
InstallationID: inst.InstallationID,
AccountLogin: pending.AccountLogin,
AccountType: coalesce(pending.AccountType, "User"),
AccountAvatarUrl: pending.AccountAvatarUrl,
ConnectedByID: inst.ConnectedByID,
})
if err != nil {
return inst, err
}
if err := h.Queries.DeletePendingGitHubInstallation(ctx, inst.InstallationID); err != nil {
return inst, err
}
return refreshed, nil
}
// fetchInstallationAccount tries to enrich the installation row with the
// account name + avatar from GitHub.
//
// GitHub's `GET /app/installations/{id}` endpoint requires GitHub App
// authentication (a JWT signed with the App's RSA private key). When the
// operator has configured GITHUB_APP_ID and GITHUB_APP_PRIVATE_KEY, we
// sign a short-lived JWT and use it; on any failure (env not set, key
// malformed, GitHub returns non-200) we fall back to the "unknown"
// placeholder. The next `installation` webhook delivery from GitHub will
// upsert the row with the real account info — see handleInstallationEvent.
//
// The HTTP call is synchronous (no independent timeout — that's a pre-
// existing wart of the install path), but we deliberately do NOT let a
// failure abort the setup callback: a network blip here just leaves the
// "unknown" placeholder in place, and the frontend re-queries on the
// realtime broadcast emitted by the webhook handler, so the UI converges
// without a manual refresh.
func fetchInstallationAccount(ctx context.Context, installationID int64) (login, accountType string, avatar *string) {
login = "unknown"
accountType = "User"
avatar = nil
endpoint := fmt.Sprintf("%s/app/installations/%d", strings.TrimRight(githubAPIBase, "/"), installationID)
req, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint, nil)
if err != nil {
return
}
req.Header.Set("Accept", "application/vnd.github+json")
if token, err := signGitHubAppJWT(time.Now()); err != nil {
// Misconfigured private key is operator-actionable — log so the
// install path doesn't silently fall back to "unknown" forever
// without leaving a breadcrumb.
slog.Warn("github: sign App JWT failed", "err", err)
} else if token != "" {
req.Header.Set("Authorization", "Bearer "+token)
}
resp, err := http.DefaultClient.Do(req)
if err != nil {
return
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
return
}
var body struct {
Account struct {
Login string `json:"login"`
Type string `json:"type"`
AvatarURL string `json:"avatar_url"`
} `json:"account"`
}
if err := json.NewDecoder(resp.Body).Decode(&body); err != nil {
return
}
if body.Account.Login != "" {
login = body.Account.Login
}
if body.Account.Type != "" {
accountType = body.Account.Type
}
if body.Account.AvatarURL != "" {
v := body.Account.AvatarURL
avatar = &v
}
return
}
// signGitHubAppJWT mints the short-lived RS256 JWT GitHub requires for
// App-authenticated REST calls (see fetchInstallationAccount). Returns
// ("", nil) when the operator hasn't configured the App identity — that's
// a soft "App auth not available" signal, not an error, so callers can
// fall through to their unauthenticated path. A malformed
// GITHUB_APP_PRIVATE_KEY surfaces as an error so the operator notices.
//
// `now` is injected for deterministic tests; production callers pass
// time.Now().
func signGitHubAppJWT(now time.Time) (string, error) {
appID := strings.TrimSpace(os.Getenv("GITHUB_APP_ID"))
pemKey := strings.TrimSpace(os.Getenv("GITHUB_APP_PRIVATE_KEY"))
if appID == "" || pemKey == "" {
return "", nil
}
key, err := jwt.ParseRSAPrivateKeyFromPEM([]byte(pemKey))
if err != nil {
return "", fmt.Errorf("parse GITHUB_APP_PRIVATE_KEY: %w", err)
}
// GitHub allows JWTs valid for up to 10 minutes. We back-date `iat`
// by 60 seconds to absorb modest clock skew between us and GitHub
// (otherwise an "iat in the future" verdict from GitHub fails the
// request) and cap `exp` at 9 minutes ahead to stay inside the cap
// even with the same skew applied.
claims := jwt.MapClaims{
"iat": now.Add(-60 * time.Second).Unix(),
"exp": now.Add(9 * time.Minute).Unix(),
"iss": appID,
}
token := jwt.NewWithClaims(jwt.SigningMethodRS256, claims)
signed, err := token.SignedString(key)
if err != nil {
return "", fmt.Errorf("sign App JWT: %w", err)
}
return signed, nil
}
// ── Listing / disconnect ────────────────────────────────────────────────────
// ListGitHubInstallations returns the workspace's connected GitHub
// installations to any workspace member. Connect/disconnect remain
// admin-only at the router level, so the response carries a `can_manage`
// hint and strips the numeric `installation_id` for non-admin callers —
// they get visibility into "is GitHub wired up, and by whom?" without the
// management handle.
func (h *Handler) ListGitHubInstallations(w http.ResponseWriter, r *http.Request) {
workspaceID := chi.URLParam(r, "id")
wsUUID, ok := parseUUIDOrBadRequest(w, workspaceID, "workspace id")
if !ok {
return
}
member, _ := middleware.MemberFromContext(r.Context())
canManage := roleAllowed(member.Role, "owner", "admin")
rows, err := h.Queries.ListGitHubInstallationsByWorkspace(r.Context(), wsUUID)
if err != nil {
writeError(w, http.StatusInternalServerError, "failed to list installations")
return
}
out := make([]GitHubInstallationResponse, 0, len(rows))
for _, row := range rows {
resp := githubInstallationToResponse(row)
if !canManage {
resp.InstallationID = nil
}
out = append(out, resp)
}
writeJSON(w, http.StatusOK, map[string]any{
"installations": out,
"configured": isGitHubConfigured(),
"repository_browse_configured": isGitHubRepositoryBrowseConfigured(),
"can_manage": canManage,
})
}
// ListGitHubInstallationRepositories returns the repositories accessible to a
// workspace-bound GitHub App installation. The route is admin-only because
// private repository names are sensitive. The path takes our installation row
// UUID (not GitHub's numeric installation id), and the workspace ownership
// check happens before any GitHub API call.
