Fix client hand-off with server

1. use client instead of PID
2. server side couting
3. harding the reboot and retry

Author: ray-x

lua/copilot_agent/config.lua

@@ -183,6 +183,9 @@ local state = {
   input_statusline_managed = false, -- true when plugin populated the input window local statusline
   service_job_id = nil,
   service_process_pid = nil,
+  client_id = nil,
+  client_registered_base_url = nil,
+  client_lease_registered = false,
   service_launch_info = nil, -- human-readable launch mode/command shown in chat header
   service_starting = false,
   shutting_down = false, -- true during VimLeavePre cleanup; suppress reconnect/recovery noise on intentional shutdown

lua/copilot_agent/http.lua

@@ -17,12 +17,20 @@ local is_connection_error = utils.is_connection_error
 
 local M = {}
 
-local function effective_base_url()
+local function effective_base_url(base_url_override)
+  local override = normalize_base_url(base_url_override)
+  if type(override) == 'string' and override ~= '' then
+    return override
+  end
+
   if state.base_url_managed ~= false then
     local ok, service = pcall(require, 'copilot_agent.service')
     if ok and type(service) == 'table' and type(service.refresh_managed_base_url) == 'function' then
-      service.refresh_managed_base_url()
+      if service.refresh_managed_base_url() then
+        return normalize_base_url(state.config.base_url)
+      end
     end
+    return ''
   end
 
   local base_url = normalize_base_url(state.config.base_url)
@@ -42,20 +50,20 @@ local function effective_base_url()
   return base_url
 end
 
-function M.build_url(path)
-  return effective_base_url() .. path
+function M.build_url(path, base_url_override)
+  return effective_base_url(base_url_override) .. path
 end
 
 local function has_base_url()
   local base_url = effective_base_url()
   return type(base_url) == 'string' and base_url ~= ''
 end
 
-local function log_http_request(mode, method, path, body)
+local function log_http_request(mode, method, path, body, base_url_override)
   if not should_log(vim.log.levels.DEBUG) then
     return
   end
-  log(string.format('http.%s request method=%s path=%s url=%s body=%s', mode, tostring(method), tostring(path), M.build_url(path), serialize_log_value(body)), vim.log.levels.DEBUG)
+  log(string.format('http.%s request method=%s path=%s url=%s body=%s', mode, tostring(method), tostring(path), M.build_url(path, base_url_override), serialize_log_value(body)), vim.log.levels.DEBUG)
 end
 
 local function log_http_response(mode, method, path, status, exit_code, payload, err)
@@ -132,12 +140,13 @@ function M.encode_json(value)
 end
 
 -- Synchronous HTTP request via vim.system / vim.fn.system.
-function M.sync_request(method, path, body)
+function M.sync_request(method, path, body, opts)
+  opts = opts or {}
   if not M.ensure_curl() then
     return nil, 'curl executable not found: ' .. state.config.curl_bin
   end
 
-  log_http_request('sync', method, path, body)
+  log_http_request('sync', method, path, body, opts.base_url)
   local args = {
     state.config.curl_bin,
     '-sS',
@@ -147,7 +156,7 @@ function M.sync_request(method, path, body)
     '\n%{http_code}',
     '-X',
     method,
-    M.build_url(path),
+    M.build_url(path, opts.base_url),
     '-H',
     'Accept: application/json',
   }
@@ -216,13 +225,14 @@ function M.sync_request(method, path, body)
 end
 
 -- Async HTTP request via vim.fn.jobstart.
-function M.raw_request(method, path, body, callback)
+function M.raw_request(method, path, body, callback, opts)
+  opts = opts or {}
   if not M.ensure_curl() then
     callback(nil, 'curl executable not found: ' .. state.config.curl_bin)
     return
   end
 
-  log_http_request('async', method, path, body)
+  log_http_request('async', method, path, body, opts.base_url)
   local stdout = {}
   local stderr = {}
   local args = {
@@ -234,7 +244,7 @@ function M.raw_request(method, path, body, callback)
     '\n%{http_code}',
     '-X',
     method,
-    M.build_url(path),
+    M.build_url(path, opts.base_url),
     '-H',
     'Accept: application/json',
   }
@@ -313,12 +323,13 @@ function M.request(method, path, body, callback, opts)
 
     -- Lazy-require to avoid circular dependency with service.lua.
     local service = require('copilot_agent.service')
+    service.forget_service_addr()
     service.ensure_service_running(function(start_err)
       if start_err then
         callback(nil, start_err, nil)
         return
       end
-      M.raw_request(method, path, body, callback)
+      M.raw_request(method, path, body, callback, opts)
     end)
     return
   end
@@ -331,14 +342,15 @@ function M.request(method, path, body, callback, opts)
       )
       -- Lazy-require to break http↔service circular dependency.
       local service = require('copilot_agent.service')
+      service.forget_service_addr()
       service.ensure_service_running(function(start_err)
         if start_err then
           log(string.format('http.request retry failed method=%s path=%s startup_error=%s', tostring(method), tostring(path), serialize_log_value(start_err, { max_len = 600 })), vim.log.levels.WARN)
           callback(nil, err .. '\n' .. start_err, status)
           return
         end
         log(string.format('http.request retrying method=%s path=%s after service start', tostring(method), tostring(path)), vim.log.levels.DEBUG)
-        M.raw_request(method, path, body, callback)
+        M.raw_request(method, path, body, callback, opts)
       end)
       return
     end

