FFTConvPDFV1 doesn't work with create_extended

[LHUecker]

Bug

When a PDF is created with FFTConvPDFV1, the create_extended method raises a Type Error.
However set_yield still works.

Current Behaviour

import zfit
obs = zfit.Space('x', (-0.8,1.3))
yield = zfit.Parameter('sig_yield', 500, step_size=1)
loc = zfit.Parameter('loc', 0, floating=False)
lam = zfit.Parameter('lam', 0, -2, 1, step_size = 0.001)
sigma = zfit.Parameter('sigma', 0.07, 0.0, 0.2, step_size = 0.001)
lower = zfit.Parameter('lower', 0.0, -0.5, 0.5, step_size = 0.0001)
upper = zfit.Parameter('upper', 0.3, -0.2, 0.8, step_size = 0.0001)

gauss = zfit.pdf.Gauss(obs=obs, mu=loc, sigma=sigma)
uniform = zfit.pdf.Uniform(obs=obs, low=lower, high=upper)

conv = zfit.pdf.FFTConvPDFV1(uniform, gauss)
conv_ext = conv.create_extended(yield)

 ---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/tmp/ipykernel_2005/35580713.py in <module>
----> 1 conv_ext = conv.create_extended(yield_c)

~/.local/lib/python3.9/site-packages/zfit/core/basepdf.py in create_extended(self, yield_, name_addition)
    445         if self.is_extended:
    446             raise AlreadyExtendedPDFError("This PDF is already extended, cannot create an extended one.")
--> 447         new_pdf = self.copy(name=self.name + str(name_addition))
    448         new_pdf.set_yield(value=yield_)
    449         return new_pdf

~/.local/lib/python3.9/site-packages/zfit/core/basepdf.py in copy(self, **override_parameters)
    618         #     parameters.update(distribution=self.distribution)
    619         yield_ = parameters.pop('yield', None)
--> 620         new_instance = type(self)(**parameters)
    621         if yield_ is not None:
    622             new_instance.set_yield(yield_)

TypeError: __init__() got an unexpected keyword argument 'pdfs'

Expected Behaviour

Should return a new PDF that is extended.

Context (Environment)

• zfit version: 0.8.2 & 0.9.0a0
• Python version: 3.9.10
• Are you using conda, pipenv, etc? : conda
• Operating System: CentOS Linux (lxplus)
• Tensorflow version: 2.5.0

Possible Solution/Implementation

set_yield works perfectly fine as an alternativ if one doesn't need to return a new PDF.

[jonas-eschle]

Hey, thank you very much for raising the issue! This is indeed a known problem which has roots that are deeper in the design to do it properly and the recommended way is to use set_yield in place for now.

[jonas-eschle]

Just to add to this @LHUecker , the preferred way is now to create an extended PDF right with the initialization, all PDFs now take an extended argument which can be a yield.
Does this work?