On 08.02.2011 23:57, Branko Čibej wrote:
> On 08.02.2011 23:47, Stefan Sperling wrote:
>> On Tue, Feb 08, 2011 at 11:37:16PM +0100, Branko Čibej wrote:
>>> On 08.02.2011 18:13, Hyrum K Wright wrote:
>>>> It shouldn't be difficult to create a large working copy ('svn co
>>>> ^/subversion') and use that for testing, but getting more insight than
>>>> just our own dataset would be very nice.
>>> Except that, as I noted elsewhere in this thread, even 'svn co
>>> ^/subversion/tags' get the (current trunk) svn process up to 2.5 gigs
>>> before the whole thing falls down, and it takes forever to do that. I'd
>>> say that's another release blocker.
>> Is this perchance due to the known serf memory leaks which were
>> fixed in serf 0.7.1?
> Well, I just now updated my trunk working copy and had to go install a
> newer serf to get it to build (and incidentally ran afoul of the diff
> bug mentioned elsewhere, it truncated a couple files, most unpleasant).
>
> I'll try a smaller checkout (just tags/ebcdic, which also failed by
> itself before) and report.Using Serf 0.7.1 definitely helps, the checkout of ^/subversion/tags/ebcdic was now successful, took 19 minutes and got svn memory usage up to 360MB. I'm not too happy about the last two points, but at least it works. -- Brane
