[darcs-users] add/record time

[Trevor Talbot]

> On Monday, Jul 28, 2003, at 03:48 US/Pacific, David Roundy wrote:
>
>> On Sun, Jul 27, 2003 at 07:29:47PM -0700, Trevor Talbot wrote:
>>> On Sunday, Jul 27, 2003, at 17:43 US/Pacific, David Roundy wrote:
>>>
>>>> There may be more than one slowdown.  I've found an O(n^2) function 
>>>> in pull, which is the part where it checks for conflicts in the 
>>>> merge, which is currently an O(n^2) process (but needn't be, except 
>>>> when there are lots of conflicts--and then it would be O(nm) where 
>>>> m is # of conflicts and n is number of patches).
>
>> In my test case (which is smaller than yours, so I won't have to wait 
>> as long, and thus closer to the usual situation), the resolution fix 
>> (which is now available) sped up my test pull from 2m30s to 1m40s, 
>> which is quite a nice improvement!
>>
>> I'd like to hear if you still have problems after this patch, since 
>> it looks from the profiler like there are no slupid slowdowns 
>> remaining... all the time seems to be spent where it ought to be.  If 
>> you still have trouble I may need to test with a larger patch, as 
>> there may be an O(n^2) bit of code with a tiny prefactor that I can't 
>> see with my moderate-sized patch.
>
> I was going to post when it finished, but it's been about 54 hours and 
> it hasn't finished yet, so :)

It finally died after 77 hours with an illegal instruction fault.  I've 
had this happen once before, with the record stuff, so it's probably 
related to gcc 3.3 or some other system oddity.  Oh well :)