When used inside Emacs with Reline, input is echoed

[dgutov]

Description

This issue is about using Pry as a Ruby REPL inside Emacs.

The effect looks very much like #616:

[1] pry(main)> [1] pry(main)> 2
[1] pry(main)> 2[1] pry(main)> 2=> 2
[2] pry(main)> [2] pry(main)> 3
[2] pry(main)> 3[2] pry(main)> 3=> 3
[3] pry(main)> [3] pry(main)> 

But there doesn't seem to be any interest in having this solved inside Reline (i.e. providing a "simple" io driver).

What are your expected behavior and actual behavior of your environment?

The expected behavior is that the input is not copied or otherwise reproduced in the output.

This problem come up more often now than Ruby 3.3 defaults to Reline.

Terminal Emulator

Comint inside Emacs.

IRB has a workaround: when launched with --nosingleline --nomultiline, it switches to a different I/O implementation, which it calls StdioInputMethod.

Should Pry carry its own solution like that as well? The hackish patch below seems to do the trick. It can be a configuration option, a separate wrapper class, etc.

@@ -179,7 +179,10 @@
           end
         end
 
-        if readline_available?
+        if ENV['TERM'] == 'dumb'
+          STDOUT.print(current_prompt)
+          STDIN.gets
+        elsif readline_available?
           set_readline_output
           input_readline(current_prompt, false) # false since we'll add it manually
         elsif coolline_available?

Downstream report: dgutov/robe#147 (comment)

[dgutov]

Sorry, wrong repo.

[freesteph]

@dgutov can you point the new one? I encounter these problems in Emacs and am also confused between irb, debug, reline, etc, etc, etc. Happy to follow your investigation.

[dgutov]

@freesteph See pry/pry#2303.

Irb has its own way of disabling Reline, so see the above and the linked PR for a similar solution for Pry.

Not sure about debug, haven't tested that yet.

Perhaps Reline itself should have a "dumb" backend with automatic switchover, or a fix for this some other way, but I don't understand the code enough to suggest anything in particular.

[dgutov]

Update for those coming from search results:

• The Irb workardound was merged in May (Switch to StdioInputMethod when TERM is 'dumb' irb#907), it's not perfect but better.
• The Pry workaround is in its master branch since July.
• Reline-native solution is tracked in Don't enable test mode by TERM=dumb #660 (see the triangle thing mentioned in the comments as one of sticking points).