Joey-on-the-change-of-FRDR-end-point

> I would like to ask you a question about git annex way to manage unused data. I
> built a git annex special remote using Globus.org as a datastore and I am
> setting up a new datalad dataset retrieving files content from Globus.
>
> So I set the files keys present with 1 and register the related urls as
> expected. Now, the input parameter (prefix) to initremote was wrong, hence my
> registered urls were all wrong, so I run 'git annex enableremote [new_prefix]'.

Did you use "git-annex registerurl" for that, or did the external
special remote use the SETURLPRESENT message?

The only way that initremote/enableremote would affect the url is if a
special remote does use SETURLPRESENT.

If the special remote does use SETURLPRESENT, there is also a SETURLMISSING
that it could use to remove the urls you want to remove. But, that might
not be the best fit for your situation, I think.

There is the `git-annex rmurl` command, that can remove an url from an
annexed file. Probably what you want, I think.

> This is all fine, git annex updates the prefix and builds new, now correct,
> urls. As you can imagine at this point, git annex has 2 urls per file (one
> wrong and one correct) and does not know what to do and fails.

Hmm, normally when git-annex has more than one url for a single file, it
will try all the urls, and only fails if all of the urls are not
accessible. So, I'm curious what this failure looks like.

> Admittedly it would be useful to have a command that clears content of internal
> git annex files (in the git annex branch) and provides a clean uninitialized
> repository without needing to set to 0 or 1 the whole previous history. For
> example, I have the case that the datasets in globus move to different
> locations and git annex enableremote must be re-run with the new updated
> parameters, and everything should work as expected. So I am thinking of
> something like 'git annex enablremote [new_parameter] --override'

Hmm, what data in the git-annex branch would that remove, would it be
all the data about that remote? Including removing information about
what keys are stored in the remote?

There is a way to do that, it's just to run `git-annex dead` with the
name or uuid of the remote, and then `git-annex initremote` a new
remote name.

But if you don't want to remove the information about what keys are
stored in the remote, using `git-annex rmurl` seems like a better
approach.

Would it help to be able to run a pipeline like this to find and delete
them all?

git-annex whereis --format '${file}\0${url}' | grep globus://wrong_prefix/ | git-annex rmurl --batch -z

(Not yet implemented but something like that could be added.)

Start by comparing the issue's proposed override workflow with the mentioned git-annex commands: rmurl, whereis, dead, initremote, and enableremote. Decide whether the requested behavior should remove URL records or all metadata for the remote; done means the intended cleanup or reconfiguration is unambiguous and works for the wrong Globus prefix case.

Written by the indexing model from the issue text.