What was the version of bzr and git? Was this windows or unix? I suspect its unix only. It would be good to see the numbers on both platforms. Is this a limited rev history or the entire history? You also don't mentioned the exact commands you run. Do you use n=0 or n=1 with bzr? I would normally use n=1 as n=1 is not of much interest. Such limited microbenchmarks merely spread FUD.
Also, while choosing a DVCS, the common operations have to be fast not all operations. So the interesting benchmark would provide numbers for all the common operation. Just running one operation can be misleading. I do much more with my DVCS than just run the log operation :-)
Also, I don't see myself doing a log of a full 20+ year history multiple times during the day (In fact I have never done that). The commands I use normally, are fast enough large repos with bzr (emacs included).
You know, I had bzr 1.17 or similar. I upgraded to 2.1 and got the same time (slightly over 25 seconds). git version is 1.7.0.3. Both on my mac (I don't do Windows). In both cases, it was whatever "<cmd> clone <url>" does by default. So pretty much, all defaults all around.
I mentioned this because someone else was saying it was as fast. I don't know anything about the n= things... I just ran "bzr log > /dev/null" and "git log > /dev/null" (my shell automatically times all commands).
I tend to use and search logs a lot in my projects. With 2s to generate a log output, you won't think about the cost of looking through the whole thing for something (I did similar in an active hg project earlier today).
I do get your point. It's not a... holistic benchmark. In another thread, someone called fowl on the article discussing size vs. bytes transferred. I thought it'd be better to think of that as time. Cloning roughly the same content via bzr and git (memcached master branch -- which is cloned at lp and contains just about the same history, though is somehow smaller at lp) takes a bit over 20 seconds and a bit under 4 seconds respectively.
I suppose it's valid to argue that you don't clone much, either. At some point, one would wonder whether you avoid certain things because they're slow.
Comments
> When I tried using it, it was painfully slow, taking 20-30 seconds for a simple 'bzr log'.
When did you use it? On what history size? bzr 2.x series is quite fast.
Here's the difference between
bzr log on my clone of emacs:
git log on my clone of emacs:What was the version of bzr and git? Was this windows or unix? I suspect its unix only. It would be good to see the numbers on both platforms. Is this a limited rev history or the entire history? You also don't mentioned the exact commands you run. Do you use n=0 or n=1 with bzr? I would normally use n=1 as n=1 is not of much interest. Such limited microbenchmarks merely spread FUD.
Also, while choosing a DVCS, the common operations have to be fast not all operations. So the interesting benchmark would provide numbers for all the common operation. Just running one operation can be misleading. I do much more with my DVCS than just run the log operation :-)
Also, I don't see myself doing a log of a full 20+ year history multiple times during the day (In fact I have never done that). The commands I use normally, are fast enough large repos with bzr (emacs included).
You know, I had bzr 1.17 or similar. I upgraded to 2.1 and got the same time (slightly over 25 seconds). git version is 1.7.0.3. Both on my mac (I don't do Windows). In both cases, it was whatever "<cmd> clone <url>" does by default. So pretty much, all defaults all around.
I mentioned this because someone else was saying it was as fast. I don't know anything about the n= things... I just ran "bzr log > /dev/null" and "git log > /dev/null" (my shell automatically times all commands).
I tend to use and search logs a lot in my projects. With 2s to generate a log output, you won't think about the cost of looking through the whole thing for something (I did similar in an active hg project earlier today).
I do get your point. It's not a... holistic benchmark. In another thread, someone called fowl on the article discussing size vs. bytes transferred. I thought it'd be better to think of that as time. Cloning roughly the same content via bzr and git (memcached master branch -- which is cloned at lp and contains just about the same history, though is somehow smaller at lp) takes a bit over 20 seconds and a bit under 4 seconds respectively.
I suppose it's valid to argue that you don't clone much, either. At some point, one would wonder whether you avoid certain things because they're slow.