Fix handling of int<0> constants, fix warning category

[vlstill]

Fixes a crash when the compiler tries to create int<0> constant:

Compiler bug triggered, please report this error:
exception type: boost::wrapexcept<std::out_of_range>
error: Can not shift by a negative value.

Also fixes the warning category of bit<N> overflow (which was mismatch and now is overflow).

[asl]

How can such values appear?

[vlstill]

How can such values appear?

Altera backend certainly hits it. It might not be directly from the code, but I think even a manual creation of such constant must not trigger exception in boost.

[asl]

Altera backend certainly hits it. It might not be directly from the code, but I think even a manual creation of such constant must not trigger exception in boost.

I am actually thinking, if creation of such types is a result of some underlying issue? Shouldn't int<0> be just disallowed?

[jafingerhut]

Altera backend certainly hits it. It might not be directly from the code, but I think even a manual creation of such constant must not trigger exception in boost.

I am actually thinking, if creation of such types is a result of some underlying issue? Shouldn't int<0> be just disallowed?

Section 7.1.6.2 "Unsigned integers (bit-strings)" of the P4 language spec includes this sentence: "Bitstrings with width 0 are allowed; they have no actual bits, and can only have the value 0." This was added to the language spec in Apr 2021 from some git log searches.

For type int<W> the spec is not quite as explicit about W=0, but it does say this: "W must be an expression that evaluates to a local compile-time known (see Section [Section 18.1]) value that is a non-negative integer." That allows 0.

[ChrisDodd]

I am actually thinking, if creation of such types is a result of some underlying issue? Shouldn't int<0> be just disallowed?

bit<0> was added to allow code like

header foo {
    bit<(DEFINED_CONST)> data;
    bit<(-(DEFINED_CONST) & 7)> pad;  // pad out to byte boundary
};

where DEFINED_CONST is defined by a macro or -D argument

int<0> was added for consistency

[asl]

int<0> was added for consistency

Then probably the meaning should be clarified? Should the spec explicitly say that bit<0> and int<0> behave similarly?

[ChrisDodd]

should use warning instead of P4::warning for consistency with below.