On Tue, 22 Jan 2008 13:30:07 +0100 Harald Welte laforge@openmoko.org wrote:
On Tue, Jan 22, 2008 at 10:32:53AM +0100, Haavard Skinnemoen wrote:
Hey, is that the thanks I get for actually commenting at your mmc driver? I don't recall rejecting anything, nor do I have the power to do so.
well, sorry. But since there were no other significant comments, it was sort of the most important/influential one.
No problem. I just thought it was a bit funny that you first complained about lack of comments, and then went on to complain that the comments you _did_ get were too thorough ;-)
I did suggest that you move the MMC protocol definitions to a common header file. That's not a huge task, but you could have said "no, I don't want to do that right now. Maybe later." and it would have been perfectly fine. Instead, you said "will do so in the next version of the patch." But no next version of the patch ever showed up, despite quite a few comments from others as well.
well, obiously I try to implement the suggestions of the people on this list, even if it means significant additional delays. There is no point in submitting the same stuff without implementing the comments from the review.
Yes, I agree in general. But I also think it should be perfectly ok to say "no, sorry, that's too much work at this point", although it may start a discussion about whether it _really_ is too much work. You should never silently ignore any comments, of course.
Review comments are really just suggestions about how to improve your work further, but you shouldn't ignore them unless you have a good reason to do so, and let the reviewer and other know.
Btw, I find the "lack of hardware" argument a bit disturbing. Doing some cleanups really shouldn't require anyone to have access to all the affected hardware
Well, the big issue is that most of those cleanups are deep down in the low-level head.S/start.S initialization, affecting the ordering of how PLL's are initialized, the resume-from-ram code paths, etc.
There's mmc code down at that level...?
Thos can be very tricky, and especially with the level of documentation that Samsung provides about chipset bugs (close to zero), it can be very easy to break something on one chip without noticing. They have done things along the lines of re-defining the meaning of the same bits in the same registers from 'pull-up' to 'pull-down' :)
If you're talking about cleanups in general, then yeah, those low-level bits can be quite fragile. I think the best you can do is to test it at your own hardware and clearly state that your patch touches some tricky low-level bits of code and needs testing.
If nobody steps up to test the code, I guess we have a problem. IMHO, I'd say screw them and commit the patch anyway, after a reasonable grace period. If it turns out to break something, you'll at least find out which boards are properly maintained and which boards aren't.
-- IMO much of the responsibility for testing that nothing breaks on real hardware lies on the custodians and other users that actually have access to the affected boards.
But a few cross toolchains are definitely required to verify that it actually compiles on a handful of configurations before submitting.
This would basically mean I had to know which toolchains other users of boards using the existing 24xx support are using.
U-boot is supposed to compile with standard toolchains isn't it?
Sometimes it doesn't, but that may be an interesting piece of information on its own, which you should report to the mailing list.
Yeah, I know, avr32 currently requires a vendor-supplied toolchain, but we are working on getting it upstream (honest!) and because of this, we really don't expect people to do compile-testing with the avr32 toolchain.
Is there somewhere a list of the toolchains that the code has to compile against, including pointers on where to obtain them?
I have a standard ARM toolchain and a standard ppc toolchain, both compiled from the Ubuntu gcc-4.2 sources. Those two architectures cover a lot of configurations, so I'd expect them to uncover the vast majority of problems with a patch.
ppc is currently broken though, since nobody seems to be interested in merging my one-line fix...
Haavard