func (h *Handler) ListGitHubInstallationRepositories(w http.ResponseWriter, r *http.Request) {
workspaceID := chi.URLParam(r, "id")
if _, ok := parseUUIDOrBadRequest(w, workspaceID, "workspace id"); !ok {
return
}
installationRowID := chi.URLParam(r, "installationId")
rowUUID, ok := parseUUIDOrBadRequest(w, installationRowID, "installation id")
if !ok {
return
}
row, err := h.Queries.GetGitHubInstallationByID(r.Context(), rowUUID)
if err != nil {
if errors.Is(err, pgx.ErrNoRows) {
writeError(w, http.StatusNotFound, "github installation not found")
return
}
writeError(w, http.StatusInternalServerError, "failed to load github installation")
return
}
if uuidToString(row.WorkspaceID) != workspaceID {
writeError(w, http.StatusNotFound, "github installation not found")
return
}
if !isGitHubRepositoryBrowseConfigured() {
writeError(w, http.StatusServiceUnavailable, "github repository browsing is not configured")
return
}
page, ok := parseGitHubPageParam(w, r, "page", 1, 1, 100000)
if !ok {
return
}
perPage, ok := parseGitHubPageParam(w, r, "per_page", 100, 1, 100)
if !ok {
return
}
repositories, err := fetchGitHubInstallationRepositories(
r.Context(),
row.InstallationID,
page,
perPage,
)
if err != nil {
slog.Warn("github: list installation repositories failed", "err", err)
writeError(w, http.StatusBadGateway, "failed to list github repositories")
return
}
writeJSON(w, http.StatusOK, repositories)
}
func parseGitHubPageParam(
w http.ResponseWriter,
r *http.Request,
name string,
defaultValue, minValue, maxValue int,
) (int, bool) {
raw := strings.TrimSpace(r.URL.Query().Get(name))
if raw == "" {
return defaultValue, true
}
value, err := strconv.Atoi(raw)
if err != nil || value < minValue || value > maxValue {
writeError(w, http.StatusBadRequest, "invalid "+name)
return 0, false
}
return value, true
}
func fetchGitHubInstallationRepositories(
ctx context.Context,
installationID int64,
page, perPage int,
) (GitHubRepositoriesResponse, error) {
appJWT, err := signGitHubAppJWT(time.Now())
if err != nil {
return GitHubRepositoriesResponse{}, err
}
if appJWT == "" {
return GitHubRepositoriesResponse{}, errors.New("github App JWT credentials unavailable")
}
client := &http.Client{Timeout: 15 * time.Second}
tokenEndpoint := fmt.Sprintf(
"%s/app/installations/%d/access_tokens",
strings.TrimRight(githubAPIBase, "/"),
installationID,
)
tokenReq, err := http.NewRequestWithContext(
ctx,
http.MethodPost,
tokenEndpoint,
strings.NewReader(`{"permissions":{"metadata":"read"}}`),
)
if err != nil {
return GitHubRepositoriesResponse{}, err
}
setGitHubAPIHeaders(tokenReq, appJWT)
tokenReq.Header.Set("Content-Type", "application/json")
tokenResp, err := client.Do(tokenReq)
if err != nil {
return GitHubRepositoriesResponse{}, fmt.Errorf("create installation token: %w", err)
}
defer tokenResp.Body.Close()
if tokenResp.StatusCode != http.StatusCreated {
_, _ = io.Copy(io.Discard, io.LimitReader(tokenResp.Body, githubAPIResponseLimit))
return GitHubRepositoriesResponse{}, fmt.Errorf("create installation token: github status %d", tokenResp.StatusCode)
}
var tokenBody struct {
Token string `json:"token"`
}
if err := json.NewDecoder(io.LimitReader(tokenResp.Body, githubAPIResponseLimit)).Decode(&tokenBody); err != nil {
return GitHubRepositoriesResponse{}, fmt.Errorf("decode installation token: %w", err)
}
if tokenBody.Token == "" {
return GitHubRepositoriesResponse{}, errors.New("github returned an empty installation token")
}
defer revokeGitHubInstallationToken(client, tokenBody.Token)
repositoriesEndpoint := fmt.Sprintf(
"%s/installation/repositories?page=%d&per_page=%d",
strings.TrimRight(githubAPIBase, "/"),
page,
perPage,
)
repositoriesReq, err := http.NewRequestWithContext(ctx, http.MethodGet, repositoriesEndpoint, nil)
if err != nil {
return GitHubRepositoriesResponse{}, err
}
setGitHubAPIHeaders(repositoriesReq, tokenBody.Token)
repositoriesResp, err := client.Do(repositoriesReq)
if err != nil {
return GitHubRepositoriesResponse{}, fmt.Errorf("list installation repositories: %w", err)
}
defer repositoriesResp.Body.Close()
if repositoriesResp.StatusCode != http.StatusOK {
_, _ = io.Copy(io.Discard, io.LimitReader(repositoriesResp.Body, githubAPIResponseLimit))
return GitHubRepositoriesResponse{}, fmt.Errorf("list installation repositories: github status %d", repositoriesResp.StatusCode)
}
var body struct {
TotalCount int64 `json:"total_count"`
Repositories []struct {
ID int64 `json:"id"`
FullName string `json:"full_name"`
HTMLURL string `json:"html_url"`
CloneURL string `json:"clone_url"`
Description *string `json:"description"`
Private bool `json:"private"`
Archived bool `json:"archived"`
DefaultBranch string `json:"default_branch"`
} `json:"repositories"`
}
if err := json.NewDecoder(io.LimitReader(repositoriesResp.Body, githubAPIResponseLimit)).Decode(&body); err != nil {
return GitHubRepositoriesResponse{}, fmt.Errorf("decode installation repositories: %w", err)
}
out := GitHubRepositoriesResponse{
Repositories: make([]GitHubRepositoryResponse, 0, len(body.Repositories)),
TotalCount: body.TotalCount,
}
for _, repository := range body.Repositories {
out.Repositories = append(out.Repositories, GitHubRepositoryResponse(repository))
}
if int64(page*perPage) < body.TotalCount {
nextPage := page + 1
out.NextPage = &nextPage
}
return out, nil
}
func setGitHubAPIHeaders(req *http.Request, token string) {
req.Header.Set("Accept", "application/vnd.github+json")
req.Header.Set("Authorization", "Bearer "+token)
req.Header.Set("X-GitHub-Api-Version", "2022-11-28")
}
func revokeGitHubInstallationToken(client *http.Client, token string) {
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
endpoint := strings.TrimRight(githubAPIBase, "/") + "/installation/token"
req, err := http.NewRequestWithContext(ctx, http.MethodDelete, endpoint, nil)
if err != nil {
return
}
setGitHubAPIHeaders(req, token)
resp, err := client.Do(req)
if err != nil {
return
}
defer resp.Body.Close()
_, _ = io.Copy(io.Discard, io.LimitReader(resp.Body, githubAPIResponseLimit))
}
func (h *Handler) DeleteGitHubInstallation(w http.ResponseWriter, r *http.Request) {
workspaceID := chi.URLParam(r, "id")
wsUUID, ok := parseUUIDOrBadRequest(w, workspaceID, "workspace id")
if !ok {
return
}
id := chi.URLParam(r, "installationId")
idUUID, ok := parseUUIDOrBadRequest(w, id, "installation id")
if !ok {
return
}
if err := h.Queries.DeleteGitHubInstallation(r.Context(), db.DeleteGitHubInstallationParams{
ID: idUUID,
WorkspaceID: wsUUID,
}); err != nil {
writeError(w, http.StatusInternalServerError, "failed to remove installation")
return
}
h.publish(protocol.EventGitHubInstallationDeleted, workspaceID, "system", "", map[string]any{
"id": id,
})
w.WriteHeader(http.StatusNoContent)
}
// ── List PRs for an issue ───────────────────────────────────────────────────
func (h *Handler) ListPullRequestsForIssue(w http.ResponseWriter, r *http.Request) {
issue, ok := h.loadIssueForUser(w, r, chi.URLParam(r, "id"))
if !ok {
return
}
rows, err := h.Queries.ListPullRequestsByIssue(r.Context(), issue.ID)
if err != nil {
writeError(w, http.StatusInternalServerError, "failed to list pull requests")
return
}
out := make([]GitHubPullRequestResponse, 0, len(rows))
for _, row := range rows {
out = append(out, issuePullRequestRowToResponse(row, h.PRRefresh.Enabled()))
// Page-visit trigger (MUL-5265): if this card's snapshot is missing or
// older than the view TTL, kick an async refresh. Non-blocking — the
// current (possibly stale) response is returned immediately and the
// fresh snapshot arrives via the pull_request:updated realtime event.
