[LLVMdev] code-owner sporks

dag at cray.com dag at cray.com
Fri Nov 16 15:50:52 PST 2012

Sean Silva <silvas at purdue.edu> writes:

>> - I have to *remember* I submitted the patch (not hard, but it is a
>>   cost).
> If you forgot, the chances are high that the patch was unimportant. I
> do my development on local git branches, so every time I do `git
> branch`, I'm reminded. There's really no overhead.

Every time you have to check it's overhead.  It may be small overhead,
but it's nonzero.

>> - I have to save that e-mail from llvm-commits so I can refer to it when
>>   the inevitable ping is necessary.
> Maybe you need a better mail-reader? In gmail I just look in my "sent
> mail" or if I have sent a lot of mail recently I search "is:sent
> has:attachment". It literally takes 10 seconds.

Ok, I'll give you this one.  :)

>> - I have to wade through tons and tons of commit e-mails searching for a
>>   response to my patch (for some reason the mailing list software often
>>   breaks threading).  This is a more general problem with the current
>>   review process, not strictly a timeliness issue.
> Sounds like you need a better mail reader. I have never heard of this
> even remotely being a problem in any way, at all, ever.

Really?  You've never heard comments about the huge volume of mail on
llvm-commits?  I just read another message about it today.

>> - I have to send a ping e-mail.
> Once again, this takes like what, 10 seconds? I find it highly
> unlikely that 10 seconds is a non-negligible amount of time compared
> to the time spent coding, building, and testing any non-obvious (i.e.
> needing review) change.

Again, it's non-zero.  I have to break out of whatever I'm doing,
context switch and send the mail.

>> - Now I have to look for responses to *two* e-mails.
> You definitely need a better mail reader. In every mail reader I've
> ever seen, both mails get clustered into a thread. Maybe this is the
> root cause of your vehement dissatisfaction with the current
> email-based system?

Even if threads are clustered, one still has to look through tons of
subject lines to find it.

> Have you considered the amount of quality code that gets produced with
> the current process? 

Code quality doesn't reflect how efficient the review system is.  It
reflects how good the reviewers are and we are fortunate to have
excellent ones.

I find llvm-commits daunting.  So much that I hesitate to do reviews.
As Chris commented, I am not very active on that list.  There's a reason
for that beyond lack of time.  By definition, making the process simpler
and more organized should result in less time for reviews and thus
hopefully more review participation.

> What kind of improvements can we expect from such a system?

A one-stop shop for reviewers to find patches.  Something searchable by
subsystem, file name, etc.  I can't review every patch that goes by but
I certainly can review patches for files I'm currently working in.  If I
could start my day by doing a patch search on whatever file I'm
currently working on, I'd make it a daily practice to do some reviews.
Right now that is tedious to do with e-mail.

> Are there advantages that email has over a more structured
> sort of review system like Phabricator? 

I don't know.  I would like to know if there are any.  I suppose it's
super convenient to open the list of messages but beyond that I get

> Are you sure that the bottleneck
> which causes high review latency isn't elsewhere in the system?

Of course I'm not sure.  I'm not sure anyone is sure.  :) All I know is
that there's a problem.

> Or, as counterintuitive as it may seem, perhaps all of the things you
> are pointing to as problematic (e.g. review latency, patches getting
> dropped) are actually *beneficial* (maybe it prevents less-determined
> people from becoming contributors?

If that's our goal then I will just say right out it is not a good one.


More information about the llvm-dev mailing list