annex init no longer generates default description

Please describe the problem.

As of a14f6ce75 (fix repo description setting bugs, 2019-05-23), git annex init without a description sets the description to a blank string. AFAICT from the description of that change and from the corresponding bug report, that isn't an intentional change.

What steps will reproduce the problem?

cd $(mktemp -dt gx-XXXXXXX)
git init
git annex init
git annex info --json | jq

On the parent of a14f6ce75, this shows

Initialized empty Git repository in /tmp/gx-BLmU7TJ/.git/
init  ok
(recording state in git...)
{
  "local annex size": "0 bytes",
  "size of annexed files in working tree": "0 bytes",
  "command": "info",
  "untrusted repositories": [],
  "success": true,
  "semitrusted repositories": [
    {
      "here": false,
      "uuid": "00000000-0000-0000-0000-000000000001",
      "description": "web"
    },
    {
      "here": false,
      "uuid": "00000000-0000-0000-0000-000000000002",
      "description": "bittorrent"
    },
    {
      "here": true,
      "uuid": "d1ef2ae8-9604-4c33-8c3c-181f81c4348f",
      "description": "kyle@hylob:/tmp/gx-BLmU7TJ"
    }
  ],
  "bloom filter size": "32 mebibytes (0% full)",
  "repository mode": "indirect",
  "backend usage": {},
  "transfers in progress": [],
  "local annex keys": 0,
  "available local disk space": "301.33 gigabytes (+1 megabyte reserved)",
  "annexed files in working tree": 0,
  "file": null,
  "trusted repositories": []
}

On a14f6ce75, this shows

Initialized empty Git repository in /tmp/gx-pK345zF/.git/
init  ok
{
  "local annex size": "0 bytes",
  "size of annexed files in working tree": "0 bytes",
  "command": "info",
  "untrusted repositories": [],
  "success": true,
  "semitrusted repositories": [
    {
      "here": false,
      "uuid": "00000000-0000-0000-0000-000000000001",
      "description": "web"
    },
    {
      "here": false,
      "uuid": "00000000-0000-0000-0000-000000000002",
      "description": "bittorrent"
    },
    {
      "here": true,
      "uuid": "00b8da69-41a5-4de8-9a6a-9ed991289f81",
      "description": ""
    }
  ],
  "bloom filter size": "32 mebibytes (0% full)",
  "repository mode": "indirect",
  "backend usage": {},
  "transfers in progress": [],
  "local annex keys": 0,
  "available local disk space": "301.22 gigabytes (+1 megabyte reserved)",
  "annexed files in working tree": 0,
  "file": null,
  "trusted repositories": []
}

Note that the description for here is a blank string.

fixed --Joey

RSS Atom

uuid.log is also not created

As of that same commit (a14f6ce75), running git annex init in a fresh repository no longer creates uuid.log:

% cd $(mktemp -dt gx-XXXXXXX)            
% git init
Initialized empty Git repository in /tmp/gx-kpnGSjg/.git/
% git annex init
init  ok
% git ls-tree -r git-annex  ;# no output

Comment by kyle — Wed Jun 5 15:18:30 2019

Remove comment

Issue with description cache?

I took another look at this thinking that maybe I could spot something obvious in the condition added in a14f6ce75

    {- Avoid overwriting existing description with a default
     - description. -}
    whenM (pure (isJust mdescription) <||> not . M.member u <$> uuidDescMap) $
        describeUUID u =<< genDescription mdescription

but that looks fine. Grepping around for uuidDescMap, I spotted a couple of places where the cache needed to be invalidated. Blindly applying that same approach I did

diff --git a/Annex/Init.hs b/Annex/Init.hs
index cb7f8905e..6303103df 100644
--- a/Annex/Init.hs
+++ b/Annex/Init.hs
@@ -76,6 +76,7 @@ genDescription Nothing = do

 initialize :: Maybe String -> Maybe RepoVersion -> Annex ()
 initialize mdescription mversion = checkCanInitialize $ do
+   uuidDescMapLoad
    {- Has to come before any commits are made as the shared
     - clone heuristic expects no local objects. -}
    sharedclone <- checkSharedClone

That fixes the problem on my end (i.e., git annex init sets the description to the default user@host:dir description) while still addressing the original issue fixed by a14f6ce75.

So it seems like it has something to do with the uuidDescMap cache, but I haven't been able to figure out what in particular.

Comment by kyle — Mon Jun 17 21:11:14 2019

Remove comment

comment 3

Thanks Kyle, that does point at the problem. In uuidDescMapLoad it implicitly adds the UUID of the current repo, with an empty description if none is set. So, the added check of uuidDescMap always finds the UUID in it already. Your uuidDescMapLoad hack avoids this by loading the map before the current repo has a UUID.

Fixed more cleanly..

Comment by joey — Fri Jun 21 00:12:51 2019

Remove comment

thanks

Thank you for the explanation and fix!

Comment by kyle — Fri Jun 21 02:05:24 2019

Remove comment

bug fix release

@joeyh Thanks for fixing this. Would it be possible to make a new release on Hackage? I'd like to make a new conda release that incorporates the improvements in version 7.20190615, but don't want to make a release that fails DataLad tests the default one. Thanks!