"make" completion is overwriting readline keybindings

[dsifford]

For whatever reason, the make completion is overwriting the globally set tab keybinding (which I have set to menu-complete).

This is easily reproducible. Check the example below.

Why could this be happening?

[scop]

No idea offhand. Is make completion the only thing this happens with?

[dsifford]

To the best of my knowledge, yes.

I haven't systematically tested this with any of the others, but I can say with some level of certainty that this is the first time that I've ever ran into this. I use a pretty wide array of tools that have associated completions on a frequent basis, so I'm somewhat certain that this issue stems from the make completion itself.

[bewuethr]

As far as I can tell, the completion for make is the only one that checks the COMP_TYPE variable, i.e., checks what type of completion is being attempted – maybe related? I don't see where the binding could be affected, though.

[dsifford]

Are either of you able to reproduce this on your end @scop @bewuethr ?

[bewuethr]

Yes, I can reproduce it; haven't tried to narrow it down, though.

[bewuethr]

I tried isolating the cause, but now (on a different machine) I am unable to reproduce. The binding isn't removed any longer. Can you reproduce with a minimal makefile, or is anything special about it?

[dsifford]

Nope, nothing special about the file.

I'm able to reproduce with the following Makefile on my machine:

.PHONY: foo
foo:
        @echo foo

.PHONY: bar
bar:
        @echo bar

$ uname -a
Linux archlinux 4.15.5-1-ARCH #1 SMP PREEMPT Thu Feb 22 22:15:20 UTC 2018 x86_64 GNU/Linux
$ bash --version
GNU bash, version 4.4.19(1)-release (x86_64-unknown-linux-gnu)
Copyright (C) 2016 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>

This is free software; you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.

[scop]

Can't reproduce using the given Makefile, current bash-completion master, and

$ uname -a
Linux uncleman 4.13.0-36-generic #40-Ubuntu SMP Fri Feb 16 20:07:48 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux
$ bash --version | head -n 1
GNU bash, version 4.4.12(1)-release (x86_64-pc-linux-gnu)

[bewuethr]

Hmmm, and now, on the same machine where I observed it yesterday, I can't see it any longer ☹️

[dsifford]

Well, shoot.

I was going to suggest that maybe this is an issue with bleeding edge bash since @scop is running 4.4.12 and I'm on 4.4.19, but now I'm not so sure.

What system/bash version were you running when you were able to trigger this, @bewuethr?

[dsifford]

Update: Just tested using the official docker bash:4 image and I can't reproduce there either. That image is also running 4.4.19....

bash-4.4# uname -a
Linux fd318dffa380 4.15.5-1-ARCH #1 SMP PREEMPT Thu Feb 22 22:15:20 UTC 2018 x86_64 Linux

The plot thickens....

[bewuethr]

I used Bash 4.4.18, I think, I have to check when I'm back at the machine.

Here's a way to narrow down the cause: if you set your PS4 to

PS4='$(bind -q menu-complete):${BASH_SOURCE##*/}:${LINENO}: '

and then attempt completion for make, you should be able to see exactly where the binding gets removed.

[dsifford]

@bewuethr Good idea!

I think I spotted it...

menu-complete can be invoked via "\C-i", "\C-n".:make:65: local mode=--
menu-complete can be invoked via "\C-i", "\C-n".:make:66: ((  COMP_TYPE != 9  ))
menu-complete can be invoked via "\C-i", "\C-n".:make:67: mode=-d
mmenu-complete can be invoked via "\C-i", "\C-n".:make:0: shopt -po posix
menu-complete can be invoked via "\C-i", "\C-n".:make:70: local 'reset=set +o posix'
menu-complete can be invoked via "\C-i", "\C-n".:make:70: set +o posix
menu-complete can be invoked via "\C-n".:make:74: COMPREPLY=($( LC_ALL=C             $1 -npq __BASH_MAKE_COMPLETION__=1                 "${makef[@]}" "${makef_dir[@]}" .DEFAULT 2>/dev/null |             command sed -nf <(_make_target_extract_script $mode "$cur") ))
mmenu-complete can be invoked via "\C-n".:make:0: command sed -nf /dev/fd/63
mmmenu-complete can be invoked via "\C-n".:make:0: _make_target_extract_script -d ''
mmenu-complete can be invoked via "\C-n".:make:0: LC_ALL=C
mmenu-complete can be invoked via "\C-n".:make:0: sed -nf /dev/fd/63
mmmenu-complete can be invoked via "\C-n".:make:1: local mode=-d

Looks like it's either the set +o posix call on line 70 or the the expression called when setting COMPREPLY on line 74 of the make completion file.

Interestingly, for whatever reason, the line numbers of those calls aren't actually 70 and 74, but 155 and 156, respectively.

Thoughts?

[bewuethr]

I can't figure out how to reproduce this separately; does it not happen if you remove set +o posix? If so, that'd feel like a Bash bug.