Am 04.07.2011 23:03, schrieb Antoine Pitrou:

Le lundi 04 juillet 2011 à 22:51 +0200, Dirkjan Ochtman a écrit :

...

On Mon, Jul 4, 2011 at 21:01, Antoine Pitrou <solipsis@pitrou.net> wrote:

...

I've upgraded the Mercurial version on hg.python.org to 1.9. Also, I've enabled the experimental "generaldelta" repository feature on the server, which makes the repository smaller (down from ~350 MB to ~190 MB), regardless of the client version.

Did you ask the hg guys about the generaldelta thing? IIRC it's actually experimental, which I wouldn't use on something like the production repository python devs use.

From my discussions with them the core functionality was pretty much stable. I had tried it locally a couple of times when 1.9 was still in development. I guess we should run "hg verify" on the server from time to time, anyway.

Also, it doesn't imply anything on the client side, where generaldelta is still disabled by default.

Which means, translated for the less Mercurial-savvy, that we can restore the python.org repo by copying any of the developers' local clones to the server, should use of this new format indeed corrupt the repo.

Georg

Am 05.07.2011 04:08, schrieb Jesus Cea:

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

On 04/07/11 21:01, Antoine Pitrou wrote:

...

I've upgraded the Mercurial version on hg.python.org to 1.9. Also, I've enabled the experimental "generaldelta" repository feature on the server, which makes the repository smaller (down from ~350 MB to ~190 MB), regardless of the client version.

"generaldelta" is marked as "experimental". Is it wise to activate?.

In fact Mercurial 1.9 dropped "experimental" 'parentdelta', so "experimental" actually is... quite experimental and not garanteed:).

parentdelta was removed in favor of generaldelta... and you can't really get *more* general :)

But even if it was removed, you can always (using clone --pull locally) create a clone that doesn't use that format before you upgrade to the hypothetical version with generaldelta removed.

And yes, I have read your posts to HG development mailinglist in May, and I have big hopes for "generaldelta" or something close to it. But I will wait until is "official" (luckily activate per default for new repositories/network clones).

See my other reply to Antoine.

Georg

My question, actually, is why to hurry using "generaldelta". We could wait three more months until it is declared "official". Maybe Antoine idea, that would make sense, is python.org being a tester to help to actually test "generaldelta" to qualify for production.

The prime motivation is that it decreases the size of a clone from 350 to 190 MB, roughly (and presumably also the download size). The fact that it can help the hg devs validate generaldelta on a widely-used setup is an added benefit, but that couldn't have been the sole motivation.

I am fine with that, if somebody with access takes care to do a "hg verify" frequently :).

As a side note, if there's a problem, the buildbots should tell us very quickly (since they pull from the repo).

Regards

Antoine.