[libcamera-devel] [PATCH v2 2/2] libcamera: pipeline: converter: return unique converter only if Valid

[Suhrid Subramaniam]

An HTML attachment was scrubbed...
URL: <https://lists.libcamera.org/pipermail/libcamera-devel/attachments/20230301/8f4a4d68/attachment.htm>
-------------- next part --------------


> -----Original Message-----
> From: Kieran Bingham <kieran.bingham at ideasonboard.com>
> Sent: Tuesday, February 28, 2023 3:39 PM
> To: Laurent Pinchart <laurent.pinchart at ideasonboard.com>; Laurent
> Pinchart via libcamera-devel <libcamera-devel at lists.libcamera.org>; Suhrid
> Subramaniam <Suhrid.Subramaniam at mediatek.com>
> Cc: libcamera-devel at lists.libcamera.org
> Subject: Re: [libcamera-devel] [PATCH v2 2/2] libcamera: pipeline: converter:
> return unique converter only if Valid
> 
> Quoting Laurent Pinchart via libcamera-devel (2023-02-28 23:13:12)
> > Hi Suhrid,
> >
> > On Tue, Feb 28, 2023 at 06:14:58PM +0000, Suhrid Subramaniam wrote:
> > > > -----Original Message-----
> > > > From: Suhrid Subramaniam
> > > > Sent: Tuesday, February 28, 2023 10:04 AM
> > > > To: Paul Elder <paul.elder at ideasonboard.com>; Suhrid Subramaniam
> > > > <suhridsubramaniam at gmail.com>
> > > > Cc: libcamera-devel at lists.libcamera.org
> > > > Subject: RE: [libcamera-devel] [PATCH v2 2/2] libcamera: pipeline:
> converter:
> > > > return unique converter only if Valid
> > > >
> > > > Hi Paul,
> > > >
> > > > Thanks for reviewing the patch.
> > > > Okay, I'll squash both patches.
> > > >
> > > > > -----Original Message-----
> > > > > From: Paul Elder <paul.elder at ideasonboard.com>
> > > > > Sent: Monday, February 27, 2023 11:06 PM
> > > > > To: Suhrid Subramaniam <suhridsubramaniam at gmail.com>
> > > > > Cc: libcamera-devel at lists.libcamera.org; Suhrid Subramaniam
> > > > > <Suhrid.Subramaniam at mediatek.com>
> > > > > Subject: Re: [libcamera-devel] [PATCH v2 2/2] libcamera: pipeline:
> > > > converter:
> > > > > return unique converter only if Valid
> > > > >
> > > > > On Mon, Feb 27, 2023 at 02:49:10PM -0800, Suhrid Subramaniam via
> > > > > libcamera-devel wrote:
> > > > > > - Use isValid() to check if m2m_ exists for the selected converter_.
> > > > > > - create an instance of the converter only if it is Valid.
> > > > > >
> > > > > > Signed-off-by: Suhrid Subramaniam
> > > > <suhrid.subramaniam at mediatek.com>
> > > > > > ---
> > > > > >  src/libcamera/converter.cpp | 7 ++++---
> > > > > >  1 file changed, 4 insertions(+), 3 deletions(-)
> > > > > >
> > > > > > diff --git a/src/libcamera/converter.cpp
> > > > > > b/src/libcamera/converter.cpp index 3de39cff..8a34d068 100644
> > > > > > --- a/src/libcamera/converter.cpp
> > > > > > +++ b/src/libcamera/converter.cpp
> > > > > > @@ -207,8 +207,9 @@
> > > > ConverterFactoryBase::ConverterFactoryBase(const
> > > > > std::string name, std::initiali
> > > > > >   * \param[in] media Name of the factory
> > > > > >   *
> > > > > >   * \return A unique pointer to a new instance of the
> > > > > > converter subclass
> > > > > > - * corresponding to the named factory or one of its alias.
> > > > > > Otherwise a null
> > > > > > - * pointer if no such factory exists
> > > > > > + * corresponding to the named factory or one of its alias if
> > > > > > + the converter
> > > > > > + * instance is valid (checked using isValid()). Otherwise a
> > > > > > + null pointer
> > > > > > + * if no such factory exists
> > > > > >   */
> > > > > >  std::unique_ptr<Converter>
> > > > > > ConverterFactoryBase::create(MediaDevice
> > > > > > *media)  { @@ -227,7 +228,7 @@ std::unique_ptr<Converter>
> > > > > > ConverterFactoryBase::create(MediaDevice *media)  <<
> > > > > > factory->name_ << " factory with "
> > > > > >  << (it == compatibles.end() ? "no" : media->driver())
> > > > > << "
> > > > > > alias.";
> > > > > >
> > > > > > -return factory->createInstance(media);
> > > > > > +return factory->createInstance(media)->isValid() ?
> > > > > > +factory->createInstance(media) : nullptr;
> > > > >
> > > > > Ah, I see, you want to move the isValid() check here. I think
> > > > > you can just squash this patch into the previous one.
> > > > >
> > > > > Also you're creating two instances if it is valid. I suppose the
> > > > > first one does get deconstructed immediately, but still I don't
> > > > > think it's a good
> > > > idea.
> > > > >
> > > >
> > > > Understood.
> > > > So does it make sense to do the following? Or is there a better
> > > > method you'd recommend?
> > > >
> > > > << (it == compatibles.end() ? "no" : media->driver()) << "
> > > > alias.";
> > > >
> > > > + converter_ = factory->createInstance(media);
> > > > -return factory->createInstance(media);
> > > > +return converter_ ? converter_->isValid() : nullptr;
> > >
> > > Oops, I meant:
> > > +return converter_->isValid() ? converter_ : nullptr;
> >
> > That looks good. As Paul said, I would squash the two patches together.
> > Otherwise you're introducing a bug in patch 1/2 that you then fix in
> > 2/2.


While trying to compile and check if this works, I came across an interesting compile error regarding unique pointer ownership.
I'd to modify the code to use std::move() as follows:

+ converter_ = factory->createInstance(media);
+ return converter_->isValid() ? std::move(converter_) : nullptr;

I hope this is okay.

Secondly,
Instead of nullptr, shouldn't we return an empty object of Converter unique pointer?

- return nullptr
+ return std::unique_ptr<Converter>()


> 
> Indeed, I believe I originally suggested this patch could preceed the other so
> that wouldn't happen - but I'm fine with them being squashed too.
> 
> --
> Kieran
> 
> 
Hi Kieran,
Got it. I'll squash the patches : )

> >
> > > > > >  }
> > > > > >
> > > > > >  return nullptr;
> >
> > Could you please drop this from mails sent to public mailing lists ?
> > Technically, it prevents us from merging your code.

Yep, done. Sorry about that!
> >
> > --
> > Regards,
> >
> > Laurent Pinchart