PIL + ReportLab vs Pillow

[petr at macports.org]

On 8 Jul 2014, at 02:00, Mihai Moldovan <ionic at ionic.de> wrote:

> * On 07.07.2014 11:32 am, Peter Danecek wrote:
>> Why not allowing that either PIL or Pillow satisfies the dependency where possible.
>> [...]
>> This would avoid potential conflicts if users need to install other (still) PIL dependent software.
> 
> That would be possible, yes. But as I understood it, Pillow is a drop-in
> replacement for PIL and is always supposed to work with software initially
> written for PIL.

Probably, yes. But I do not know the details about compatibility, so would not make the statement that Pillow can *always* replace PIL.

> However, I'm not a python programmer and might be wrong.
> 
> My thinking was that, as Pillow is supposed to be a PIL-compatible drop-in
> replacement, we could avoid all problems like
> - package1 -> Pillow
> - package2 -> PIL
> => package1 can't be installed at the same time with package2 (and vice versa)

This is why I am proposing the solution above. If at least one of the above packages (e.g. package1) can install with both alternatives, you will be able to install, but the sequence is relevant. If both support alternative, you have no problem at all.

Of cause, if you would replace PIL as dependency in all ports at one time, no conflicts arise.

> (Note that while PIL is supposed to always be substitutable with Pillow, the
> other way is not possible, if specialized Pillow features are being used.)
> 
> Does that make sense?

I this is true, PIL could be phased out at some point.

~petr