Series of one patch?

[Stephen Finucane]

On Thu, 2017-06-22 at 13:03 -0700, Sean Farley wrote:
> Stephen Finucane <stephen at that.guru> writes:
> 
> > One of my Red Hat colleagues showed me a tool this morning - patches [1] -
> > that
> > the QEMU community use. They seem to be building a huge JSON blob of
> > seriesified patches as part of a cron job [2], and that tool can
> > download/parse
> > this blob and apply locally stored patches.
> > 
> > I noted that in that blob, _everything_ is considered to be part of a
> > series.
> > I've seen this design before in the freedesktop instance [3] and wasn't too
> > pushed on it at the time as it seemed like a bit of a lie (those patches
> > aren't
> > actually in a series). However, the ability to see *all* patches in series-
> > patch manner, rather than the inverted patch-series manner, is actually
> > rather
> > nifty. If would also mean testing of patches could happen without needing
> > to
> > filter for both 'patch-created' (for non-series patches) and 'series-
> > completed' 
> > events (for series patches).
> > 
> > What do folks think? Any reason not to create a series for _all_ patches?
> 
> I kinda assumed this was already done. Just to clarify, as of right now,
> patchwork doesn't not put a single patch into a series?

The key differentiator is whether the subject has series markers (i.e. [N/M]).
This means that if the patch has a cover letter (and therefore has a [1/1]
marker) then it would go into a series. Without this, however, the 'series'
attribute is unset.

> I like to think of all series (even those of one patch) as a branch. So,
> a bit like saying a branch can only have >1 commit and those of one
> commit are not on a branch. Weird to me.

That's not a bad analogy. I was thinking of them more as auto-generated
bundles, and there wasn't a need to "bundle" a single patch. A branch might be
a better approach.

Overall so, I'm guessing you're in favour of this?

Stephen