how to structure some disruptive patches

[Frank Rowand <frowand.list-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]

Hi Rob,

I've been working on cleaning up the docbook function documentation
in the device tree files.  The resulting patches could interfere with
code patches, so I want to figure out how to structure the documentation
patches to avoid that.  I would like your opinion and suggestions on
the subject.

First premise: this set of documentation changes is a second class
citizen compared to code changes.  Any patch from this set can be
dropped any time it interferes with a code patch.

I started the project with the simple intent of fixing what was
clearly broken.
  - fix syntax errors
  - fix mismatches between actual function arguments vs docbook
    list of function arguments
  - fix mismatches between function return type (void vs anything) and
    values vs docbook return types and values
  - fix factually incorrect descriptions of what the function does

When I was making those changes, I found also found documentation
that was somewhat cryptic or otherwise hard to understand.  I also
found that the description of the same item was described in a
wide variety of ways in different docbook headers.  I made an
effort to also clean that up.

The result was a large patch series (that was not totally complete):

   15 files changed, 1414 insertions(+), 730 deletions(-)

plus around 150 lines of change to docbook files so that
the device tree man pages would be created.  I did not
add any other significant docbook documentation of device tree.

There are at least a couple of approaches I could take for
submitting patches (and would be glad to hear of any other
approach that I did not think of):

1) The big ugly patchset referenced above, one patch per
   file.
     pro: exposes what changes were made to each function
          documentation header
     con: very dense and not very readable
     con: the mechanical corrections create a lot of noise if
          the reviewer wants to view actual content changes
     con: probably more likely to conflict with code patches
          in a way that is not easily fixed

2) Split method 1 into stages (one patch per file for each stage).
   Examples of some possible stages are:
   - white space fixes
   - syntax fixes
   - incorrect argument list fixes
   - incorrect return type fixes
   - incorrect return value fixes
   - incorrect behavior description fixes
   - confusing behavior description fixes
     pro: it would be easier to review patches for many of the stages
     con: a lot more patches
     con: maybe difficult to handle conflicts with code patches
     con: maybe difficult to rework patches for review comments (changes
          for an earlier stage are likely to impact a patch for a later
          stage)

3) Do one patch series to remove all docbook function header documentation.
   Do a second patch series to add the updated docbook function header
   documentation.  For each of the two series, one patch per file.
     pro: the patches are much easier to read
     pro: it might be easier to resolve conflicts with source patches
          (either you dropping a few hunks or me redoing the docbook patch
          to remove the conflict)
     con: the actual changes to what the documentation says are not visible

My current version of the patches is against 4.7-rc2.  I will have to update
this to 4.7 or 4.8-rc1 (I would assume that 4.8-rc1 would make more sense).

One comment that applies to all of the above approaches is that I could
attack subsets of the device tree files instead of trying to do all of
them at the same time.

What do you think of all of this?

-Frank
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html