h.PRRefresh.MaybeEnqueueOnView(
row.InstallationID, row.RepoOwner, row.RepoName, row.PrNumber,
row.SnapshotFetchedAt.Time,
row.SnapshotFetchedAt.Valid &&
row.SnapshotHeadSha != "" &&
row.SnapshotHeadSha == row.HeadSha,
)
}
// PRs from token-based providers (Forgejo / Gitea / GitLab) share the same
// card list. They live in their own provider-tagged tables, so they merge
// in here mapped to the same response shape; the combined list is re-sorted
// newest-first.
vcsRows, err := h.Queries.ListVCSPullRequestsByIssue(r.Context(), issue.ID)
if err != nil {
writeError(w, http.StatusInternalServerError, "failed to list pull requests")
return
}
for _, row := range vcsRows {
out = append(out, vcsPullRequestRowToResponse(row))
}
sort.SliceStable(out, func(i, j int) bool {
return out[i].PRCreatedAt > out[j].PRCreatedAt
})
writeJSON(w, http.StatusOK, map[string]any{"pull_requests": out})
}
// broadcastPRSnapshotApplied is the ghsnapshot pipeline's onApplied callback:
// once an API snapshot is written to a PR row, re-broadcast the PR so every
// open issue detail page re-queries its PR list and picks up the fresh CI /
// mergeability state. Runs on a background pipeline goroutine.
func (h *Handler) broadcastPRSnapshotApplied(ctx context.Context, prID pgtype.UUID) {
pr, err := h.Queries.GetGitHubPullRequestByID(ctx, prID)
if err != nil {
return
}
issueIDs, err := h.Queries.ListIssueIDsForPullRequest(ctx, prID)
if err != nil {
return
}
linked := make([]string, 0, len(issueIDs))
for _, id := range issueIDs {
linked = append(linked, uuidToString(id))
}
h.publish(protocol.EventPullRequestUpdated, uuidToString(pr.WorkspaceID), "system", "", map[string]any{
"pull_request": githubPullRequestToResponse(pr, h.PRRefresh.Enabled()),
"linked_issue_ids": linked,
})
}
// ── Webhook ─────────────────────────────────────────────────────────────────
// identifierRe extracts identifiers like "MUL-1510" from text. Case-insensitive
// because branch names are conventionally lowercase but issue prefixes are
// uppercase. Word boundary on the left prevents matching inside email-style
// strings (e.g. "abc@MUL-1") and the digit anchor on the right rules out
// version numbers like "v1.2-3".
var identifierRe = regexp.MustCompile(`(?i)\b([a-z][a-z0-9]{1,9})-(\d+)\b`)
// closingIdentifierRe extracts identifiers that appear immediately after a
// GitHub-style closing keyword ("close[sd]?", "fix(e[sd])?", "resolve[sd]?"),
// optionally separated by a colon and whitespace. Matching is intentionally
// strict on adjacency — "Fix MUL-1" closes MUL-1, but "Fix login MUL-1"
// does not. This mirrors GitHub's own closing-keyword grammar and is the
// gate the webhook uses to decide whether to auto-advance an issue to
// `done` after a PR merges. References like "Follow up in MUL-2" and bare
// title prefixes like "MUL-1: ..." link the PR (via identifierRe) but
// never auto-close.
var closingIdentifierRe = regexp.MustCompile(
`(?i)\b(?:close[sd]?|fix(?:e[sd])?|resolve[sd]?)[:\s]+([a-z][a-z0-9]{1,9})-(\d+)\b`,
)
// HandleGitHubWebhook (POST /api/webhooks/github) is GitHub's destination for
// every event from a connected installation. We verify HMAC signature, route
// on X-GitHub-Event, and either upsert PR rows + auto-link to issues or
// remove the installation on uninstall.
func (h *Handler) HandleGitHubWebhook(w http.ResponseWriter, r *http.Request) {
body, err := io.ReadAll(io.LimitReader(r.Body, 10<<20)) // 10 MiB cap
if err != nil {
writeError(w, http.StatusBadRequest, "read body failed")
return
}
secret := githubWebhookSecret()
if secret == "" {
// Refusing to process webhooks at all is safer than treating an
// unconfigured deployment as "all signatures valid".
writeError(w, http.StatusServiceUnavailable, "github webhooks not configured")
return
}
sigHeader := r.Header.Get("X-Hub-Signature-256")
if !verifyWebhookSignature(secret, sigHeader, body) {
writeError(w, http.StatusUnauthorized, "invalid signature")
return
}
event := r.Header.Get("X-GitHub-Event")
ctx := r.Context()
switch event {
case "ping":
writeJSON(w, http.StatusOK, map[string]string{"ok": "pong"})
return
case "installation":
h.handleInstallationEvent(ctx, body)
case "pull_request":
h.handlePullRequestEvent(ctx, body)
case "check_suite", "check_run", "status":
// CI events are pure triggers under Plan C (MUL-5265): their payload is
// never read for display. Each just asks the API pipeline to re-fetch
// the authoritative snapshot for the PR(s) it concerns.
h.triggerPRRefreshFromCIEvent(ctx, body)
default:
// Acknowledge every event so GitHub doesn't mark the endpoint failing,
// but ignore types we don't model.
