The bisection didn't catch the correct commit actually.
My commit broke it, yes. but we reverted part of it later so that the code started working again.

The commit that breaks this is actually rBcbca71a7cff3
There is a mistake where the rctf that is passed to BLI_rctf_isect had its max and min values swapped.

The older code explicitly stated which variables it would set, however the raw order is not the same.
So the clean up changes set the xmin AND xmax to zero instead of xmin and ymin.

I think this is the correct fix:

diff --git a/source/blender/editors/space_node/node_gizmo.cc b/source/blender/editors/space_node/node_gizmo.cc
index 41d13976209..9aa4f299774 100644
--- a/source/blender/editors/space_node/node_gizmo.cc
+++ b/source/blender/editors/space_node/node_gizmo.cc
@@ -292,7 +292,7 @@ static void gizmo_node_crop_prop_matrix_set(const wmGizmo *gz,
   const bool ny = rct.ymin > rct.ymax;
   BLI_rctf_resize(&rct, fabsf(matrix[0][0]), fabsf(matrix[1][1]));
   BLI_rctf_recenter(&rct, (matrix[3][0] / dims[0]) + 0.5f, (matrix[3][1] / dims[1]) + 0.5f);
-  const rctf rct_isect{0, 0, 1, 1};
+  const rctf rct_isect{0, 1, 0, 1};
   BLI_rctf_isect(&rct_isect, &rct, &rct);
   if (nx) {
     SWAP(float, rct.xmin, rct.xmax);

Can you guys confirm?

That seems to work for me, sorry for the typo.

This works, although I'm increasingly of the opinion that we should avoid positional arguments when initializing struct members especially when moving C code to C++, it's bitten us a few times, is less readable - and can potentially backfire if struct members are ever recorded.

@Campbell Barton (campbellbarton) how would you like it to be initialized?

Until MSVC supports C99's struct initialization - this is the closest match AFAICS.

rctf rct_isect{};
rct_isect.xmin = 0;
rct_isect.xmax = 1;
rct_isect.ymin = 0;
rct_isect.ymax = 1;

• The {} are important, to zero members not assigned later (in the case of rctf it's not so important but as a convention I think we should stick to this as it's caused multiple bugs when moving from C to C++ (Unless leaving other members uninitialized is intended, in that case it can be left out).
• It's unfortinate the struct cant remain const but I don't think there are better alternatives.

@Sebastian Parborg (zeddb): for 3.0.1 corrective release, rB5703efab886d: Fix T94784: Crop node gizmo doesn't work doesnt apply (rBcbca71a7cff3: Cleanup: Move remaning node editor files to C++ is not in 3.0)

In T94784#1287023, @Sebastian Parborg (zeddb) wrote:

My commit broke it, yes. but we reverted part of it later so that the code started working again.

What was reverted partially? What should go into 3.0.1? Are we sure this is actually broken in blender-v3.0-release? (I cannot reproduce there... neither in 3.0 release for some reason, pretty sure I could repro initially though)

@Philipp Oeser (lichtwerk) my bad, the clean up commit from Hans is indeed not in the 3.0 branch.
So disregard this fix.