Feature Request [Quality of life]: Remove memory addresses/ids of objects from Op versioning

[TeoZosa]

Background

Right now, memory addresses in the repr of instantiated objects cause spurious diffs when versioning Ops in Weave. This makes it difficult to track meaningful changes between runs.

Workaround

I use the following utility method to remove at <memory_address> from reprs to eliminate those spurious diffs:

import re
def _strip_memory_address_from_repr(obj: object) -> None:
    original_repr = obj.__class__.__repr__
    obj.__class__.__repr__ = lambda self: re.sub(r" at 0x[0-9a-fA-F]+", "", original_repr(self))  # type: ignore[method-assign]

---

Question

Is there something more foundational that can be done to cover this issue globally?

Code?

Digging around the codebase, are these the appropriate places?

weave/weave/trace/serialize.py

Lines 98 to 111 in a76414b

	def stringify(obj: Any, limit: int = MAX_STR_LEN) -> str:
	"""This is a fallback for objects that we don't have a better way to serialize."""
	rep = None
	try:
	rep = repr(obj)
	except Exception:
	try:
	rep = str(obj)
	except Exception:
	rep = f"<{type(obj).__name__}: {id(obj)}>"
	if isinstance(rep, str):
	if len(rep) > limit:
	rep = rep[: limit - 3] + "..."
	return rep

weave/weave/trace/serialize.py

Lines 214 to 226 in a76414b

	def fallback_encode(obj: Any) -> Any:
	rep = None
	try:
	rep = repr(obj)
	except Exception:
	try:
	rep = str(obj)
	except Exception:
	rep = f"<{type(obj).__name__}: {id(obj)}>"
	if isinstance(rep, str):
	if len(rep) > MAX_STR_LEN:
	rep = rep[:MAX_STR_LEN] + "..."
	return rep

But I guess that wouldn't cover the case of a bad repr in which case an object's id could create a spurious diff as well, ex.

[gtarpenning]

@TeoZosa Thanks for the writeup, this is a great feature request/bug. We had already noted this internally as low-hanging fruit to improve the comparison page. As you have noted, these memory addresses are generated pretty deep in our code capture handling, which is rather delicate.

For now, looks like you have a decent workaround, but we will be working on this in the near future.

[TeoZosa]

@gtarpenning Got it, thanks for getting back to this so fast. Will keep my workaround for the time being. The team has been putting out some pretty major features recently and this one is definitely not as urgent.