}
w.WriteHeader(http.StatusAccepted)
}
func verifyWebhookSignature(secret, header string, body []byte) bool {
const prefix = "sha256="
if !strings.HasPrefix(header, prefix) {
return false
}
want, err := hex.DecodeString(strings.TrimPrefix(header, prefix))
if err != nil {
return false
}
mac := hmac.New(sha256.New, []byte(secret))
mac.Write(body)
return hmac.Equal(mac.Sum(nil), want)
}
type ghInstallationPayload struct {
Action string `json:"action"`
Installation struct {
ID int64 `json:"id"`
Account struct {
Login string `json:"login"`
Type string `json:"type"`
AvatarURL string `json:"avatar_url"`
} `json:"account"`
} `json:"installation"`
}
func githubInstallationAccountFromPayload(p ghInstallationPayload) (login, accountType string, avatar *string, ok bool) {
login = strings.TrimSpace(p.Installation.Account.Login)
if login == "" {
return "", "", nil, false
}
accountType = coalesce(p.Installation.Account.Type, "User")
avatar = strPtrOrNil(p.Installation.Account.AvatarURL)
return login, accountType, avatar, true
}
func (h *Handler) handleInstallationEvent(ctx context.Context, body []byte) {
var p ghInstallationPayload
if err := json.Unmarshal(body, &p); err != nil {
slog.Warn("github: bad installation payload", "err", err)
return
}
switch p.Action {
case "deleted", "suspend":
// User removed/suspended the App on GitHub — trust in this
// installation_id is gone entirely, so drop every workspace binding.
// We DELETE … RETURNING so each broadcast can be scoped to its
// workspace; events without WorkspaceID are dropped by the realtime
// listener and would leave already-open Settings tabs stale.
deleted, err := h.Queries.DeleteGitHubInstallationByInstallationID(ctx, p.Installation.ID)
if err != nil {
slog.Warn("github: delete installation failed", "err", err, "installation_id", p.Installation.ID)
return
}
if err := h.Queries.DeletePendingGitHubInstallation(ctx, p.Installation.ID); err != nil {
slog.Warn("github: delete pending installation failed", "err", err, "installation_id", p.Installation.ID)
}
// Broadcast the internal row id only — the numeric installation_id is
// a management handle that non-admin members are not allowed to see.
// The frontend invalidates the installations query on this event and
// does not read the broadcast payload directly. One broadcast per
// deleted binding so every affected workspace's Settings tab refreshes.
for _, row := range deleted {
h.publish(protocol.EventGitHubInstallationDeleted, uuidToString(row.WorkspaceID), "system", "", map[string]any{
"id": uuidToString(row.ID),
})
}
case "created", "new_permissions_accepted", "unsuspend":
login, accountType, avatar, ok := githubInstallationAccountFromPayload(p)
if !ok {
slog.Warn("github: installation payload missing account login", "installation_id", p.Installation.ID)
return
}
// We don't know which workspace(s) this maps to from the webhook
// alone. If no setup callback has created a workspace binding yet,
// keep the account metadata and let the callback consume it after it
// creates github_installation.
existing, err := h.Queries.ListGitHubInstallationsByInstallationID(ctx, p.Installation.ID)
if err != nil {
slog.Warn("github: lookup installation failed", "err", err, "installation_id", p.Installation.ID)
return
}
if len(existing) == 0 {
if _, err := h.Queries.UpsertPendingGitHubInstallation(ctx, db.UpsertPendingGitHubInstallationParams{
InstallationID: p.Installation.ID,
AccountLogin: login,
AccountType: accountType,
AccountAvatarUrl: ptrToText(avatar),
}); err != nil {
slog.Warn("github: store pending installation failed", "err", err, "installation_id", p.Installation.ID)
}
return
}
// Refresh the account display metadata across every workspace binding;
// workspace_id and connected_by_id are left untouched.
refreshed, err := h.Queries.UpdateGitHubInstallationAccountByInstallationID(ctx, db.UpdateGitHubInstallationAccountByInstallationIDParams{
InstallationID: p.Installation.ID,
AccountLogin: login,
AccountType: accountType,
AccountAvatarUrl: ptrToText(avatar),
})
if err != nil {
slog.Warn("github: refresh installation failed", "err", err)
return
}
if err := h.Queries.DeletePendingGitHubInstallation(ctx, p.Installation.ID); err != nil {
slog.Warn("github: delete pending installation failed", "err", err, "installation_id", p.Installation.ID)
}
// Broadcast so any open Settings → GitHub tab re-queries the
// installations list. Without this, a row created by the setup
// callback with the "unknown" placeholder (e.g. because GitHub
// App JWT auth wasn't configured, or this webhook arrived after
// the user already loaded the page) would stay visibly stale
// until the user manually refreshes. One broadcast per bound workspace.
for _, inst := range refreshed {
h.publish(protocol.EventGitHubInstallationCreated, uuidToString(inst.WorkspaceID), "system", "", map[string]any{
"installation": githubInstallationToBroadcast(inst),
})
}
}
}
type ghPullRequestPayload struct {
Action string `json:"action"`
PullRequest struct {
Number int32 `json:"number"`
HTMLURL string `json:"html_url"`
Title string `json:"title"`
Body string `json:"body"`
State string `json:"state"`
Draft bool `json:"draft"`
Merged bool `json:"merged"`
MergedAt string `json:"merged_at"`
ClosedAt string `json:"closed_at"`
CreatedAt string `json:"created_at"`
UpdatedAt string `json:"updated_at"`
MergeableState string `json:"mergeable_state"`
Additions int32 `json:"additions"`
Deletions int32 `json:"deletions"`
ChangedFiles int32 `json:"changed_files"`
Head struct {
Ref string `json:"ref"`
SHA string `json:"sha"`
} `json:"head"`
User struct {
Login string `json:"login"`
AvatarURL string `json:"avatar_url"`
} `json:"user"`
} `json:"pull_request"`
Changes *ghPRChanges `json:"changes"`
Repository struct {
Name string `json:"name"`
Owner struct {
Login string `json:"login"`
} `json:"owner"`
} `json:"repository"`
Installation struct {
ID int64 `json:"id"`
} `json:"installation"`
}
func (h *Handler) handlePullRequestEvent(ctx context.Context, body []byte) {
var p ghPullRequestPayload
if err := json.Unmarshal(body, &p); err != nil {
slog.Warn("github: bad pull_request payload", "err", err)
return
}
if p.Installation.ID == 0 {
return
}
insts, err := h.Queries.ListGitHubInstallationsByInstallationID(ctx, p.Installation.ID)
if err != nil {
slog.Warn("github: lookup installation failed", "err", err)
return
}
if len(insts) == 0 {
// Webhook from an installation we never wired up — nothing we
// can attribute to a workspace, so drop it silently.
return
}
// #4855 lets one GitHub App installation bind to several workspaces. A
// repo's events belong to every bound workspace, so fan the delivery out:
// each workspace independently mirrors the PR and auto-links it against its
// own issues (its own prefix + github toggle). Repo scope is whatever GitHub
// authorized the installation for; we deliberately don't gate on the
// workspace.repos registry — that list is "code the agent clones", not a
// webhook subscription (MUL-4343).
