[ggorman]

Stacked on #3061; the base switches to main once that merges.

The copy-back skip from #3050 evaluates the later write's guard at the earlier copy-back. If anything that guard reads (indices included) is written in between, the only firing write can be skipped. Single-writer reproducer (every Function has at most one firing write per step, slot and point):

cm = ConditionalDimension(name='cm', parent=time, relation='strict',
                          condition=And(CondEq(time, 3), T[0, 0, 0] < 3.5))
eqs = [Eq(f, T), Eq(T, T + 1), Eq(f, T + 1, implicit_dims=cm)]  # f save=4, T not saved

At time = 3 the increment makes cm false, so only Eq(f, T) fires. f[:, 1, 1] is [0, 1, 2, 3] with opt='noop' and [0, 1, 2, 0] with buffering on main. The indirect-guard variant (the guard reads T through an index Function) gives [2, 3, 4, 5] vs [2, 3, 4, 0].

The skip now applies only if nothing the later guard reads is written between the two copy-backs. Test: test_buffering.py::test_buffering_same_slot_skip[mutable|indirect]; both fail with only buffering.py reverted.

Per the #3050 discussion, the cases with two firing writes to a slot in one step are not used as evidence here.

🤖 Generated with Claude Code

[FabioLuporini]

@mloubout can this one be closed now?

[mloubout]

@FabioLuporini not sure, this was open after merging the prvious fix. This is still an odd "MFE" that writes two different values to the same saved slot so not sure what case this is supposed to represent.