You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
aft_callgraph while the store is building: report progress, allow a bounded wait, or return marked lexical candidates #606
While the call-graph store is building, every aft_callgraph op returns at once:
callers: callgraph_building — callers: callgraph store is building in the background; retry shortly
data: {"index":{"callgraph":{"status":"building","reason":null}},"results":null}
The reply has no progress and no estimate. The agent can't tell whether "shortly" means seconds or ten minutes. On a large repository we got this answer twice, about four minutes apart, for the same query. So agents either loop on the call (wasting turns), or give up and grep without saying the result is unverified. Checked on main (396d0bf): no bounded wait and no fallback.
Proposal
Any one of these would help; together they would remove the dead end:
Progress in the reply: files done / total, or elapsed time, so the agent can decide whether to wait.
A bounded wait: an optional wait_ms (capped by config) that blocks until the store is ready or the wait expires, then answers or reports building with progress.
A marked fallback: for callers and impact, return the lexical candidates (files that reference the name), clearly labelled unverified: lexical candidates while the call graph builds, with complete: false.
Option 3 must never present lexical candidates as the call-graph answer.
Activity