bug#65318: 29.1.50; `comint-run' should read-string with specified history list

bug#65318: 29.1.50; `comint-run' should read-string with specified history list

[bruno cuconato]

Not really a bug, but a feature request. When I run `comint-run' I use
the history to find the switches I need, and often have to waddle
through a bunch of unrelated input. As I understand it, that is
because the interactive `read-string' call does not specify a history
list (and thus a global one is used, I assume.¹) I'm not sure what the
best solution is: do we want to add a history list just for the
switches, or do we want it to be shared with other commands (like
`shell-command'?), or would any change to the history list argument
break people's workflows? I for one would prefer the first option.

¹: An unrelated mini-bug: `read-string's docstring does not say what
 happens when the argument is nil, although maybe the manual might do
 so (I haven't checked). Obviously, one can simply check the C source.

bug#65318: 29.1.50; `comint-run' should read-string with specified history list

[Eli Zaretskii]

> From: bruno cuconato <bcclaro@gmail.com>
> Date: Tue, 15 Aug 2023 17:15:41 -0300
> 
> 
> Not really a bug, but a feature request. When I run `comint-run' I use
> the history to find the switches I need, and often have to waddle
> through a bunch of unrelated input. As I understand it, that is
> because the interactive `read-string' call does not specify a history
> list (and thus a global one is used, I assume.¹) I'm not sure what the
> best solution is: do we want to add a history list just for the
> switches, or do we want it to be shared with other commands (like
> `shell-command'?), or would any change to the history list argument
> break people's workflows? I for one would prefer the first option.

Since the best solution is not clear, I think we should provide a user
option that would name the history to use, and make it nil by default,
to preserve the current behavior.  Patches welcome.

> ¹: An unrelated mini-bug: `read-string's docstring does not say what
>  happens when the argument is nil, although maybe the manual might do
>  so (I haven't checked). Obviously, one can simply check the C source.

The doc string says to check the documentation of read-from-minibuffer
for the details of HISTORY, and read-from-minibuffer says that if the
argument is nil or omitted, it defaults to minibuffer-history.  So I
think we are okay in that department.

Thanks.

bug#65318: 29.1.50; `comint-run' should read-string with specified history list

[Juri Linkov]

> Not really a bug, but a feature request. When I run `comint-run' I use
> the history to find the switches I need, and often have to waddle
> through a bunch of unrelated input. As I understand it, that is
> because the interactive `read-string' call does not specify a history
> list (and thus a global one is used, I assume.¹) I'm not sure what the
> best solution is: do we want to add a history list just for the
> switches, or do we want it to be shared with other commands (like
> `shell-command'?), or would any change to the history list argument
> break people's workflows? I for one would prefer the first option.

Instead of adding new history variables with user options
to every use of 'read-string' such as in 'comint-run':
(read-string "Run program: ") and (read-string "Switches: "),
maybe better to add a single common option with a mapping
from a command and a prompt regexp to the history variable.

Then you could customize it to e.g.:

  (setq read-history-variables
   '((comint-run "Run program" shell-command-history)
     (comint-run "Switches" my-switches-history)))