[Intel-gfx] [PATCH] gpu/drm/i915: globally replace dev_priv with i915

[Jani Nikula]

On Thu, 13 Jun 2019, Rodrigo Vivi <rodrigo.vivi at intel.com> wrote:
> On Thu, Jun 13, 2019 at 09:48:16AM -0700, Lucas De Marchi wrote:
>> On Thu, Jun 13, 2019 at 09:29:48AM -0700, Lucas De Marchi wrote:
>> > On Thu, Jun 13, 2019 at 04:12:37PM +0300, Jani Nikula wrote:
>> > > On Wed, 12 Jun 2019, Lucas De Marchi <lucas.demarchi at intel.com> wrote:
>> > > > We are slowly converting dev_priv to i915 everywhere, spread into
>> > > > smaller series. While this is good to avoid unrelated breakages to other
>> > > > inflight patches, it's bad because inflight patches on nearby paths keep
>> > > > breaking. Paired with other code moves and refactores this is becoming a
>> > > > nightmare.
>> > > > 
>> > > > Now that I915_{READ,WRITE} are gone and implicit access to dev_priv no
>> > > > longer exists we can simply sed all files and be done with the
>> > > > conversion. This was generated with the following commands with no
>> > > > additional fixups:
>> > > > 
>> > > > 	git ls-files --full-name -z -- drivers/gpu/drm/i915/ | \
>> > > > 		xargs -0 sed -i 's/\bdev_priv\b/i915/g'
>> > > > 
>> > > > Any pending series can apply the same search and replace when rebasing.
>> > > 
>> > > I'm pretty strongly against renaming the implicit dev_priv local
>> > > variable before we've gotten rid of it. Renaming s/dev_priv/i915/ upon
>> > > converting a function to not use the implicit dev_priv helps us by
>> > > catching any leftover references.
>> > 
>> > I don't think this is strong enough reason to block the conversion. The
>> > conversion is taking forever and is gonna take year+ if it continues at
>> > this pace. It affects multiple developers rebasing their work and
>> > introduce bugs on pending series.
>> > 
>> > Removing any file-scope reference (I didn't review yet if we still have
>> > any) and reviewing the macros is sufficient. And if we later we find out
>> > we missed one, we just go and fix it. I see zero advantage on slow and
>> > forever. You have a mixed code base and new code following what's around
>> > will just propagate more the mixed code base :(
>> 
>> btw, let me be clear the proposal is not "this should be done now". I do
>> think we should be smart and think on the best opportunity to do it.
>> Probably like after the big code moves gt/, gem/, display/, etc end.
>
> yeap, doing per folder is probably a good thing, specially because gt and gem
> are far ahead this conversion.

In general I think getting rid of the implicit dev_priv local is much
more important than a global s/dev_priv/i915/ rename. In the end, it's
just a local variable; the problematic part is the dependency on the
implicit name.

IMO you don't even have to do the rename at the time of getting rid of
the local. It's just a nice bonus to be able to freely choose the
variable name.

For the renames, sure, please do them in gt and gem subdirs first,
because that's actually feasible:

$ git grep -w dev_priv -- gt | wc -l
94

$ git grep -w dev_priv -- gem | wc -l
168

$ git grep -w dev_priv -- display | wc -l
6241


BR,
Jani.


>
>> 
>> Lucas De Marchi
>> 
>> > 
>> > Lucas De Marchi
>> > 
>> > > 
>> > > BR,
>> > > Jani.
>> > > 
>> > > 
>> > > -- 
>> > > Jani Nikula, Intel Open Source Graphics Center

-- 
Jani Nikula, Intel Open Source Graphics Center