Boolean function parser breaks for alternative notation of AND, OR

[RenWal]

The wiki states:

In general, Boolean functions support NOT ("!", "'"), AND ("&", "*", " "), OR ("|", "+"), and XOR ("^") operations as well as brackets ("(", ")") when being parsed from a string.

The following parses OK:

hal_py.BooleanFunction.from_string("(A & B) | C")

The following should parse according to the documentation, but doesn't:

hal_py.BooleanFunction.from_string("(A * B) + C")

The latter produces the following error:

[06.11.2024 19:50:24] [python_context] [error] [hal/src/python_bindings/bindings/boolean_function.cpp:1251] [hal/src/netlist/boolean_function.cpp:1079] could not parse Boolean function from string: no parser available for '(A...

Can we fix the parser, or update the docs?

[RenWal]

The reason seems to be that HAL switches to the Liberty parser when encountering the "*" or "+" symbols. While the HAL-internal parser has no issue with spaces, the Liberty parser treats spaces as AND. Thanks @fyrbiak for pointing this out.

Hence, what does work is the following:

hal_py.BooleanFunction.from_string("(A*B)+C")

I updated the documentation to include a caution about spaces in equations.

[RenWal]

This issue is not fixed, and it breaks the Liberty gate library parser.

# parses OK
hal_py.BooleanFunction.from_string("SE*SI")
hal_py.BooleanFunction.from_string("SE & SI")
hal_py.BooleanFunction.from_string("SE SI")
hal_py.BooleanFunction.from_string("SE+SI")

# does not parse
hal_py.BooleanFunction.from_string("SE * SI")
hal_py.BooleanFunction.from_string("SE + SI")
hal_py.BooleanFunction.from_string("SE  SI")

Attempting to parse a gate library containing the Boolean function ((SE * SI) + (D * !SE)) causes parsing to fail due to this bug, which also means that netlists shipped with accompanying .lib files using that notation cannot be loaded.

[RenWal]

This issue arises because the Liberty reference manual is sloppy in how it defines whitespace handling, and we follow the rules as written and not as intended. The manual (Version R-2020.09, September 2020) says:

Operator	Description
&	logical AND
space	logical AND

Taken literally, this means that strictly following the standard's wording we're supposed to tokenize the expression SE * SI as SE AND AND AND SI, which we do, but of course that makes no sense.

What the manual presumably means is that the absence of an explicit operator should be parsed as an implicit AND, and not that every space is to be treated as an AND.

The solution here will be to remove the space handling from our lexer in src/netlist/boolean_function/parser_liberty.cpp and instead handle this case on a grammar level, i.e. "if you encounter two consecutive variable lexemes, then imply an AND token between them".