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
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
|
Goal: Web interface to allow reverting of changes.
Interface:
At least at first, it will be exposed via the recentchanges
page, with revert icons next to each change. We may want a dynamic
per-page interface that goes back more than 100 changes later.
Limiting assumptions:
* No support for resolving conflicts in reverts; such a revert would just
fail and not happen.
* No support for reset-to-this-point; initially the interface would only
revert a single commit, and if a bunch needed to go, the user would have
to drive that one at a time.
Implementation plan:
* `rcs_revert` hook that takes a revision to revert.
* CGI: `do=revert&rev=foo`
* recentchanges plugin adds above to recentchanges page
* prompt user to confirm (to avoid spiders doing reverts),
check that user is allowed to make the change, commit reversion,
and refresh site.
Peter Gammie has done an initial implementation of the above.
[[!template id=gitbranch branch=peteg/revert author="[[users/peteg]]"]]
>> It is on a separate branch now. --[[users/peteg]]
> Review: --[[Joey]]
>
> The revert commit will not currently say what web user did the revert.
> This could be fixed by doing a --no-commit revert first and then using
> rcs_commit_staged.
>> Fixed, I think. --[[users/peteg]]
>
> So I see one thing I completly forgot about is `check_canedit`. Avoiding users
> using reverting to make changes they would normally not be allowed to do is
> tricky. I guess that a easy first pass would be to only let admins do it.
> That would be enough to get the feature out there..
>
> I'm thinking about having a `rcs_preprevert`. It would take a rev and look
> at what changes reverting it would entail, and return the same data
> structure that `rcs_recieve` does. This could be done by using `git revert
> --no-commit`, and then examining the changes, and then `git reset` to drop
> them.
>> We can use the existing `git_commit_info` with the patch ID - no need to touch the working directory. -- [[users/peteg]]
>
> Then the code that is currently in IkiWiki/Receive.pm, that calls
> `check_canedit` and `check_canremove` to test the change, can be
> straightforwardly refactored out, and used for checking reverts too.
>> Wow, that was easy. :-) -- [[users/peteg]]
>
> (The data from `rcs_preprevert` could also be used for a confirmation
> prompt -- it doesn't currently include enough info for diffs, but at
> least could have a list of changed files.)
>
> Note that it's possible for a git repo to have commits that modify wiki
> files in a subdir, and code files elsewhere. `rcs_preprevert` should
> detect changes outside the wiki dir, and fail, like `rcs_receive` does.
>> Taken care of by refactoring `rcs_receive` in `git.pm`
>> I've tested it lightly in my single-user setup. It's a little nasty that the `attachment` plugin
>> gets used to check whether attachments are allowed -- there really should be a hook for that.
>>> I agree, but have not figured out a way to make a hook work yet.
>>> --[[Joey]]
>>
>> Please look it over and tell me what else needs fixing... -- [[users/peteg]]
>>> I have made my own revert branch and put a few^Wseveral fixes in there.
>>> All merged to master now! --[[Joey]]
[[done]]
|