[patch,gfortran] PR 21625: Initialization of derived types at allocation

[Tobias Schlüter]

Paul Thomas wrote:
>>No very strong reason, really.  I had to put it somewhere, and
>>resolve_code seemd to be one poosibility.  Since I could make it
>>work reasonably easily there, I decided to go for that
>>alternative. If it's belived that match.c(gfc_match_allocate)
>>would be a better place, I can see if I can move it there.
> 
> I have been doing my fair share of polluting resolve.c but have been 
> trying to keep to resolutions that span namespaces (eg. require host 
> associated symbols) or that use information that is not available until 
> then... not always very successfully, I might add!  However, I did sweep 
> out a lot of the constraints that should have been elsewhere (at Tobi's 
> urging, I might add.).
> 
> What do others think about this?

I don't think that what Erik's patch does can be done earlier than during
resolution -- after all he's inserting runtime code which is executed after an
allocate statement.  Unless I misunderstood something.

- Tobi