//
// Fanning out means an identifier can resolve in more than one workspace at
// once (issue prefixes are not globally unique and issue numbers restart at
// 1 per workspace, so two bound workspaces can both own a real "ABC-100").
// Settle who — if anyone — may act on a closing keyword BEFORE any workspace
// links, and treat that verdict as authoritative for the whole delivery, so
// the mirror pass cannot re-derive a different answer. See closeIntentPolicy.
closePolicy := h.resolveCloseIntentPolicy(ctx, insts, &p)
for _, inst := range insts {
h.mirrorPullRequestForWorkspace(ctx, inst.WorkspaceID, inst.InstallationID, &p, closePolicy)
}
// The PR row(s) now carry the new head; ask the API pipeline for the
// authoritative CI + mergeability snapshot for that head. The webhook is
// only the doorbell — its own mergeable/checks payload is not used for
// display anymore (MUL-5265).
h.PRRefresh.Enqueue(p.Installation.ID, p.Repository.Owner.Login, p.Repository.Name, p.PullRequest.Number)
}
// closeIntentPolicy decides which (closing identifier, workspace) pairs this
// delivery is allowed to act on, and is the single authority for that decision:
// the per-workspace mirror pass consults it instead of re-deriving the answer
// from its own reads.
//
// It is an allowlist, not a denylist, so it fails closed by construction — an
// identifier we could not positively attribute to exactly one workspace is
// simply absent, and absence denies. The zero value therefore permits nothing,
// which is what every indeterminate read returns.
type closeIntentPolicy struct {
// unrestricted marks the single-binding case: cross-workspace ambiguity is
// impossible with one bound workspace, so every closing identifier keeps
// its pre-#6804 behavior and the scan does no reads at all.
unrestricted bool
// owner maps a closing identifier to the one workspace proven to resolve
// it. Recording the winner (rather than a bare "allowed") also closes the
// window between the scan and the mirror pass: if a second workspace grows
// a same-numbered issue in between, it is not the recorded owner, so it
// still cannot act.
owner map[string]string
}
// permits reports whether workspaceID may carry close intent for identifier.
func (c closeIntentPolicy) permits(identifier, workspaceID string) bool {
if c.unrestricted {
return true
}
owner, ok := c.owner[identifier]
return ok && owner == workspaceID
}
// resolveCloseIntentPolicy determines, before any workspace links, which
// closing identifiers on this PR may advance an issue and in which workspace.
//
// Issue prefixes are deliberately not globally unique (#2797) and issue numbers
// restart at 1 in every workspace, so once #5183 fanned PR webhooks out to every
// bound workspace, "Closes ABC-100" could resolve to a genuine — but different —
// issue in each of them. Linking in both is recoverable noise; auto-advancing
// both to done is not, because it silently rewrites the status of an issue in a
// workspace that has nothing to do with this PR (#6804).
//
// An identifier is allowed only when exactly one bound workspace was *proven*
// to resolve it. Anything else — two resolvers, or a read we could not complete
// — leaves it out, so no workspace acts on it. That asymmetry is deliberate:
// misjudging "unique" as "ambiguous" costs an auto-complete a human can perform,
// while misjudging "ambiguous" as "unique" silently closes someone else's issue.
//
// Only closing identifiers are examined, and only when the installation has more
// than one binding, so a PR without a closing keyword — or the overwhelmingly
// common single-workspace installation — does no extra work.
func (h *Handler) resolveCloseIntentPolicy(ctx context.Context, insts []db.GithubInstallation, p *ghPullRequestPayload) closeIntentPolicy {
if len(insts) < 2 {
return closeIntentPolicy{unrestricted: true}
}
idents := extractClosingIdentifiers(p.PullRequest.Title, p.PullRequest.Body)
if len(idents) == 0 {
return closeIntentPolicy{}
}
prNumber := p.PullRequest.Number
repo := p.Repository.Owner.Login + "/" + p.Repository.Name
// Collect every workspace that resolves each identifier. Any read we cannot
// complete abandons the whole event: a workspace we failed to inspect might
// be a second resolver, and without ruling that out we cannot call any other
// workspace the unique one.
resolvers := make(map[string][]string, len(idents))
for _, inst := range insts {
ws, err := h.Queries.GetWorkspace(ctx, inst.WorkspaceID)
if err != nil {
slog.Warn("github: cannot load bound workspace, withholding close intent for this delivery",
"err", err, "installation_id", p.Installation.ID, "repo", repo, "pr_number", prNumber)
return closeIntentPolicy{}
}
// A workspace with auto-link off never writes a link row, so it is not a
// competing claimant and must not suppress one that would legitimately
// act. A settings blob we cannot parse is not evidence of either, so it
// fails closed rather than defaulting to "enabled".
autoLink, err := autoLinkPRsEnabledForWorkspace(ws)
if err != nil {
slog.Warn("github: cannot read workspace auto-link setting, withholding close intent for this delivery",
"err", err, "workspace_id", uuidToString(inst.WorkspaceID),
"installation_id", p.Installation.ID, "repo", repo, "pr_number", prNumber)
return closeIntentPolicy{}
}
if !autoLink {
continue
}
prefix := issuePrefixForWorkspace(ws)
for _, id := range idents {
number, ok := issueNumberForPrefix(id, prefix)
if !ok {
continue
}
if _, err := h.Queries.GetIssueByNumber(ctx, db.GetIssueByNumberParams{
WorkspaceID: inst.WorkspaceID,
Number: number,
}); err != nil {
if errors.Is(err, pgx.ErrNoRows) {
// Proven miss: this workspace has no such issue.
continue
}
slog.Warn("github: cannot resolve identifier in bound workspace, withholding close intent for this delivery",
"err", err, "identifier", id, "workspace_id", uuidToString(inst.WorkspaceID),
"installation_id", p.Installation.ID, "repo", repo, "pr_number", prNumber)
return closeIntentPolicy{}
}
resolvers[id] = append(resolvers[id], uuidToString(inst.WorkspaceID))
}
}
policy := closeIntentPolicy{owner: make(map[string]string, len(resolvers))}
for id, wss := range resolvers {
if len(wss) == 1 {
policy.owner[id] = wss[0]
continue
}
// The symptom this prevents (an issue advancing to done for no visible
// reason) is otherwise untraceable back to GitHub, so leave a breadcrumb
// naming the collision an operator has to resolve by changing a prefix.
slog.Warn("github: ambiguous closing identifier across bound workspaces, withholding close intent",
"identifier", id,
"workspaces", len(wss),
"installation_id", p.Installation.ID,
"repo", repo,
"pr_number", prNumber,
)
}
return policy
}
// ghCIEventPayload captures the shared shape of the check_suite / check_run /
// status webhooks — enough to resolve which PR to refresh. These events are
// pure triggers under Plan C: their payload data is never read for display.
type ghCIEventPayload struct {
Installation struct {
ID int64 `json:"id"`
} `json:"installation"`
Repository struct {
Name string `json:"name"`
Owner struct {
Login string `json:"login"`
} `json:"owner"`
} `json:"repository"`
// status events: top-level commit SHA, no PR number.
