[v1,1/2] PCI: Document patch submission hints

From: Bjorn Helgaas <bhelgaas@google.com>

Add hints about how to write PCI patches and changelogs.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
---
 Documentation/PCI/00-INDEX               |    2 
 Documentation/PCI/submitting-patches.txt |  153 ++++++++++++++++++++++++++++++
 2 files changed, 155 insertions(+)
 create mode 100644 Documentation/PCI/submitting-patches.txt

On Fri, Jun 29, 2018 at 03:27:39PM -0500, Bjorn Helgaas wrote:
> +  - Wrap changelogs to fit in 80 columns when shown by "git show", which
> +    adds 4 spaces.  I use "textwidth=75" in vim.

I guess the ideal width is subjective, I usually wrap at 72 chars because
then you've got 4 blanks on either side when viewed with "git log", which
I find neater than maxing out the horizontal width.

In some cases I deliberately wrap at less than 72 chars or allow 73 chars
if it avoids vastly unequal widths in a paragraph.  Often when I later
doublecheck what you've committed, I find that you've rewrapped everything
to 75 chars and the result doesn't look as neat as I wanted it to be.
Not a big deal, but thought I'd mention it now that you're codifying this
"rule" somewhat more formally.

Lukas

On Fri, Jun 29, 2018 at 03:27:39PM -0500, Bjorn Helgaas wrote:
> --- /dev/null
> +++ b/Documentation/PCI/submitting-patches.txt
> @@ -0,0 +1,153 @@
> +Start with Documentation/process/submitting-patches.rst for general
> +guidance.
> +
> +These are things I look at when reviewing patches.

For an uninitiated reader who doesn't know that you're currently the
(sole) maintainer of the PCI subsystem, this sentence might look odd.
Who's "I"?  What happens if you onboard co-maintainers, are you
going to change this to "we"?


> +  - Wrap code and comments to fit in 80 columns.  Exception: I prefer
> +    printk strings to be in one piece for searchability, so don't split
> +    quoted strings to make them fit in 80 columns.

This is a duplication of Documentation/process/coding-style.rst, section 2.


> +  - Follow the existing convention  Run "git log --oneline <file>" and make
> +    your subject line match previous changes in format, capitalization, and
> +    sentence structure.  For example, native host bridge driver patch
> +    titles look like this:
> +
> +      PCI: vmd: Remove IRQ affinity so we can allocate more IRQs
> +      PCI: mediatek: Add MSI support for MT2712 and MT7622
> +      PCI: rockchip: Remove IRQ domain if probe fails

A quick "git log --oneline --no-merges drivers/pci" shows that the prefixes
in use aren't consistent at all:  Sometimes a slash is used to separate
"PCI" from the subpart touched by the patch, sometimes a colon, e.g.
"PCI/AER: " versus "PCI: shpchp: ".  Your own patches aren't consistent
in that respect.  Sometimes, only "PCI: " is given as prefix, even though
the commit only touches a subpart such as "sysfs", so could easily specify
more precisely what it's touching.

If you value consistency, it would be good to codify the preferred form
right here.


> +  - Include specific details, e.g., write "Add XYZ controller support"
> +    instead of "add support for new generation controller".

Why not simply "Support XYZ controller"?  One word less, more succinct.


> +  - Always copy linux-pci@vger.kernel.org and linux-kernel@vger.kernel.org.

I'd drop linux-kernel here.  The volume on that list is already like
drinking from a firehose, I doubt it adds much value to cc it.

Thanks,

Lukas

On Sun, Jul 01, 2018 at 07:45:08PM +0200, Lukas Wunner wrote:
> On Fri, Jun 29, 2018 at 03:27:39PM -0500, Bjorn Helgaas wrote:
> > --- /dev/null
> > +++ b/Documentation/PCI/submitting-patches.txt
> > @@ -0,0 +1,153 @@
> > +Start with Documentation/process/submitting-patches.rst for general
> > +guidance.
> > +
> > +These are things I look at when reviewing patches.
> 
> For an uninitiated reader who doesn't know that you're currently the
> (sole) maintainer of the PCI subsystem, this sentence might look odd.
> Who's "I"?  What happens if you onboard co-maintainers, are you
> going to change this to "we"?

You're absolutely right.  That was appropriate for email (where I
originally posted this) but certainly not for the tree.

> > +  - Follow the existing convention  Run "git log --oneline <file>" and make
> > +    your subject line match previous changes in format, capitalization, and
> > +    sentence structure.  For example, native host bridge driver patch
> > +    titles look like this:
> > +
> > +      PCI: vmd: Remove IRQ affinity so we can allocate more IRQs
> > +      PCI: mediatek: Add MSI support for MT2712 and MT7622
> > +      PCI: rockchip: Remove IRQ domain if probe fails
> 
> A quick "git log --oneline --no-merges drivers/pci" shows that the prefixes
> in use aren't consistent at all:  Sometimes a slash is used to separate
> "PCI" from the subpart touched by the patch, sometimes a colon, e.g.
> "PCI/AER: " versus "PCI: shpchp: ".  Your own patches aren't consistent
> in that respect.  Sometimes, only "PCI: " is given as prefix, even though
> the commit only touches a subpart such as "sysfs", so could easily specify
> more precisely what it's touching.
> 
> If you value consistency, it would be good to codify the preferred form
> right here.

