On Tue, Aug 16, 2011 at 12:27 PM, Pedro Giffuni <giffu...@tutopia.com> wrote: > On Tue, 16 Aug 2011 18:44:30 +0300, Daniel Shahaf <d...@daniel.shahaf.name> > wrote: > ... >>> >>> Perhaps I am missing something, but couldn't you just go ahead >>> and commit it directly in the server? >> >> No, the two copies Rob mentioned are the dump and the repository, but >> you could >> >> curl | gzip -dc | svnadmin load >> >> to avoid one of them. > > I meant that, given that we are not preserving any history a dump seems > overhead. Instead of having people downloading dumps, setting up their > own repositories and presumably have infra@ load it afterwards, we > should just commit it (assuming initial issues were fixed). >
Consider this to be a form of Review-Then-Commit (RTC). Those who wish to review the dump file are welcome to do so. Those who wish to object to the approach are welcome to do so take the lead on an alternative approach. But if I don't receive objections, then I'll assume lazy consensus and go forward with this. Since we are talking about 76 thousand files, it did not seem like the best place to apply CTR. I'm not certain whether I have permissions to load the dump file directly, without going through Apache Infra. I'd like to avoid doing this as a giant commit of 76K files because of difficulty of dealing with partial failures. I also want to avoid sending a 1.8 GB commit note to your inbox! I'm assuming that on the Infra side, they would first lock SVN from writes, do a dump of the existing AOOo repository, as a backup, then load my dump file. That way, if anything goes wrong, they can restart from a known state. > cheers, > > Pedro. >