summaryrefslogtreecommitdiff
path: root/doc/bugs/feature_request:_addhash.mdwn
blob: fca74200ec8904760ccecd3702d89921fb1d0136 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
### Use case 1

I have a big repo using a SHA256E back-end. Along comes a new shiny SKEIN512E back-end and I would like to transition to using that, because it's faster and moves from ridiculously to ludicrously low risk of collisions.

I can set `.gitattributes` to use the new back-end for any new files added, but then I when I import some arbitrary mix of existing and new files to the repo it will not deduplicate any more, it will add all the files under the new hash scheme.

### Use case 2

I have a big repo of files I have added using `addurl --fast`. I download the files, and they are in the repo.

 - I cannot verify later that none of them have been damaged.
 - If I come across an offline collection of some of the files, I cannot easily get them into the annex by a simple import.

### Workaround

In both these cases, what I can do is unlock (maybe?) or unannex (definitely) all of the files, and then re-add them using the new hash. In both cases this means I now risk having duplicates in various clones of the repo, and would have to clean them up with `drop-unused` -- after first having re-copied them from a repo that has them under the new hash. I also lose the continuity of the history of that object.

In use case 2 I also lose the URLs of the files and would have to re-add them using `addurl`.

... which brings me to the proposed feature.

### addhash

Symmetrical to `addurl`, which in one form can take an existing hashed (or URL-sourced) file and add an URL source to it, `addhash` can take an existing URL-sourced (or hashed) file and add a hash reference to it (given that the file is in the annex so that the hash may be calculated) -- an alias under which it may also be identified, in addition to the existing URL or hash.

 - Any file added to the annex after `addhash` will use the symlink name of the original hash if their hash matches the `addhash`ed one.
 - An `fsck` run will use one of the available hashes to verify the integrity of the file, maybe according to some internal order of preference, or possibly a configurable one.