Bug #17493: Calling Reline::Unicode.calculate_width appears to lock up when called from emacs or if output is redirected

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-darwin17]
• Backport: 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN

[rsharman (Richard Sharman)]

In https://github.com/ruby/reline/pull/224 there is a test program

require "reline"

p Reline::Unicode.calculate_width("√")

as a test case for bug #17405. This program hangs when called from an emacs shell.

The problem is that cursor_pos in reline/ansi.rb outputs ESC [ 6 n and expects to be able read in raw mode the cursor position. This doesn't work when ruby is invoked from a shell inside emacs.

--> #1 /Users/richard/.rbenv/versions/3.0/lib/ruby/3.0.0/reline/ansi.rb:175:in `cursor_pos'
    #2 /Users/richard/.rbenv/versions/3.0/lib/ruby/3.0.0/reline.rb:373:in `may_req_ambiguous_char_width'
    #3 /Users/richard/.rbenv/versions/3.0/lib/ruby/3.0.0/reline.rb:359:in `ambiguous_width'
    #4 /Users/richard/.rbenv/versions/3.0/lib/ruby/3.0.0/forwardable.rb:238:in `ambiguous_width'
    #5 /Users/richard/.rbenv/versions/3.0/lib/ruby/3.0.0/reline/unicode.rb:115:in `get_mbchar_width'
    #6 /Users/richard/.rbenv/versions/3.0/lib/ruby/3.0.0/reline/unicode.rb:143:in `calculate_width'
    #7 /Users/richard/rubystuff/bugs/cursor_pos/test.rb:4

Even when run from an actual terminal, a similar problem occurs if stdout is redirected:

$ ruby test2.rb > T

The program appears to have locked up at this point; pressing Control-C will not terminate it.

Is it valid for a routine that calculates a character's width to assume that stdin is actually an ANSI sequence compatible terminal?

(This also occurs with ruby version 2.7.2p137.)

[rsharman (Richard Sharman)]

• File Patch Patch added

Another way the problem can happen is displaying an EN DASH character, such as in a CSV file produced by LibreOffice.

When a shell is running inside emacs, including the case of running inf-ruby, ANSI sequences are not honoured. In most cases this is harmless; however, in routine cursor_pos it causes problems because it expects to read back characters indicating the cursor position.

A simple fix is to disable this by setting Reline::IOGate to Reline::GeneralIO if the environment variable TERM is set to "dumb", as it in inside emacs; see file Patch. (This supersedes the previous Patch file.)

[jeremyevans0 (Jeremy Evans)]

The provided patch appears to only affect Windows, but I confirmed the problem also happens on OpenBSD. Presumably it happens on other operating systems too. I added a similar fix of using GeneralIO for the non-Windows case, and submitted a pull request to reline for it: https://github.com/ruby/reline/pull/285

[jeremyevans0 (Jeremy Evans)]

• Status changed from Open to Closed