SHA string `json:"sha"`
CheckSuite struct {
HeadSHA string `json:"head_sha"`
PullRequests []struct {
Number int32 `json:"number"`
} `json:"pull_requests"`
} `json:"check_suite"`
CheckRun struct {
PullRequests []struct {
Number int32 `json:"number"`
} `json:"pull_requests"`
CheckSuite struct {
HeadSHA string `json:"head_sha"`
} `json:"check_suite"`
} `json:"check_run"`
}
// triggerPRRefreshFromCIEvent enqueues an API refresh for the PR(s) a
// check_suite / check_run / status webhook concerns. check_suite/check_run
// carry the PR numbers directly; status events carry only a commit SHA, so we
// map it back to the mirrored head_sha to find the PR(s).
func (h *Handler) triggerPRRefreshFromCIEvent(ctx context.Context, body []byte) {
if !h.PRRefresh.Enabled() {
return
}
var p ghCIEventPayload
if err := json.Unmarshal(body, &p); err != nil {
return
}
if p.Installation.ID == 0 || p.Repository.Name == "" {
return
}
owner, repo := p.Repository.Owner.Login, p.Repository.Name
seen := map[int32]struct{}{}
enqueue := func(number int32) {
if number == 0 {
return
}
if _, ok := seen[number]; ok {
return
}
seen[number] = struct{}{}
h.PRRefresh.Enqueue(p.Installation.ID, owner, repo, number)
}
for _, pr := range p.CheckSuite.PullRequests {
enqueue(pr.Number)
}
for _, pr := range p.CheckRun.PullRequests {
enqueue(pr.Number)
}
if len(seen) > 0 {
return
}
// No PR number in the payload (status event, or a check event whose
// pull_requests array was empty) — resolve by head SHA.
sha := p.SHA
if sha == "" {
sha = coalesce(p.CheckSuite.HeadSHA, p.CheckRun.CheckSuite.HeadSHA)
}
if sha == "" {
return
}
numbers, err := h.Queries.ListGitHubPRNumbersByHeadSHA(ctx, db.ListGitHubPRNumbersByHeadSHAParams{
InstallationID: p.Installation.ID,
RepoOwner: owner,
RepoName: repo,
HeadSha: sha,
})
if err != nil {
return
}
for _, number := range numbers {
enqueue(number)
}
}
// mirrorPullRequestForWorkspace mirrors a pull_request webhook into a single
// workspace: it upserts the PR row, replays any check_suite events that
// arrived before the PR was mirrored, auto-links referenced issues (gated by
// the workspace's github toggles), advances issues on terminal events, and
// broadcasts the change. Invoked once per workspace bound to the delivering
// installation.
//
// closePolicy is the delivery-wide verdict on which closing identifiers this
// workspace may act on; identifiers it does not permit still link, but never
// carry close_intent, so they can never advance an issue to done. This function
// only ever narrows that verdict — it cannot grant close intent the policy
// withheld.
func (h *Handler) mirrorPullRequestForWorkspace(ctx context.Context, wsID pgtype.UUID, installationID int64, p *ghPullRequestPayload, closePolicy closeIntentPolicy) {
state := derivePRState(p.PullRequest.State, p.PullRequest.Draft, p.PullRequest.Merged)
mergeable, clearMergeable := derivePRMergeableState(p.Action, p.PullRequest.MergeableState, baseRefChanged(p.Changes))
pr, err := h.Queries.UpsertGitHubPullRequest(ctx, db.UpsertGitHubPullRequestParams{
WorkspaceID: wsID,
InstallationID: installationID,
RepoOwner: p.Repository.Owner.Login,
RepoName: p.Repository.Name,
PrNumber: p.PullRequest.Number,
Title: p.PullRequest.Title,
State: state,
HtmlUrl: p.PullRequest.HTMLURL,
Branch: ptrToText(strPtrOrNil(p.PullRequest.Head.Ref)),
AuthorLogin: ptrToText(strPtrOrNil(p.PullRequest.User.Login)),
AuthorAvatarUrl: ptrToText(strPtrOrNil(p.PullRequest.User.AvatarURL)),
MergedAt: parseGHTime(p.PullRequest.MergedAt),
ClosedAt: parseGHTime(p.PullRequest.ClosedAt),
PrCreatedAt: parseGHTimeRequired(p.PullRequest.CreatedAt),
PrUpdatedAt: parseGHTimeRequired(p.PullRequest.UpdatedAt),
HeadSha: p.PullRequest.Head.SHA,
MergeableState: mergeable,
ClearMergeableState: pgtype.Bool{Bool: clearMergeable, Valid: true},
Additions: p.PullRequest.Additions,
Deletions: p.PullRequest.Deletions,
ChangedFiles: p.PullRequest.ChangedFiles,
})
if err != nil {
slog.Warn("github: upsert pr failed", "err", err)
return
}
workspaceID := uuidToString(wsID)
resp := githubPullRequestToResponse(pr, h.PRRefresh.Enabled())
// Auto-link: scan title/body/branch for issue identifiers, look them
// up in this workspace, attach the link rows. Idempotent (ON CONFLICT
// upserts the close_intent flag — see LinkIssueToPullRequest) so
// re-firing the webhook doesn't duplicate.
//
// RFC MUL-2414 §4.8: the PR mirror upsert above always runs (so re-enabling
// GitHub features restores history without backfill), but the link rows
// are a "new side-effect" and must be gated by the workspace's auto-link
// flag (which itself short-circuits when the master `github_enabled`
// switch is off).
linkedIssueIDs := make([]string, 0)
if h.workspaceAutoLinkPRsEnabled(ctx, wsID) {
idents := extractIdentifiers(p.PullRequest.Title, p.PullRequest.Body, p.PullRequest.Head.Ref)
// closingIdents is the subset of identifiers that this PR explicitly
// declared via a closing keyword ("Closes/Fixes/Resolves MUL-X").
