Fix bug due to __props__ in OpFromGraph subclasses

[ricardoV94]

When specified, Ops with identical __props__ are considered identical, in that they can be swapped and given the original inputs to obtain the same output.

Description

This was causing distinct OFG to be reused incorrectly for different nodes during function inner graph cloning. When provided Ops with the same __props__ are considered identical and can be swapped / recycled. It would be fine for OFG to have __props__ but would need to include the fgraph as well. However, then, the fgraph should be frozen because __props__ are also used for hashing and can't be mutable. For now I removed the __props__ from the two use cases.

The relevant function is

pytensor/pytensor/graph/basic.py

Lines 1259 to 1308 in a3f0a4e

	def clone_node_and_cache(
	node: Apply,
	clone_d: dict[Union[Apply, Variable, "Op"], Union[Apply, Variable, "Op"]],
	clone_inner_graphs=False,
	**kwargs,
	) -> Apply | None:
	"""Clone an `Apply` node and cache the results in `clone_d`.
	This function handles `Op` clones that are generated by inner-graph
	cloning.
	Returns
	-------
	``None`` if all of `node`'s outputs are already in `clone_d`; otherwise,
	return the clone of `node`.
	"""
	if all(out in clone_d for out in node.outputs):
	# If all of `node`'s outputs already have replacements or clones in
	# `clone_d`, then there's likely no need to clone it
	return None
	# Use a cached `Op` clone when available
	new_op: Op | None = cast(Optional["Op"], clone_d.get(node.op))
	cloned_inputs: list[Variable] = [cast(Variable, clone_d[i]) for i in node.inputs]
	new_node = node.clone_with_new_inputs(
	cloned_inputs,
	# Only clone inner-graph `Op`s when there isn't a cached clone (and
	# when `clone_inner_graphs` is enabled)
	clone_inner_graph=clone_inner_graphs if new_op is None else False,
	**kwargs,
	)
	if new_op:
	# If we didn't clone the inner-graph `Op` above, because
	# there was a cached version, set the cloned `Apply` to use
	# the cached clone `Op`
	new_node.op = new_op
	clone_d[node] = new_node
	if new_node.op is not node.op:
	clone_d.setdefault(node.op, new_node.op)
	for old_o, new_o in zip(node.outputs, new_node.outputs):
	clone_d.setdefault(old_o, new_o)
	return new_node

[codecov]

Codecov Report

Attention: Patch coverage is 75.00000% with 1 line in your changes missing coverage. Please review.

Project coverage is 81.72%. Comparing base (1509cee) to head (fbd1e4d).
Report is 110 commits behind head on main.

Files with missing lines	Patch %	Lines
pytensor/tensor/einsum.py	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #981      +/-   ##
==========================================
+ Coverage   81.70%   81.72%   +0.02%     
==========================================
  Files         182      182              
  Lines       47596    47598       +2     
  Branches    11589    11589              
==========================================
+ Hits        38889    38901      +12     
+ Misses       6518     6508      -10     
  Partials     2189     2189              

Files with missing lines	Coverage Δ
pytensor/tensor/basic.py	91.42% <100.00%> (+<0.01%)	⬆️
pytensor/tensor/einsum.py	96.41% <50.00%> (-0.50%)	⬇️

... and 4 files with indirect coverage changes

---- 🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests