ivy-resume restarts at the beginning of counsel-ag and counsel-rg

[eflanigan00]

I type counsel-ag, type some characters then scroll down a few lines and pick a selection. When I invoke ivy-resume I get the same character search properly but the selected line starts back at the beginning of the selection list instead of resuming from my last selection.

[amosbird]

+1. It seems this is due to :dynamic-collection t.

[abo-abo]

This is indeed due to :dynamic-collection. When you resume, you can't be sure that the candidate that you selected before even exists. And then the candidates start coming it asynchronously. This is a problem: suppose you had selected candidate "foo", 101st on the list of 2000. When you resume, and the first chunk of 50 candidates comes in, what should be selected?

I'll think about making some improvement here, but it's not straightforward.

[neojski]

Not sure it's because of :dynamic-collection. When I resume SPC f f it doesn't seem to highlight the previous search either. @abo-abo, is it supposed to work or it's not supported?

[abo-abo]

@neojski counsel-find-file is supported, if that's what you mean.

[hisnawi]

Any possibility this issue gets resurrected?
Seems like a very important feature to have when the search results are in the hundreds.

[abo-abo]

@hisnawi The issue isn't closed. Anyone is free to contribute.

Just understand that this isn't an easy fix, at least as far as I can see. I estimate this issue at 1-2 hours of uninterrupted time and low priority; so it won't get done any time soon unless someone else steps up with a contribution.

[hisnawi]

Thanks for the response, I am a beginner when it comes to lisp otherwise I would have given it a go. But what I noticed is that helm already has this feature and it is very similar to ivy in terms of showing results dynamically. Wouldn't it be possible to leverage some of their code to integrate it in ivy?

It maybe a completely different issue in ivy I am just speculating.

[neojski]

@neojski counsel-find-file is supported, if that's what you mean.

I just tried with make plain version and it doesn't seem to work, i.e. I did:

1. C-x C-f C-n C-g (select ..)
2. M-: (ivy-resume) ret

Expected: .. is selected
Result: . is selected

[abo-abo]

@neojski Thanks, I can reproduce it now.

[neojski]

@abo-abo, you expected this to work. What part of ivy code was supposed to be responsible for this? If you pointed me to the right place in the source code I could try to fix it.

[abo-abo]

@neojski You can look at the source code of ivy-resume. The interesting part is:

:preselect (ivy-state-current ivy-last)

So there are two things to check:

• (ivy-state-current ivy-last) is the right value
• ivy-read using it does the right thing

[ericdanan]

In ivy-resume I don't see

:preselect (ivy-state-current ivy-last)

but

:preselect (unless (eq (ivy-state-collection ivy-last)
                       'read-file-name-internal)
             (regexp-quote ivy--current))

which seems to be there since commit 84e20a3 . So it seems to be on purpose that preselect is not set for counsel-find-file but I don't know why.

[abo-abo]

So it seems to be on purpose that preselect is not set for counsel-find-file but I don't know why.

I can't tell from the message as well. Maybe this has something to do with TRAMP. I think I'll just revert it to simplify the code.

[ericdanan]

I just tested and it works if ivy-state-current is a file but not if it is a directory. Maybe something going on in ivy-reset-state.

Neither does it work if I navigate to another directory before exiting with C-g, the directory I navigated to is not remembered. Perhaps ivy--cd should update ivy-state-directory.

Sorry that I can't investigate more at the moment.

[ericdanan]

Indeed I was able to make it work (I think) by modifying ivy--reset-state and ivy--cd. See #1508.

[neojski]

I thin this issue is not actually fixed for counsel-ag or counsel-rg.

[peterwvj]

I also experience this issue with counsel-ag.

[abo-abo]

@neojski @peterwvj I reopen this issue, since the preselect is not yest fixed for dynamic collections like counsel-ag.

[peterwvj]

Thanks a lot for working on this @abo-abo

So it seems like ivy-resumeis now able highlight the previously selected line, assuming that the search results produced using counsel-ag did not change in the meantime. Am I correct? This works very well assuming that you don't need to edit the search results.

For comparison helm-ag will by fault "freeze" the search results making it possible to resume to the previously highlighted line. Obviously this is easily done because helm-ag does not use a dynamic collection like counsel-ag does. I can see advantages to both approaches. However, I was wondering if it would be possible to configure counsel-ag to use a collection that enables a helm-ag-like behavior. I recently switched from helm to ivy so please bear with me if there's an obvious solution to this.

[abo-abo]

@peterwvj I don't understand what "freeze" means. Please open a new issue with a step-by-step scenario.

[peterwvj]

To clarify, helm-ag does not use dynamic collections. In consequence, the helm-ag search results are just produced ones, and not automatically updated when you call helm-resume.

So if you perform a search (by calling helm-ag), then navigate to the location of one your matches, edit that line, and call helm-resume then helm takes you back to the exact same line. This is always possible because the search results do not get updated. Not even if your edits would somehow "invalidate" the search results.

Does this make sense?

So my question is: would it be possible to easily configure counsel-ag to behave in this way?

[abo-abo]

So my question is: would it be possible to easily configure counsel-ag to behave in this way?

Yes. If you do C-c C-o (ivy-occur), your matches will be saved to a new buffer that represents the current search. You can have as many buffers as you like.
The results that they contain may become stale, which will make f go to a stale line number.
But you can refresh each buffer with g to make the results non-stale again.

[peterwvj]

Great, this is exactly what I need! Thanks a lot for explaining this!