bug#50470: 27.1; 'company-mode' 'eshell'

[Dmitry Gutov]

Hi Stefan,

On 23.01.2022 05:23, Stefan Monnier wrote:
> And the 100% untested patch below is a suggestion for how to try and fix
> those kinds of bugs.
> Can someone try and maybe make it work?

I've tried the patch, and it seems to work already, as well as fix this 
particular scenario. (Thanks!)

Might as well install it, I think.

There is a scenario that is more noticeably broken (yet actually better 
with this patch): a modification of bug#18951. Instead of

  ls *

try

  ls ~/Docu*

...and [on master] the result is that the asterisk is replaced with the 
"common part" of the possible completions automatically. If there is 
nothing to expand with, the asterisk is similarly deleted.

With your patch, we get the "Buffer is read-only" error in *Messages* 
instead, which is probably an improvement. Because it doesn't modify the 
input, nor break Company completions long-term (after the asterisk is 
removed).

The offending functions is pcomplete-parse-arguments. There is some 
complex global state going on there, but the following addition seems to 
fix the problem:

@@ -790,6 +804,9 @@ pcomplete-parse-arguments
                   (common-stub (car completions))
                   (c completions)
                   (len (length common-stub)))
+              (unless pcomplete-allow-modifications
+                (setq pcomplete-stub (buffer-substring begin (point)))
+                (throw 'pcomplete-completions completions))
              (while (and c (> len 0))
                (while (and (> len 0)
                            (not (string=


Not sure if this new value of pcomplete-stub is always TRT, but it has 
passed a bunch of my experiments successfully.