// Linking still happens for every mention (idents above), but the
// link row's close_intent column — and therefore whether the
// auto-advance gate eventually fires — is only set for keyword-
// declared identifiers. Bare title prefixes and branch-name
// references are link-only.
closingIdents := map[string]struct{}{}
for _, c := range extractClosingIdentifiers(p.PullRequest.Title, p.PullRequest.Body) {
closingIdents[c] = struct{}{}
}
// qualifyingIdents are the identifiers that genuinely tie this PR to an
// issue: a title prefix, a branch-name reference, or a body closing
// keyword. Any identifier that is linked but NOT in this set was matched
// only by a bare mention in the PR body ("Related MUL-1", "Follow up in
// MUL-1"). Those links are still recorded (auto-link stays generous so
// close_intent can be tracked across edits) but are flagged
// reference_only and hidden from the issue's PR list — a passing mention
// should not surface the PR as a working PR for that issue (MUL-3739).
qualifyingIdents := map[string]struct{}{}
for _, id := range extractIdentifiers(p.PullRequest.Title, p.PullRequest.Head.Ref) {
qualifyingIdents[id] = struct{}{}
}
for c := range closingIdents {
qualifyingIdents[c] = struct{}{}
}
// close_intent should follow the PR title/body while the PR is still
// editable before its terminal close event. Once GitHub has delivered
// a terminal event, later edit/synchronize webhooks must not rewrite
// the merge-time close decision.
preserveCloseIntent := p.Action != "closed" && (state == "merged" || state == "closed")
prefix := h.getIssuePrefix(ctx, wsID)
// reevalIssues collects each issue whose link row we just touched so
// we can re-run the auto-advance gate against the persisted aggregate
// after every link upsert in this event. Driving the gate off
// persisted state (instead of "did *this* webhook declare closing
// intent?") is what fixes the multi-PR sibling case: a PR with
// `Closes MUL-1` merges first while a link-only sibling is still
// open, then the sibling closes later — its webhook has no closing
// keyword, but the earlier link row carries close_intent=true, so
// MUL-1 still advances.
reevalIssues := make([]db.Issue, 0, len(idents))
for _, id := range idents {
issue, ok := h.lookupIssueByIdentifier(ctx, wsID, prefix, id)
if !ok {
continue
}
_, declared := closingIdents[id]
if declared && !closePolicy.permits(id, workspaceID) {
// The delivery-wide scan did not prove this workspace is the one
// this closing keyword refers to, so it does not act on it.
// Falling through with declared=false also clears close_intent
// on a row a pre-#6804 delivery had already set, so a re-fired
// webhook heals the stored decision.
declared = false
}
closeIntent := declared && !preserveCloseIntent
_, qualifies := qualifyingIdents[id]
referenceOnly := !qualifies
if err := h.Queries.LinkIssueToPullRequest(ctx, db.LinkIssueToPullRequestParams{
IssueID: issue.ID,
PullRequestID: pr.ID,
CloseIntent: closeIntent,
ReferenceOnly: referenceOnly,
PreserveCloseIntent: preserveCloseIntent,
LinkedByType: strToText("system"),
LinkedByID: pgtype.UUID{},
}); err != nil {
slog.Warn("github: link failed", "err", err)
continue
}
linkedIssueIDs = append(linkedIssueIDs, uuidToString(issue.ID))
reevalIssues = append(reevalIssues, issue)
}
// A terminal PR event (`merged` or `closed`) may be the moment the
// last in-flight sibling resolves. We re-evaluate every issue we
// just linked once both the PR row and the link row are persisted,
// so the aggregate query sees the freshest state. We advance the
// issue to done when:
// 1. the issue isn't already terminal (`done` / `cancelled`);
// 2. no linked PR is still `open` / `draft`;
// 3. at least one merged linked PR declared close_intent (a
// "Closes/Fixes/Resolves" keyword on its link row).
// Rule (3) is what prevents "Follow up in MUL-2" / "Unblocks MUL-3"
// references from being treated the same as "Closes MUL-1", and
// also prevents an "all closed-without-merge" sequence from
// silently auto-closing the issue — if nothing carrying closing
// intent was ever delivered, the user should decide manually.
if state == "merged" || state == "closed" {
for _, issue := range reevalIssues {
// A custom terminal status counts as terminal here. (MUL-6243)
if s := issuestatus.Effective(ctx, h.Queries, issue.WorkspaceID, issue.Status); s == "done" || s == "cancelled" {
continue
}
// Combined across providers: an issue may also carry a still-open
// self-hosted VCS PR, which must block auto-advance here just as
// an open GitHub PR blocks it on the VCS webhook path.
counts, err := h.Queries.GetIssueCombinedPullRequestCloseAggregate(ctx, issue.ID)
if err != nil {
slog.Warn("github: count linked pr states failed", "err", err, "issue_id", uuidToString(issue.ID))
continue
}
if counts.OpenCount == 0 && counts.MergedWithCloseIntentCount > 0 {
h.advanceIssueToDone(ctx, issue, workspaceID)
}
}
}
}
// Broadcast PR change to the workspace so any open issue detail page
// re-queries its PR list.
h.publish(protocol.EventPullRequestUpdated, workspaceID, "system", "", map[string]any{
"pull_request": resp,
"linked_issue_ids": linkedIssueIDs,
})
}
// derivePRMergeableState resolves the upsert behaviour for the PR row's
// mergeable_state column on a `pull_request` webhook. It returns three
// states encoded as (value, clear):
//
// - clear=true → force the column to NULL. State-changing actions (`opened`,
// `synchronize`, `reopened`, or a base-branch swap) must blank the value
// because GitHub re-computes mergeability asynchronously; the payload may
// still carry the previous head's clean/dirty answer, and trusting it
// would surface a stale verdict against the new head.
// - clear=false, value valid → write the value. The event carried a
// concrete verdict we should persist.
// - clear=false, value invalid → preserve the existing column. Metadata
// events (labeled/assigned/edited-without-base-swap) ship pull_request
// payloads with mergeable_state empty even when the previous verdict is
// still accurate, and silently overwriting clean/dirty with NULL would
// drop information GitHub only refreshes lazily.
func derivePRMergeableState(action, payload string, baseRefChanged bool) (pgtype.Text, bool) {
if action == "opened" || action == "synchronize" || action == "reopened" {
return pgtype.Text{}, true
}
if action == "edited" && baseRefChanged {
return pgtype.Text{}, true
}
if payload == "" {
return pgtype.Text{}, false
}
return pgtype.Text{String: payload, Valid: true}, false
}
// ghPRChanges captures the only field of `pull_request.edited`'s `changes`
// payload we care about: a base-branch swap. Everything else (title, body)
// leaves mergeability intact.
type ghPRChanges struct {
Base *struct {
Ref *struct {
From string `json:"from"`
} `json:"ref"`
} `json:"base"`
}
// baseRefChanged returns true when a pull_request.edited event indicates the
// PR's base branch was swapped. Only this kind of edit invalidates the
// existing mergeable_state.
