Comments on: The Locality of Code Changes

By: Dazed & Confused » The cost and benefits of cleaning up code

Dazed & Confused » The cost and benefits of cleaning up code — Tue, 10 Nov 2009 14:57:59 +0000

[...] though it is not very scientifically conclusive Joakim Karlsson has some very interesting results on the locality of code changes. As Joakim discusses this relates closely to how you evaluate wether the code you are currently [...]

By: Weekend miscellany — The Endeavour

Weekend miscellany — The Endeavour — Sat, 04 Jul 2009 12:18:03 +0000

[...] Karlsson says in his post The Locality of Code Changes “The probability that you will change a piece of code in the near future increases when you [...]

By: Joakim

Joakim — Sat, 04 Jul 2009 07:24:12 +0000

Yes, I think I see your point. My data just shows that there is a distribution of waiting times between changes to code, but it fails to show anything about whether changes to dormant files really lead to a higher frequency of revisits. Right?

By: Brandon Moore

Brandon Moore — Fri, 03 Jul 2009 22:59:44 +0000

I don’t think that histogram is very conclusive. You get an exponential falloff if you plot a histogram of waiting times between events from a Poisson process. A scatter plot comparing successive commit intervals might show the correlation you are looking for.

By: Chris Sterling

Chris Sterling — Wed, 01 Jul 2009 23:57:12 +0000

Great information! Thank you for the research that you did and there seems to be something to the results you found.

By: Joakim

Joakim — Wed, 01 Jul 2009 19:51:35 +0000

@Brett, @Stefano:

Interesting tip. Thanks.

I also wonder if there is a way to display the distribution between what files ends up where in the histogram. Are we just seeing the same few files being changed over and over again, or is there a more interesting distribution?

By: Joakim

Joakim — Wed, 01 Jul 2009 19:33:13 +0000

@MSR:

Wow! I never heard of MSR before. Thanks for the info.

Are there any specific articles or other materials about this that you could point me at?

By: MSR

MSR — Wed, 01 Jul 2009 18:54:42 +0000

This kind of behaviour has been observed and described repeatedly at conferences like Mining Software Repositories, International Conference of Program Comprehension, PROMISE, ICSE, ICSM, etc.

Co-change is also observed. If you change two files together in the past, you will probably change those two files again in the future.

By: Brett L. Schuchert

Brett L. Schuchert — Wed, 01 Jul 2009 16:36:19 +0000

Stefano wrote:

…hystogram is related to the file sizes

I’d like to see a histogram for files < 50 lines, 50 – 100, 100 – 500, etc. (or some such grouping.

I suspect the size would strongly correlate with change.

By: Joakim

Joakim — Wed, 01 Jul 2009 13:00:18 +0000

@SuperlativeMan:
Thanks for the feedback.

I am not sure what you mean the difference between active development and maintenance is, though?

How do you think that would affect the interpretation of the data?