Operator precedence of `<=` misleading when using `if-then`

In Python <= has higher precedence than not, or, and, if-then, and lambda. In cocotb <= is reused as an assignment operator, which leads to some highly surprising behavior when compared to the “normal” assignment operator =.

	Example	Expected	Actual
1	a <= b or c	a <= (b or c)	(a <= b) or c	Edit: Mitigation implemented, see below
2	a <= b and c	a <= (b and c)	(a <= b) and c	Edit: Mitigation implemented, see below
3	a <= b if c else d	a <= (b if c else d)	(a <= b) if c else d

I recently had my third encounter with this behavior. Although I had seen it twice before, it still took me some time to realize it was an operator precedence issue thanks to the pervasiveness of “normal” assignments.

Partial solution

Edit: this has already been implemented. With a recent version of cocotb only `if-then` is still an issue. Click to expand original text.

To prevent (1) and (2) <= could return a singleton DontUse like:

class DontUse:
    def __bool__(self):
        raise ValueError("Don't use me like this dummy")

DontUse = DontUse()

This would make Python throw the error when trying to evaluate or and and. There’s no such workaround available for if-then AFAIK.

Full solution

I don’t think there’s much we can do. I think using <= should be deprecated and eventually removed entirely. There’s very little benefit, but it can cause some pretty annoying debug sessions. Assignment on simulator objects should probably replace <='s current function.

Old	New
dut.a <= b if c else d	dut.a = b if c else d
a <= b if c else d	a.value = b if c else d

The good news is that there’s a good upgrade path:

1. Deprecation warning on <=
2. Hard error in __le__
3. Removal of __le__ (?)

Workarounds

1. Use parentheses: a <= (b if c else d)
2. Use .value: a.value = b if c else d

[eric-wieser]

We already implement the partial solution you suggest (as of #766, cocotb 1.2.0):

https://github.com/cocotb/cocotb/blob/fba7b854b00375500552ff8804757f468152b111/cocotb/handle.py#L476-L483

https://github.com/cocotb/cocotb/blob/fba7b854b00375500552ff8804757f468152b111/cocotb/handle.py#L418-L434

[martijnbastiaan]

Works for me. It’s non-surprising, performant, and solves this issue. It’s slightly annoying that it makes testbenches more verbose, but I 💯% prefer that over the status quo.