I was trying to make the point that whenever we're doing *anything* in
a shared project like Linux, it's better to look at existing practice
and follow it than it is to slavishly follow rules.

So I purposely didn't enumerate all the cases.  I think if you look at
logs of individual files, they're pretty consistent.  I generally use
"PCI/XXX" for things in the core (mostly capabilities like MSI, AER,
DPC, etc) and "PCI: xxx:" for drivers (shpchp, pciehp, etc).  IIRC, I
adopted this from previous practice.

Of course there are things I don't apply that are different.  And
Rafael does most of the PM stuff, so I try to follow *his* convention
for that.

> > +  - Always copy linux-pci@vger.kernel.org and linux-kernel@vger.kernel.org.
> 
> I'd drop linux-kernel here.  The volume on that list is already like
> drinking from a firehose, I doubt it adds much value to cc it.

It's useful as an archive and for people not subscribed to linux-pci
who happen to see the occasional thing they want to respond to.
Nobody expects to be able to follow everything on LKML.
get_maintainer.pl reports LKML for everything, and I'm not trying to
innovate by being different.

But on reflection, I think the overall value of this writeup is
minimal.  It's a lot of repetition of things already documented
elsewhere and most of it boils down to "pay attention to existing
practice and don't do things differently unless you're innovating and
adding value."  That *should* be obvious, and if it's not, I doubt
that adding one more thing to read is going to make it more obvious.

Bjorn

On Thu, Jul 12, 2018 at 10:59:46AM -0500, Bjorn Helgaas wrote:
> But on reflection, I think the overall value of this writeup is
> minimal.  It's a lot of repetition of things already documented
> elsewhere and most of it boils down to "pay attention to existing
> practice and don't do things differently unless you're innovating and
> adding value."  That *should* be obvious, and if it's not, I doubt
> that adding one more thing to read is going to make it more obvious.

So my opinion is that your writeup does contain valid points that are
worth documenting:  For an open source project, a top priority is to
attract and retain contributors who improve the bus factor, who keep
the code base alive and maintained, thereby avoiding bit rot.

Knowledge diffusion, including documentation of best practices and
conventions, goes a long way towards that goal.  Your writeup was
mainly from a maintainer perspective: "consistency makes maintenance
easier".  But consistency is also valuable from a contributor
perspective:  It makes it easier to dive into a code base and find
your way around, and that includes changelogs in the git history.

There are important bits of knowledge in the writeup, if those can
be distilled, the result would very much be valuable to have in the tree.

Example:

> I generally use
> "PCI/XXX" for things in the core (mostly capabilities like MSI, AER,
> DPC, etc) and "PCI: xxx:" for drivers (shpchp, pciehp, etc).

That was in fact unknown to me.

If you find it difficult to put yourself in the shoes of a contributor,
I could try to rework the document and distill the points I find important.

Thanks,

Lukas

On Mon, Jul 30, 2018 at 04:31:36PM +0200, Lukas Wunner wrote:
> On Thu, Jul 12, 2018 at 10:59:46AM -0500, Bjorn Helgaas wrote:
> > But on reflection, I think the overall value of this writeup is
> > minimal.  It's a lot of repetition of things already documented
> > elsewhere and most of it boils down to "pay attention to existing
> > practice and don't do things differently unless you're innovating and
> > adding value."  That *should* be obvious, and if it's not, I doubt
> > that adding one more thing to read is going to make it more obvious.
> 
> So my opinion is that your writeup does contain valid points that are
> worth documenting:  For an open source project, a top priority is to
> attract and retain contributors who improve the bus factor, who keep
> the code base alive and maintained, thereby avoiding bit rot.
> 
> Knowledge diffusion, including documentation of best practices and
> conventions, goes a long way towards that goal.  Your writeup was
> mainly from a maintainer perspective: "consistency makes maintenance
> easier".  But consistency is also valuable from a contributor
> perspective:  It makes it easier to dive into a code base and find
> your way around, and that includes changelogs in the git history.
> 
> There are important bits of knowledge in the writeup, if those can
> be distilled, the result would very much be valuable to have in the tree.
> 
> Example:
> 
> > I generally use
> > "PCI/XXX" for things in the core (mostly capabilities like MSI, AER,
> > DPC, etc) and "PCI: xxx:" for drivers (shpchp, pciehp, etc).
> 
> That was in fact unknown to me.
> 
> If you find it difficult to put yourself in the shoes of a contributor,
> I could try to rework the document and distill the points I find important.

I definitely want to make it easy and attractive for people to
contribute to Linux in general.  I'd be glad for any hints.  You're
right, it's hard for me to put myself in the shoes of a contributor,
at least a new one.

I guess I'm hesitant because a lot of the things I included there seem
like they border on the obsessive, and I don't want to ask
contributors to make trivial changes to fit in with my personal
quirks.  Since they're my quirks, I'd rather just silently fix them up
as I apply patches.

If there were a wider consensus on some things and they could be
folded into Documentation/process/ somehow, that would be ideal, but
I'm not sure there would be enough of a consensus, and I don't really
want to start a big discussion about it.  But maybe there are a few
things that wouldn't be controversial, I dunno.

Bjorn