func baseRefChanged(c *ghPRChanges) bool {
return c != nil && c.Base != nil && c.Base.Ref != nil && c.Base.Ref.From != ""
}
func derivePRState(state string, draft, merged bool) string {
if merged {
return "merged"
}
if state == "closed" {
return "closed"
}
if draft {
return "draft"
}
return "open"
}
func parseGHTime(s string) pgtype.Timestamptz {
if s == "" {
return pgtype.Timestamptz{}
}
t, err := time.Parse(time.RFC3339, s)
if err != nil {
return pgtype.Timestamptz{}
}
return pgtype.Timestamptz{Time: t, Valid: true}
}
func parseGHTimeRequired(s string) pgtype.Timestamptz {
t := parseGHTime(s)
if !t.Valid {
return pgtype.Timestamptz{Time: time.Now().UTC(), Valid: true}
}
return t
}
// extractIdentifiers pulls every "PREFIX-NUMBER" match across the supplied
// fields, deduplicating in input order.
func extractIdentifiers(parts ...string) []string {
seen := map[string]struct{}{}
out := []string{}
for _, src := range parts {
for _, m := range identifierRe.FindAllStringSubmatch(src, -1) {
ident := strings.ToUpper(m[1]) + "-" + m[2]
if _, dup := seen[ident]; dup {
continue
}
seen[ident] = struct{}{}
out = append(out, ident)
}
}
return out
}
// extractClosingIdentifiers pulls every "PREFIX-NUMBER" identifier that
// appears immediately after a GitHub-style closing keyword in the supplied
// fields, deduplicating in input order. Identifiers in branch names are
// intentionally excluded — callers should pass only title and body — because
// branch names are not natural-language fields and treating "mul-1/fix-login"
// as a close declaration would silently re-open the bug this gate is meant
// to fix.
func extractClosingIdentifiers(parts ...string) []string {
seen := map[string]struct{}{}
out := []string{}
for _, src := range parts {
for _, m := range closingIdentifierRe.FindAllStringSubmatch(src, -1) {
ident := strings.ToUpper(m[1]) + "-" + m[2]
if _, dup := seen[ident]; dup {
continue
}
seen[ident] = struct{}{}
out = append(out, ident)
}
}
return out
}
// autoLinkPRsEnabledForWorkspace reports whether the workspace allows the
// GitHub webhook to create issue ↔ PR link rows. Defaults to true so that
// workspaces predating RFC MUL-2414 keep the historical "auto-link on"
// behavior, and short-circuits to false whenever the master GitHub switch
// is explicitly off — mirroring the precedence used on the client side.
//
// An unparseable settings blob is surfaced as an error instead of being folded
// into the permissive default. Callers deciding whether to write a link row
// keep taking the default, but the close-intent scan has to tell "auto-link is
// on" apart from "we could not find out" — see closeIntentPolicy.
func autoLinkPRsEnabledForWorkspace(ws db.Workspace) (bool, error) {
if len(ws.Settings) == 0 {
return true, nil
}
var s struct {
GitHubEnabled *bool `json:"github_enabled"`
GitHubAutoLinkPRsEnabled *bool `json:"github_auto_link_prs_enabled"`
}
if err := json.Unmarshal(ws.Settings, &s); err != nil {
return true, err
}
if s.GitHubEnabled != nil && !*s.GitHubEnabled {
return false, nil
}
if s.GitHubAutoLinkPRsEnabled == nil {
return true, nil
}
return *s.GitHubAutoLinkPRsEnabled, nil
}
func (h *Handler) workspaceAutoLinkPRsEnabled(ctx context.Context, workspaceID pgtype.UUID) bool {
ws, err := h.Queries.GetWorkspace(ctx, workspaceID)
if err != nil {
return true
}
enabled, _ := autoLinkPRsEnabledForWorkspace(ws)
return enabled
}
// issueNumberForPrefix returns the issue number encoded in a "PREFIX-NUMBER"
// identifier, and false when the identifier is malformed or carries a prefix
// that is not the workspace's. Case-insensitive: branch names are
// conventionally lowercase while issue prefixes are uppercase.
func issueNumberForPrefix(identifier, prefix string) (int32, bool) {
idx := strings.LastIndex(identifier, "-")
if idx < 0 {
return 0, false
}
gotPrefix, numStr := identifier[:idx], identifier[idx+1:]
if !strings.EqualFold(gotPrefix, prefix) {
return 0, false
}
n, err := strconv.Atoi(numStr)
if err != nil {
return 0, false
}
return int32(n), true
}
// lookupIssueByIdentifier looks up an issue in the given workspace by its
// "PREFIX-NUMBER" identifier. Returns the row + true if the prefix matches
// the workspace's configured prefix and the number resolves to a real issue.
func (h *Handler) lookupIssueByIdentifier(ctx context.Context, workspaceID pgtype.UUID, prefix, identifier string) (db.Issue, bool) {
number, ok := issueNumberForPrefix(identifier, prefix)
if !ok {
return db.Issue{}, false
}
issue, err := h.Queries.GetIssueByNumber(ctx, db.GetIssueByNumberParams{
WorkspaceID: workspaceID,
Number: number,
})
if err != nil {
return db.Issue{}, false
}
return issue, true
}
func (h *Handler) advanceIssueToDone(ctx context.Context, issue db.Issue, workspaceID string) {
updated, err := h.Queries.UpdateIssueStatus(ctx, db.UpdateIssueStatusParams{
ID: issue.ID,
Status: "done",
WorkspaceID: issue.WorkspaceID,
})
if err != nil {
slog.Warn("github: advance issue to done failed", "err", err)
return
}
// Fire the platform parent-notification path on the same transition the
// HTTP UpdateIssue / BatchUpdateIssues paths use. A merged PR is one of
// the most common ways a sub-issue actually reaches `done`, and skipping
// it here would leave the parent silent for the dominant completion path.
// notifyParentOfChildDone re-checks every guard (prev != done, parent
// exists, parent not terminal), so calling it unconditionally is safe.
h.notifyParentOfChildDone(ctx, issue, updated)
prefix := h.getIssuePrefix(ctx, issue.WorkspaceID)
resp := issueToResponse(updated, prefix)
h.fillStatusCategory(ctx, updated.WorkspaceID, &resp)
h.publish(protocol.EventIssueUpdated, workspaceID, "system", "", map[string]any{
"issue": resp,
"status_changed": true,
"prev_status": issue.Status,
"creator_type": issue.CreatorType,
"creator_id": uuidToString(issue.CreatorID),
"source": "github_pr_merged",
})
}
// ── Helpers ─────────────────────────────────────────────────────────────────
func parseStrictUUID(s string) (pgtype.UUID, error) {
var u pgtype.UUID
if err := u.Scan(s); err != nil {
return pgtype.UUID{}, err
}
return u, nil
}
func coalesce(a, fallback string) string {
if strings.TrimSpace(a) == "" {
return fallback
}
return a
}
func strPtrOrNil(s string) *string {
if s == "" {
return nil
}
v := s
return &v
}