On Fri, Apr 23, 2010 at 07:07, Stefan Küng <tortoisesvn_at_gmail.com> wrote:
>...
>>> On 23.04.2010 00:06, Hyrum K. Wright wrote:
>>>>
>>>> With the increased integration of build tools and other notification
>>>> systems
>>>> which desire knowledge of commit activity, it would be useful to provide
>>>> an
>>>> easier mechanism of installing post-commit notification, without having
>>>> to
>>>> use the hook infrastructure. Imagine a user being able to set up commit
>>>> mails, CIA notifications, and buildbot notifications *without* having to
>>>> have access to the repository. To that end, I propose the following.
>>>>
>>>> Simply a versioned property, which, when encountered during the course
>>>> of
>>>> a
>>>> commit, causes the server to emit a notification. This property would
>>>> be
>>>> a
>>>> list of URLs, to which the repository would send a specially formatted
>>>> POST
>>>> with the information about the commit. (The idea being that a committer
>>>> to
>>>> the project could set up this property, as well as the server which
>>>> receives
>>>> these notifications, all without the intervention of the repository
>>>> administrator.) As part of the bubble up, the repo would queue these
>>>> URLs,
>>>> and then POST to them during the post-commit phase of the commit.
>>>>
>>>> Thoughts?
>...
> After re-reading the proposal: doesn't Google Code already have something
> like this implemented?
> http://code.google.com/p/support/wiki/PostCommitWebHooks
Yup!
Tho I can guarantee they have already isolated those outgoing requests
from their serving/internal networks and infrastructure. Would your
average IT administrator do the same? :-P
Regardless of the security stuff, I think your reference is quite
valid. It would be great for svn to provide a script to follow that
model/specification.
And I'll go one further: also provide a (sample) CGI script to
*accept* those POST requests, and "do something". At the ASF, our
website is updated after a commit. The code that does that today is
apparently a little flaky, so having an officially
supported/tested/maintained solution for that kind of scenario would
be in our best-interest, let alone helping umpteen million downstream
users.
Cheers,
-g
Received on 2010-04-23 13:18:01 CEST