Allow the console to contain "Error" (or even a backtrace) on startup

[Hardtack]

Robe checks the initialization of subprocess with text of subprocess' stdout. It checks the existence of word that ends with "Error". But some projects print initial message (before the robe starts), they can print the word "Error" to the stdout.

For example,

WARNING: Some ignorable error occurred during initialization.
#<SomeError cannot load optional module 'liboptional.so'>
... [stack trace]

[1] pry(main)> => "robe on 12345"

In this case, the robe was started normally in ruby process, but emacs-side's robe thinks that it was not started well.

Using external file like .robe-port would be possible solution

[dgutov]

Hi! Thanks for the report, but I'd expect fixing this will require some major changes. We're looking for the string Error\\> in the loop that waits until Robe is started in the current REPL buffer.

Its goal is to protect against looping infinitely when something goes wrong (see e.g. #9 (comment)). Most of the time, this approach works better than the obvious alternative (using timeouts).

Even if we create a file upon successful launch, that would happen after the error backtrace is printed to the REPL. So either we don't wait and declare the command as failed, or wait for... some?.. amount of time to see robe on.

The easiest workaround for you would be to launch the REPL and Robe server separately: first M-x inf-ruby-console-auto, and then M-x robe-start.

Going a bit more advanced route, we could try to process all the output coming from the REPL's startup first, before launching Robe, like:

diff --git a/robe.el b/robe.el
index dbe84f8..ddf6da0 100644
--- a/robe.el
+++ b/robe.el
@@ -120,7 +120,10 @@ project."
             (when (buffer-live-p ruby-buffer)
               (kill-buffer ruby-buffer))
             (inf-ruby-console-auto)
-            (set-window-configuration conf))
+            (set-window-configuration conf)
+            (while (not (buffer-local-value 'comint-last-prompt
+                                            (get-buffer inf-ruby-buffer)))
+              (accept-process-output (inf-ruby-proc))))
         (error "Aborted"))))
   (when (not robe-running)
     (let* ((proc (inf-ruby-proc))

The downside: if the REPL has a prompt that doesn't match inf-ruby-prompt-pattern, the command will hang forever.

[Hardtack]

Actually, I also tried to launch REPL and Robe separately, but it is very annoying.
I have to launch REPL & Robe separately in some projects, and I can use normally with robe-start in some project.
Since robe doesn't kill ruby REPL after catching Error\\>, reusing this REPL can be less-annoying solution. All I have to do is just call robe-start twice!
I'm currently doing that with following modification

diff --git a/a/lib/robe.rb b/b/lib/robe.rb
index 8208e3d..a1e254f 100644
--- a/a/lib/robe.rb
+++ b/b/lib/robe.rb
@@ -6,7 +6,7 @@ module Robe
     attr_accessor :server

     def start(port = 0)
-      return if @server
+      return "robe on #{@server.port}" if @server

       @server = Server.new(Sash.new, port)

Without this modification, emacs will hang forever when re-using ruby REPL created by previous robe-start.

[dgutov]

I'm currently doing that with following modification

Good idea, pushed.