The importance of patch pools

[John A Meinel]

Kevin Smith wrote:
> Matthieu Moy wrote:
>
>> Kevin Smith <yarcs at qualitycode.com> writes:
>>
>>> I read that whole thread, but it wasn't clear to me that centralized
>>> storage would still be possible with weaves. If it is, that's great
>>> news.
>>
>>
>>
>> Re-read
>>
>>
>> http://article.gmane.org/gmane.comp.version-control.bazaar-ng.general/2098
>>
>
>
> I'm only responding in the hopes of salvaging my reputation as one who
> does RTFM (or RTF Thread) before posting a question. Hopefully the
> thread will die here.
>
> Even after re-reading that message (and the wiki page it links to), it
> really doesn't say whether centralized storage would be possible with
> weaves. Really.

I agree, it isn't very clear from that message. Though the message does
discuss having weaves and centralized storage at the same time.

>
> Fortunately, I got a direct answer ("yes") in Aaron's reply, so whether
> or not the above thread potentially answered my question is moot.

Basically, with a weave, each time you add a new full-text it gets
associated with some identifier. In the current code, these are simple
integers per file, which also has a revision-id => simple integer map
(though this second part is not fully implemented :), hence why we
aren't using weaves yet.

What I don't know, though, is how cheap it will be to get a specific
revision using a remote weave. You can certainly pull across the entire
file, but it would be nice if you didn't have to pull everything when
you already have half (or more) of it.

And if you are using a dumb server (sftp/http), you can't have the
remote end do any work.

For myself, a weave is fine, as long as the above is handled, and the
weave format ends up append-only. I really like arch's *don't mess with
history* policy, so that as long as it worked in the past, it should
still work today, even if today's tools screwed something up. (compare
with CVS where because it stores current + back-patches, if you modify
the current manually, you potentially just destroyed all of the history,
since the back-patches may not apply).

I think monotone might suffer from something similar, though I'm not
positive. I know they have hashes so they can *detect* when something
goes wrong, but history might still be lost. (I'm not an expert on
monotone, so they might have something else to make it okay).

I really liked the revfile format, since it was append only, and because
each commit was stored in a particular range, with an index to look it
up, if I had revisions 1-50, it should be a reasonably small download to
get revision 51. I'm sure something will get worked out, though.

John
=:->


>
> Thanks,
>
> Kevin
>
>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 253 bytes
Desc: OpenPGP digital signature
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20050823/890f9f75/attachment.pgp