Multi-commit PRs vs. multiple single-commit PRs

[Andrei Alexandrescu via Digitalmars-d]

On 3/23/17 4:57 AM, Vladimir Panteleev wrote:
> On Wednesday, 22 March 2017 at 17:16:09 UTC, Andrei Alexandrescu wrote:
>> I'm a bit confused. This got settled a while ago, in part to avoid
>> silly debates over the inconsequential. Our organization prefers
>> squash before commit in the majority of cases. For a minority of pull
>> requests (that touch many files, are semi-mechanical etc) multiple
>> commits in one PR are fine within reason. These would be about one
>> order of magnitude less frequent. -- Andrei
>
> Well, I don't think we shouldn't keep researching for ways to improve
> wolkflow. I certainly don't think it's inconsequential, and anyone who
> has time and thinks they can bring fresh arguments to the table is
> welcome to do so.

Of course, new research is always welcome! The more, the better. Bring 
it over!

There's a spectrum at work; at one extreme there's be close-mindedness 
that keeps a rigid status quo and refuses to accept new evidence. At the 
other end of the spectrum there's frequent reopening of the same debate 
with the same arguments, then repeatedly agreeing to close it just to 
repeat the cycle at the next opportunity.

Walter and I think the better course of action for the community is to 
favor small pull requests that are squashed upon committing. There are 
reasons and a body of evidence that has been hashed over several times. 
Clearly there are extreme cases that don't do well with this flow, which 
confirm our understanding that no rule is a replacement of good 
judgment. Such rare exceptions are fine with us.

But we can't afford this incessant challenge of the slightest authority 
and this reopening of old decisions with no new arguments. Far as I 
understand (and please do correct me if I'm wrong) what's being 
discussed now does not qualify as new research and is a reopening of a 
previous discussion with no new evidence, in which one side of the 
dialog accuses the other of appeal to authority, while simultaneously 
invoking appeal to its own authority.

I have spent a long time this day thinking how to reply to this so as to 
close the argument once and for all, after I had already spent more time 
than is reasonable thinking and discussing this in public and in 
private. There really is no time for this kind of stuff if we want to 
scale. We should discuss how to make exceptions @nogc and reference 
counted strings and a bunch of other important and urgent things. What 
steps can we take on this particular matter so we save everybody 
involved time and cognitive load?


Thanks,

Andrei