From b932a814caec526f536ec7ae663a85a470148b47 Mon Sep 17 00:00:00 2001
From: Erich Blume <blume.erich@gmail.com>
Date: Tue, 2 Jun 2026 10:23:42 -0700
Subject: [PATCH] heph.nvim: fix daemon.wait_ready deadlock on a stale socket
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

wait_ready ran the rpc health probe inside a `vim.wait` predicate, and the probe
itself uses `vim.wait` — nesting vim.wait inside another vim.wait's predicate
deadlocks Neovim. It only bit when the socket file existed: the first launch's
socket doesn't exist yet (probe short-circuits), but the second launch hit the
stale socket left by the prior daemon and froze in setup().

- wait_ready now probes in a plain Lua loop (deadline via uv.hrtime + a bare
  vim.wait(50) yield) — never a vim.wait inside a vim.wait predicate.
- stop_spawned now unlinks the socket on exit, so a clean exit leaves no stale
  socket (a crash still can — the wait_ready fix handles that too).

Verified: two-launch repro no longer hangs; a crash-left stale socket recovers
in ~460ms. 10 e2e specs green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 heph.nvim/lua/heph/daemon.lua | 42 ++++++++++++++++++++++-------------
 1 file changed, 26 insertions(+), 16 deletions(-)

diff --git a/heph.nvim/lua/heph/daemon.lua b/heph.nvim/lua/heph/daemon.lua
index 7e4968f..9792b92 100644
--- a/heph.nvim/lua/heph/daemon.lua
+++ b/heph.nvim/lua/heph/daemon.lua
@@ -38,25 +38,30 @@ end
 
 --- Wait until `socket` both exists and accepts a real RPC (`health`). The
 --- existence check alone races the daemon's bind→accept, so we prove liveness
---- with a round-trip on a throwaway session. Returns `true`, or `false, reason`.
+--- with a round-trip. Returns `true`, or `false, reason`.
+---
+--- The probe runs in a **plain Lua loop**, never inside a `vim.wait` predicate:
+--- the rpc round-trip itself uses `vim.wait`, and nesting `vim.wait` inside
+--- another `vim.wait`'s predicate deadlocks Neovim (a stale socket made the
+--- inner connect-wait re-enter and hang).
 function M.wait_ready(socket, timeout)
   timeout = timeout or 5000
-  if not vim.wait(timeout, function()
-    return uv.fs_stat(socket) ~= nil
-  end, 20) then
-    return false, "socket never appeared: " .. socket
+  local rpc = require("heph.rpc")
+  local deadline = uv.hrtime() + timeout * 1e6 -- ns
+  while uv.hrtime() < deadline do
+    if uv.fs_stat(socket) ~= nil then
+      local session = rpc.new_session(socket)
+      local ok = pcall(function()
+        session:call("health", vim.empty_dict(), { timeout = 200 })
+      end)
+      session:close()
+      if ok then
+        return true
+      end
+    end
+    vim.wait(50) -- yield ~50ms; no predicate, so not nested
   end
-  local session = require("heph.rpc").new_session(socket)
-  local ok = vim.wait(timeout, function()
-    return pcall(function()
-      session:call("health", vim.empty_dict(), { timeout = 200 })
-    end)
-  end, 50)
-  session:close()
-  if not ok then
-    return false, "socket present but not accepting rpc: " .. socket
-  end
-  return true
+  return false, "daemon not ready at " .. socket
 end
 
 --- Ensure a daemon is reachable at `opts.socket`. If one is already serving the
@@ -124,6 +129,11 @@ function M.stop_spawned()
       m.handle:close()
     end
   end)
+  -- hephd doesn't unlink its socket on SIGTERM; remove it so the next launch
+  -- doesn't probe a stale socket. (A crash still leaves one — wait_ready copes.)
+  pcall(function()
+    uv.fs_unlink(m.socket)
+  end)
 end
 
 return M