Thursday, 26 November 2009

Sometimes it don't come easy, sometimes it don't come at all

My work sometimes causes me to have reason to change small things in open source projects which I then have to submit for consideration upstream. Often this is as simple as providing a patch on the relevant mailing list and moving on.

Sometimes however the change is larger and I really would like to see the feature or bug fix accepted by the project. I am fortunate that my employer is enlightened enough to not only accept this but actively encourage participation in open source communities.

Most of the time the patches are accepted, with changes or updates, in a reasonable timescale. Certainly for kernel work I have got used to a submission/accept time measured in hours!

On occasion the change gets dropped on the floor or missed and you have to resubmit (even if its just to get told its uninteresting and to go away ;-). Unfortunately there are a small number of projects where getting something merged is an experience in frustration.

The particular project which has caused me the most trouble along these lines is Qemu. Daniel and I did the work to support the Samsung ARM based 2410 and 2440 System on Chip (SoC) and a couple of boards they were used on. Of course we then presented the patch to the Qemu mailing list. That was in November...No, not this year, or even last, we are talking 2006!

Unsurprisingly there were numerous issues with our first cut, we improved the patches and resubmitted them and received absolutely no feedback. Eventually we provoked the maintainer to respond, we took his feedback into account and resubmitted. After a year of so of this Daniel got bored and gave up.

Being a complete idiot and apparently sucker for punishment I decided to re-try. I completely restructured the port and started by just trying to get the absolute minimal core changed accepted. It took some time to provoke a response (oh and they altered RCS which meant I had to change my submission again, but they did change to git so that is acceptable ;-)

Earlier this year resulted in my all time favorite maintainer response. Eventually I managed to find out he actually objected to the patch because the SoC we were using actually required a slightly different processor capability set (the emulation is ARMv5 by default and the SoC in question is ARM v4te, hey I had not noticed any practical difference, and Qemu is an emulator not a simulator...but OK)

Remember this is the first time this objection has been raised in 30months of patch submission. Fine, I create a patch which adds V4te emulation and submit that. I eventually get a response, to which my reply is a little frustrated I will admit.

I have since resubmitted the patch with the corner cases fixed, it has been ignored, and I never have received a reply to my direct questions. On this three year anniversary of the original submission I am simply going to give up, something I detest doing.

I am one of those dull people that finishes what he starts, that is pretty much my role in Simtec these days, ensuring stuff we start gets finished or at least drawing the line and calling it complete. To end up with a complete failure offends my personal values.

Now please do not think I hold any ill feelings towards anyone concerned, open source maintainers perform their roles without pay and, by extension, without responsibility to anyone but themselves. I am sure people are too busy to care about the itch I wanted scratched and that overall their project has not really lost all that much by not having me as a contributor.

I know I could have forked the project, or maintained my own tree or any number of other technical solutions but at the end of 1,100 days I have finally exhausted my enthusiasm and can move on. Speaking of which, time to resubmit a kbuild change which got dropped ;-)

No comments:

Post a Comment