Statistics from the 3.19 development cycle

As of this writing, the 3.19-rc7 prepatch is out and things look on track for the final 3.19 release in the near future. So it seems about time for a look at the changes that went into this release and where they came from. Once again, it has been a busy development cycle with few real surprises.

There have been 12,461 changesets merged into the mainline for the 3.19 release. That is up a bit from 3.18 and fits well with the general trend toward larger changeset counts in each release, as can be seen in this simple plot:

If the final 3.19 release does happen on February 8, as seems likely, the development cycle will have been 63 days long — the third cycle in a row of that length. Development cycle length has been decreasing for some time, but, at this point, we may be close to the minimum time it takes to produce a production-quality kernel release.

The changes in the 3.19 release were contributed by 1,422 developers — almost the same as the number of contributors to 3.18. The most active of these developers are:

Most active 3.19 developers

By changesets H Hartley Sweeten 463 3.7% Wolfram Sang 333 2.7% Alexander Aring 193 1.5% Benjamin Romer 176 1.4% Lars-Peter Clausen 154 1.2% Malcolm Priestley 126 1.0% Al Viro 123 1.0% Thierry Reding 121 1.0% Takashi Iwai 118 0.9% Daniel Vetter 103 0.8% Geert Uytterhoeven 103 0.8% SF Markus Elfring 99 0.8% Johan Hedberg 98 0.8% Michael S. Tsirkin 95 0.8% Hans Verkuil 84 0.7% Peter Hurley 82 0.7% Ville Syrjälä 79 0.6% Damien Lespiau 76 0.6% Fabian Frederick 73 0.6% Steven Rostedt 71 0.6%

By changed lines Malcolm Priestley 31115 4.5% Jeff Kirsher 26025 3.7% Hans Verkuil 22602 3.3% Nicolas Ferre 22580 3.2% Paolo Bonzini 14706 2.1% Hante Meuleman 13694 2.0% H Hartley Sweeten 12445 1.8% Ley Foon Tan 10222 1.5% Takashi Sakamoto 8704 1.3% Thierry Reding 8260 1.2% Philipp Zabel 7477 1.1% Joe Perches 7268 1.0% Daniel Vetter 6584 0.9% Pratik Patel 6573 0.9% Alexander Aring 6382 0.9% Vinod Koul 6085 0.9% Finn Thain 5663 0.8% Daeseok Youn 5554 0.8% Hannes Reinecke 4951 0.7% Christoph Hellwig 4818 0.7%

There may indeed come a day, in some distant, unimaginable future, when the Comedi drivers are done and Hartley Sweeten does not contribute the most changesets to a kernel release. But that day won't happen during the 3.19 cycle. Wolfram Sang eliminated unnecessary assignments to the owner field of struct platform_driver, Alexander Aring did a lot of work on the kernel's 802.15.4 implementation, Benjamin Romer cleaned up the unisys driver in the staging tree, and Lars-Peter Clausen did a bunch of sound-driver work.

In the "lines changed" column, Malcolm Priestley topped the list with a lot of work (including the removal of a lot of code) on the vt6655 driver in the staging tree. Jeff Kirsher contributed three patches, one of which removed the old and unloved bcm driver, while Hans Verkuil, as usual, did a lot of work within the Video4Linux tree. Nicolas Ferre removed a bunch of old ARM code, and Paolo Bonzini rearranged the KVM x86 code.

The developers working on 3.19 were supported by just over 200 companies, a typical number for recent development cycles. The list of the most active companies holds few surprises:

Most active 3.19 employers

By changesets Intel 1490 12.0% (None) 1363 10.9% Red Hat 1029 8.3% (Unknown) 935 7.5% (Consultant) 617 5.0% Vision Engraving Systems 463 3.7% SUSE 437 3.5% Samsung 420 3.4% Linaro 362 2.9% Texas Instruments 284 2.3% Google 270 2.2% IBM 267 2.1% Renesas Electronics 261 2.1% Unisys 208 1.7% AMD 205 1.6% Free Electrons 205 1.6% Freescale 202 1.6% NVidia 153 1.2% ARM 153 1.2% Broadcom 145 1.2%

By lines changed Intel 95283 13.7% (None) 78635 11.3% Red Hat 57378 8.3% (Unknown) 43819 6.3% Cisco 25757 3.7% (Consultant) 25511 3.7% Atmel 25239 3.6% Broadcom 18189 2.6% Linaro 18071 2.6% SUSE 17898 2.6% Samsung 16295 2.3% AMD 16186 2.3% Google 15152 2.2% Vision Engraving Systems 13307 1.9% Free Electrons 11559 1.7% Altera 11369 1.6% IBM 11324 1.6% Code Aurora Forum 10632 1.5% Texas Instruments 10377 1.5% NVidia 9639 1.4%

The percentage of developers working on their own time is holding steady at around 11%, suggesting that the long-term decline in volunteer developers may have come to an end.

To finish things out, it has been a while since we have taken a look at the longevity of code going into the kernel — in other words, how much of the current kernel came from each development cycle? The current picture looks like this:

Surviving lines of code from each development cycle

This chart was generated by running "git blame" on each file in the kernel tree and identifying the commit that created each line within the file. It starts at 2.6.13, omitting the beginning of the git era at 2.6.12 for clarity; that bar, representing the 3 million lines that have not been modified since 2.6.12, would dwarf all the others.

The story that emerges from that chart is that recent development cycles do not dramatically overshadow earlier cycles. That suggests that, once a line of code is put in place, it tends not to change much in the following cycles. Sometimes it seems that there is a lot of churn, frequently rewriting the same code, but the truth is that most of the code in the kernel has not been changed in years.

In other words, most kernel developers are busy adding new code, which is not a particularly surprising result. It takes a lot of new code to keep up with the demands for new features and support for new hardware. The evidence suggests that the kernel community is continuing to grow in response to those demands and the process is working reasonably well.

Index entries for this article
Kernel	Releases/3.19

---

to post comments