bug#47012: xref copies keymap properties to minibuffer

[Dmitry Gutov <dgutov@yandex.ru>]

On 01.04.2021 21:45, Juri Linkov wrote:

>> Additionally, what do you think about toning down 'match''s background
>> color? Maybe use some subtler yellow like "lemon chiffon" or "khaki1"? Or
>> "light goldenrod".
> 
> Such toning down is welcome since currently match's background is too intense.
> Actually, I customized it long ago to "#ffff88" on one display,
> and to "#ffffbb" on another display.  I guess "#ffffbb" is too radical,
> but "#ffff88" should be fine and close to "khaki1" that is nicely looking as well.
> Another variant is to update gradually, i.e. start with "#ffff66",
> then after some time to "#ffff88".

I like either of these, just not #ffff66 (still too bright). Perhaps 
some darker variant would be a better option if being easy-to-notice 
becomes a problem (as Eli's response seems to indicate). Maybe we should 
ask others as well.

Anyway, let's hold off on this part of the change until further discussion.

>>>> Please try (setq xref-file-name-display 'project-relative).
>>> Thanks, I didn't know about this.  Shouldn't this be the default value
>>> since this is what's displayed by grep and ripgrep.
>>
>> I wouldn't mind, personally.
> 
> This is added to the patch below too.

LGTM.

>>> Actually, there is no exact option for what grep and ripgrep do,
>>> because they display file names relative to the search directory.
>>> But currently there is no xref option to display file names
>>> relative to the subdirectory specified by 'C-u C-x p g'.
>>
>> This issue is tricky because xref-find-definitions does not assume the
>> presence of a project, or even of any kind of containing directory. And
>> yet, it's handy to show its results with relative file names when possible,
>> too. So I picked "relative to the project" as the option value, and the
>> corresponding logic.
>>
>> I think what you're talking about is only a problem when the directory has
>> no containing project at all. In that case we could probably default to the
>> value of default-directory as the reference.
> 
> Maybe it would be nice to default to default-directory even when
> 'C-u C-x p g' is used in a project.

That's not the question, though. The question is whether we default to 
default-directory when outside of recognizable projects. Or how to 
differentiate 'C-u C-x p g' from other cases, such as xref-find-definitions.

If this is not urgent, let's leave it for a separate bug report.

> What is the real problem for me is that after navigating to
> a project's subdirectory (with e.g. dired) and typing 'C-u C-x p g',
> it doesn't provide the current directory as the default value.
> It inserts the project root by default, not its subdirectory:
> 
>    Base directory: /project/root/
> 
> whereas 'M-x rgrep' conveniently provides default-directory as default.

Makes sense, fixed in 4798dc0c51, please check it out.

> BTW, is it possible to make 'project-find-regexp' more compatible with 'rgrep'
> in other features too?  What is missing is a way to modify the constructed
> command line.  For example, often I need to add "-w" to the constructed command
> to match words only.  In 'C-u M-x rgrep', this is easy to do,
> but not in 'C-u C-x p g'.

That's not so easy to do: the exact command is concealed inside the 
helper function in another package (xref). I suppose it's possible to 
rearrange things such that command creation and its execution are two 
different phases, but TBH I wouldn't love the result. Though I agree it 
might be handy.

What I've been thinking we should have instead, is some kind of 
graphical prompt with multiple fields, where you can by default input 
the regexp and press RET, but you can also see the other options (like 
the file name glob to filter by, whether to search the "external roots" 
or not, whether to search only a particular directory, whether to ignore 
case, whether to search in the project-ignored files as well; options 
which modify the regexp or matching logic like your -w could be added too).

Note that several of the options enumerated above are not something we 
could expose in the "edit the command" interface, because the command 
gets the list of files from stdin.

Maybe it would be presented like one-line prompt where you can reach 
further fields using TAB, and maybe expand into some multiline pane 
(still inside the minibuffer) if some options can't fit on the same 
line, and you reach the end of that line by TAB-bing.

To sum up, if we managed to create some visual interface for specifying 
the options that project-find-regexp has control over, maybe it would 
both result in a less complex interaction between packages, as well as 
in a more powerful UI which more people will be happy with.