ti.ndrange does not check for number of parameters

[houkensjtu]

Describe the bug
ti.ndrange does not report an error or warning even when the number of parameters does not match.

To Reproduce

import taichi as ti

ti.init(arch=ti.cpu)

@ti.kernel
def iter():
    for i in ti.ndrange(1, nx):  # Taichi scope
        print(i)

iter()  # 0,1,2,3,4,5,6,7

for i in ti.ndrange(1, nx):  # Python scope
    print(i)  # (0,0) (0,1) (0,2) (0,3) (0,4) (0,5) (0,6) (0,7)

As shown in the code example above, a user who is unfamiliar with ti.ndrange's syntax might expect the printing result of iter() to be 1, 2, 3, 4, 5, 6, 7; however, since ti.ndrange requires a parentheses for each index and the users forgot to include one, the results might be unexpected. It's probably better to give a warning or error message when the number of index does not match the parameters in ti.ndrange.

It's also confusing that the same syntax yields different results for Taichi scope and Python scope. If this kind of mismatch is allowed (with a warning but not an error), then the current behavior in Python scope seems to be more reasonable than the results in Taichi scope.

[lin-hitonami]

In fact the use of the ndrange-for in Python scope is a use of grouped ndrange-for... I've come up with an idea that we treat the use of ndrange-for with only one loop variable but has many dimensions automatically as a grouped ndrange-for (maybe we can do the same thing on the struct for too). In this way the syntax of Python and Taichi can be united, and the ti.grouped decorator can be deprecated. WDYT @yuanming-hu @ailzhang @strongoier

[yuanming-hu]

For ndrange deprecating ti.grouped may be an option.

For struct-fors, if we deprecate ti.grouped, what would be the type for i in for i in x where x is a 1-D field? Is that a scalar or a 1D vector? :-)

If we only deprecate it for ndrange then there may be some inconsistency between two for loops.

[lin-hitonami]

For struct-fors, if we deprecate ti.grouped, what would be the type for i in for i in x where x is a 1-D field? Is that a scalar or a 1D vector? :-)

What about a subclass of ti.Expr which can have subscript? We can let i[0] return i?

[yuanming-hu]

TBH, that would be a little tricky to understand by users, given it's neither a scalar nor a vector...

[lin-hitonami]

TBH, that would be a little tricky to understand by users, given it's neither a scalar nor a vector...

It is actually a scalar and a vector (with one element).

[lin-hitonami]

Given this only happens when there is only one loop variable and the field has only one dimension, I think it is not that hard to understand......

[yuanming-hu]

While it may be understandable by some of the users, it does break the uniformity. If we use ti.grouped then it would be crystal clear that you get a vector when you use ti.grouped, and scalars when you don't.

[strongoier]

I agree that we should not mix scalar and vector together.