feat(core): improve troubleshooter behavior on critical errors

espanso/src/cli/daemon/mod.rs

@@ -90,20 +90,26 @@ fn daemon_main(args: CliModuleArgs) -> i32 {
   // at a given time.
   let mut _current_troubleshoot_guard = None;
 
-  let config_store = match troubleshoot::load_config_or_troubleshoot(&paths, &paths_overrides) {
-    troubleshoot::LoadResult::Correct(load_result) => load_result.config_store,
-    troubleshoot::LoadResult::Warning(load_result, guard) => {
-      _current_troubleshoot_guard = guard;
-      load_result.config_store
-    }
-    troubleshoot::LoadResult::Fatal(mut guard) => {
-      guard
-        .wait()
-        .expect("unable to wait for troubleshooting guard");
-      error!("critical error while loading config");
-      return DAEMON_FATAL_CONFIG_ERROR;
-    }
-  };
+  let (watcher_notify, watcher_signal) = unbounded::<()>();
+
+  watcher::initialize_and_spawn(&paths.config, watcher_notify)
+    .expect("unable to initialize config watcher thread");
+
+  let config_store =
+    match troubleshoot::load_config_or_troubleshoot_until_config_is_correct_or_abort(
+      &paths,
+      &paths_overrides,
+      watcher_signal.clone(),
+    ) {
+      Ok((result, guard)) => {
+        _current_troubleshoot_guard = guard;
+        result.config_store
+      }
+      Err(err) => {
+        error!("critical error while loading config: {}", err);
+        return DAEMON_FATAL_CONFIG_ERROR;
+      }
+    };
 
   info!("espanso version: {}", VERSION);
   // TODO: print os system and version? (with os_info crate)
@@ -126,13 +132,6 @@ fn daemon_main(args: CliModuleArgs) -> i32 {
   ipc::initialize_and_spawn(&paths.runtime, exit_notify.clone())
     .expect("unable to initialize ipc server for daemon");
 
-  let (watcher_notify, watcher_signal) = unbounded::<()>();
-
-  if config_store.default().auto_restart() {
-    watcher::initialize_and_spawn(&paths.config, watcher_notify)
-      .expect("unable to initialize config watcher thread");
-  }
-
   if cli_args.is_present("show-welcome") {
     ui::show_welcome_screen(&preferences);
   }
@@ -140,6 +139,10 @@ fn daemon_main(args: CliModuleArgs) -> i32 {
   loop {
     select! {
       recv(watcher_signal) -> _ => {
+        if !config_store.default().auto_restart() {
+          continue;
+        }
+
         info!("configuration change detected, restarting worker process...");
 
         // Before killing the previous worker, we make sure there is no fatal error

espanso/src/cli/daemon/troubleshoot.rs

@@ -18,9 +18,13 @@
  */
 
 use std::process::{Child, Command};
+use std::time::Duration;
 
-use anyhow::{Result};
+use anyhow::{bail, Result};
+use crossbeam::channel::Receiver;
+use crossbeam::select;
 use espanso_path::Paths;
+use log::info;
 
 use crate::cli::util::CommandExt;
 use crate::cli::PathsOverrides;
@@ -47,10 +51,15 @@ impl TroubleshootGuard {
   pub fn new(child: Child) -> Self {
     Self { child }
   }
+  #[allow(dead_code)]
   pub fn wait(&mut self) -> Result<()> {
     self.child.wait()?;
     Ok(())
   }
+  pub fn try_wait(&mut self) -> Result<bool> {
+    let result = self.child.try_wait()?;
+    Ok(result.is_some())
+  }
 }
 
 impl Drop for TroubleshootGuard {
@@ -97,3 +106,44 @@ pub fn load_config_or_troubleshoot(paths: &Paths, paths_overrides: &PathsOverrid
     }
   }
 }
+
+pub fn load_config_or_troubleshoot_until_config_is_correct_or_abort(
+  paths: &Paths,
+  paths_overrides: &PathsOverrides,
+  watcher_receiver: Receiver<()>,
+) -> Result<(ConfigLoadResult, Option<TroubleshootGuard>)> {
+  let mut _troubleshoot_guard = None;
+
+  loop {
+    // If the loading process is fatal, we keep showing the troubleshooter until
+    // either the config is correct or the user aborts by closing the troubleshooter
+    _troubleshoot_guard = match load_config_or_troubleshoot(paths, paths_overrides) {
+      LoadResult::Correct(result) => return Ok((result, None)),
+      LoadResult::Warning(result, guard) => return Ok((result, guard)),
+      LoadResult::Fatal(guard) => Some(guard),
+    };
+
+    loop {
+      select! {
+        recv(watcher_receiver) -> _ => {
+          info!("config change detected, reloading configs...");
+
+          break
+        },
+        default(Duration::from_millis(500)) => {
+          if let Some(guard) = &mut _troubleshoot_guard {
+            if let Ok(ended) = guard.try_wait() {
+              if ended {
+                bail!("user aborted troubleshooter");
+              }
+            } else {
+              bail!("unable to wait for troubleshooter");
+            }
+          } else {
+            bail!("no troubleshoot guard found");
+          }
+        }
+      }
+    }
+  }
+}