lua/copilot_agent/init.lua

@@ -25,6 +25,7 @@ local cfg = require('copilot_agent.config')
 local defaults = cfg.defaults
 local state = cfg.state
 local notify = cfg.notify
+local log = cfg.log
 local normalize_base_url = cfg.normalize_base_url
 
 local service = require('copilot_agent.service')
@@ -144,6 +145,14 @@ function M.setup(opts)
   end
   state.config.base_url = normalize_base_url(state.config.base_url)
   state.service_launch_info = describe_service_launch()
+  state.service_job_id = nil
+  state.service_process_pid = nil
+  state.client_registered_base_url = nil
+  state.service_starting = false
+  state.service_addr_known = false
+  state.pending_service_callbacks = {}
+  state.service_output = {}
+  state.lsp_client_id = nil
   state.shutting_down = false
   -- Initialize runtime permission mode from config.
   state.permission_mode = state.config.permission_mode or 'interactive'
@@ -263,21 +272,46 @@ function M.setup(opts)
     end,
   })
   events.remember_open_buffer_disk_state()
-  service.register_client_lease()
   -- Clean up clipboard temp files if Neovim exits before they were sent.
   vim.api.nvim_create_autocmd('VimLeavePre', {
     group = vim.api.nvim_create_augroup('CopilotAgentCleanup', { clear = true }),
     callback = function()
+      log(string.format('VimLeavePre cleanup start pid=%d', vim.fn.getpid()), vim.log.levels.DEBUG)
       state.shutting_down = true
-      events.stop_event_stream()
-      local ok, logger = pcall(require, 'copilot_agent.log')
-      if ok and type(logger.flush_pending) == 'function' then
-        logger.flush_pending()
+      local ok, err = pcall(events.stop_event_stream)
+      if ok then
+        log('VimLeavePre stopped event stream', vim.log.levels.DEBUG)
+      else
+        log('VimLeavePre stop_event_stream failed: ' .. tostring(err), vim.log.levels.WARN)
+      end
+      local logger_ok, logger = pcall(require, 'copilot_agent.log')
+      local attachments_ok, attachments_err = pcall(session.discard_pending_attachments)
+      if attachments_ok then
+        log('VimLeavePre discarded pending attachments', vim.log.levels.DEBUG)
+      else
+        log('VimLeavePre discard_pending_attachments failed: ' .. tostring(attachments_err), vim.log.levels.WARN)
+      end
+      local unregister_ok, unregister_err = pcall(service.unregister_client)
+      if unregister_ok then
+        log('VimLeavePre unregistered client', vim.log.levels.DEBUG)
+      else
+        log('VimLeavePre unregister_client failed: ' .. tostring(unregister_err), vim.log.levels.WARN)
+      end
+      local stop_ok, stop_err = pcall(service.stop_service)
+      if stop_ok then
+        log('VimLeavePre stop_service completed', vim.log.levels.DEBUG)
+      else
+        log('VimLeavePre stop_service failed: ' .. tostring(stop_err), vim.log.levels.WARN)
+      end
+      if logger_ok and type(logger.flush_pending) == 'function' then
+        local flush_ok, flush_err = pcall(logger.flush_pending)
+        if flush_ok then
+          log('VimLeavePre flushed pending logs', vim.log.levels.DEBUG)
+          pcall(logger.flush_pending)
+        else
+          log('VimLeavePre flush_pending failed: ' .. tostring(flush_err), vim.log.levels.WARN)
+        end
       end
-      session.discard_pending_attachments()
-      service.unregister_client_lease()
-      service.maybe_shutdown_detached_service_if_last_client({ nonblocking = true })
-      service.stop_service()
     end,
   })
   vim.schedule(function()
@@ -465,6 +499,7 @@ function M.statusline_reasoning(max_len)
 end
 
 function M.status()
+  service.refresh_managed_base_url()
   local snapshot = service.debug_snapshot()
   local lines = {
     'service: ' .. normalize_base_url(state.config.base_url),
@@ -489,6 +524,7 @@ function M.status()
 end
 
 function M.debug_service()
+  service.refresh_managed_base_url()
   local snapshot = service.debug_snapshot()
   local lines = {
     'Service debug:',

lua/copilot_agent/lsp.lua

@@ -660,7 +660,8 @@ local function lsp_only_command(service_url)
   if type(cmd) ~= 'table' or vim.tbl_isempty(cmd) then
     return nil
   end
-  return vim.list_extend(cmd, { '-lsp-only', '--service-url', service_url })
+  vim.list_extend(cmd, { '-lsp-only', '--service-url', service_url })
+  return cmd
 end
 
 function M.paste_clipboard_image()

lua/copilot_agent/render.lua

@@ -3305,6 +3305,7 @@ function M.render_chat()
     state.entry_row_index = {}
 
     lines[#lines + 1] = state.config.chat.title
+    service.refresh_managed_base_url()
     local service_url = normalize_base_url(state.config.base_url)
     if service_url == '' then
       service_url = '<discovering via control socket>'