[lang] Return None when parent() exceeds its maximum depth

[k-ye]

Seems like https://github.com/taichi-dev/taichi/pull/1214/files#r440768343 wasn't addressed. I'd suggest us not to make root circular and keep the parent chain linear. Following parent until null is a fairly common op, e.g.

taichi/taichi/transforms/offload.cpp

Lines 87 to 95 in 37976e7

	auto leaf = for_stmt->snode;
	// make a list of nodes, from the leaf block (instead of 'place') to root
	std::vector<SNode *> path;
	// leaf is the place (scalar)
	// leaf->parent is the leaf block
	// so listgen should be invoked from the root to leaf->parent
	for (auto p = leaf; p; p = p->parent) {
	path.push_back(p);
	}

This PR also handles the case where n exceeds the hierarchy depth.

Related issue = #1214

[Click here for the format server]

---

[codecov]

Codecov Report

Merging #1319 into master will increase coverage by 0.08%.
The diff coverage is 90.24%.

@@            Coverage Diff             @@
##           master    #1319      +/-   ##
==========================================
+ Coverage   66.41%   66.49%   +0.08%     
==========================================
  Files          37       37              
  Lines        5133     5146      +13     
  Branches      931      932       +1     
==========================================
+ Hits         3409     3422      +13     
  Misses       1562     1562              
  Partials      162      162              

Impacted Files	Coverage Δ
python/taichi/lang/transformer.py	93.63% <ø> (ø)
python/taichi/misc/gui.py	24.65% <0.00%> (ø)
python/taichi/lang/matrix.py	90.70% <88.88%> (+0.03%)	⬆️
python/taichi/lang/expr.py	89.82% <91.66%> (ø)
python/taichi/lang/snode.py	87.50% <94.73%> (+1.47%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 477b1e1...e6bc6db. Read the comment docs.

[archibate]

Took a quick look, lgtmig!

[yuanming-hu]

Thank you! LGTM.

[yuanming-hu]

Offtopic discussion: maybe we should make ti.root a real SNode instead of ti.Root?

[archibate]

Offtopic discussion: maybe we should make ti.root a real SNode instead of ti.Root?

Agree.
If make ti.root a SNode too hard for you, you may at least make SNode a base class of Root.

[k-ye]

Offtopic discussion: maybe we should make ti.root a real SNode instead of ti.Root?

Actually, probably not. I think root was implemented this way to do lazy-evaluation, see

taichi/python/taichi/lang/impl.py

Lines 250 to 253 in 13462ff

	def __getattribute__(self, item):
	import taichi as ti
	ti.get_runtime().create_program()
	root = SNode(ti.get_runtime().prog.get_root())

While we can say one should only access ti.root after ti.init(), we have to define root before ti.init(). This initialization requires one level of indirection.. OTOH, changing to ti.root() may solve the problem.

As a side effect, I found it useless to define parent() on root, because it's not called anyway.

[yuanming-hu]

Offtopic discussion: maybe we should make ti.root a real SNode instead of ti.Root?

Actually, probably not. I think root was implemented this way to do lazy-evaluation, see

taichi/python/taichi/lang/impl.py

Lines 250 to 253 in 13462ff

	def __getattribute__(self, item):
	import taichi as ti
	ti.get_runtime().create_program()
	root = SNode(ti.get_runtime().prog.get_root())

While we can say one should only access ti.root after ti.init(), we have to define root before ti.init(). This initialization requires one level of indirection.. OTOH, changing to ti.root() may solve the problem.

As a side effect, I found it useless to define parent() on root, because it's not called anyway.

Good point. How about just setting ti.root in ti.init()? We are enforcing users to call ti.init in the very beginning of every taichi program anyway. It's fine to let ti.root be attribute not found before ti.init.

[k-ye]

Allow me to merge this one first. I've encountered some problems when modifying global root in the module level. Seems like I've encountered the same problem in https://stackoverflow.com/questions/3536620/how-to-change-a-module-variable-from-another-module.

[yuanming-hu]

Allow me to merge this one first. I've encountered some problems when modifying global root in the module level. Seems like I've encountered the same problem in https://stackoverflow.com/questions/3536620/how-to-change-a-module-variable-from-another-module.

Ah you are right... I think that's the reason why I did a Root class a few months ago... (Sorry about my terrible memory...)