Using enum as a type generates syntax error.

[hanw]

 enum HashAlgo {
    CRC32
}
extern Hash<Algo, O> {
    Hash();
    O getHash<T, D, M>(in T base, in D data, in M max);
}
control ingress<H, M> (inout H hdr, inout M meta);
package PSA<H, M> (ingress<H,M> ig);
// user program
struct ParsedHeaders {
    bit<32> hdr;
}
struct Metadata {
    bit<32> md;
}
control MyIngress (inout ParsedHeaders p, inout Metadata m) (Hash<HashAlgo.CRC32, bit<32>> h) {
    action compute_hash() {
        bit<32> b = p.hdr;
        tuple<bit<32>, bit<32>> d = {p.hdr, m.md};
        bit<32> max = m.md;
        bit<32> hash;
        hash = h.getHash(b, d, max);
    }
    apply{
        compute_hash();
    }
}
Hash<HashAlgo.CRC32, bit<32>> hash;
MyIngress(hash) ig;
PSA(ig) main;

Issue:
action_selector does not accept an enum as the type for instantiation.

Error message:

../test/gtest/arch_test.cpp(559):syntax error, unexpected '.', expecting ',' or '>'
        control MyIngress (inout ParsedHeaders p, inout Metadata m) (Hash<HashAlgo.

[mihaibudiu]

You see what a terrible error message this is from gtest, it does not really tell you where the error is.

I cannot reproduce this error, I am getting the following:

issue492.p4(30): warning: Table test1 does not have an `default_action' property
    table test1 {
          ^^^^^
issue492.p4(35): warning: selector shadows selector
        selector = action_selector(HashAlgorithm.crc16, 32w1024, 32w10);
        ^^^^^^^^
issue492.p4(12)
    selector
    ^^^^^^^^

[sethfowler]

What do you mean "from GTest"? That's just p4c's normal error reporting. The line that's being reported is correct.

[mihaibudiu]

../test/gtest/arch_test.cpp(559):syntax error, unexpected '.', expecting ',' or '>'

[sethfowler]

Yes, that is correct. That error message points to the line where the error is in the P4 program, which is also a line in a .cpp file since the P4 program is defined inline in the test.

[mihaibudiu]

Yes, but I don't have that C++ program, since the code is not checked-in.
How about the error message I posted which happens if you don't use gtest?

[jnfoster]

Mihai, I am on 482a3bd.

I see this error message for Han's original code snippet, which is like yours but also dumps core:

han.p4(30): warning: Table test1 does not have an `default_action' property
    table test1 {
          ^^^^^
han.p4(35): warning: selector shadows selector
        selector = action_selector(HashAlgorithm.crc16, 32w1024, 32w10);
        ^^^^^^^^
han.p4(12)
    selector
    ^^^^^^^^
han.p4(42): error: PackageBlock: Expected 6 arguments for main package
PSA(ig) main;
        ^^^^
terminate called after throwing an instance of 'Util::CompilerBug'
  what():  COMPILER BUG: ../ir/ir.cpp:172
No parameter named eg

Aborted (core dumped)

With the newer code since he updated the issue, I see the kind of error he first reported:

han.p4(17):syntax error, unexpected '.', expecting ',' or '>'
control MyIngress (inout ParsedHeaders p, inout Metadata m) (Hash<HashAlgo.
                                                                          ^
error: 1 errors encountered, aborting compilation

[mihaibudiu]

I don't understand where this error is coming from. Look at the code, there is no line like the one quoted in the sources. There is no identifier HashAlgo.

[jnfoster]

He updated the code to

 enum HashAlgo {
    CRC32
}
extern Hash<Algo, O> {
    Hash();
    O getHash<T, D, M>(in T base, in D data, in M max);
}
control ingress<H, M> (inout H hdr, inout M meta);
package PSA<H, M> (ingress<H,M> ig);
// user program
struct ParsedHeaders {
    bit<32> hdr;
}
struct Metadata {
    bit<32> md;
}
control MyIngress (inout ParsedHeaders p, inout Metadata m) (Hash<HashAlgo.CRC32, bit<32>> h) {
    action compute_hash() {
        bit<32> b = p.hdr;
        tuple<bit<32>, bit<32>> d = {p.hdr, m.md};
        bit<32> max = m.md;
        bit<32> hash;
        hash = h.getHash(b, d, max);
    }
    apply{
        compute_hash();
    }
}
Hash<HashAlgo.CRC32, bit<32>> hash;
MyIngress(hash) ig;
PSA(ig) main;

[mihaibudiu]

@jnfoster: the compiler is dumping core because you can't compile this with p4c-bm2-ss; this is not a program for v1model.
I will file a separate bug for that.

[hanw]

I pasted the wrong test case in the original email. Thanks Nate for pointing that out.

[mihaibudiu]

This example is not syntactically correct; you can only use types as type arguments when specializing a type. However, HashAlgo.CRC32 is not a type, it is a constant.

[cc10512]

However, @hanw's example is using HashAlgo as the the type, and HashAlgo.CRC32 as a constant parameter. That doesn't work either.

[mihaibudiu]

The example posted won't work in Java or C# either; the generics in P4 are modeled after those languages.
A constant can be used perhaps as a constructor parameter, but not as a template parameter.
Maybe there is another example I have not seen?

[cc10512]

Here is the source for all the examples that @hanw is trying to get through the compiler: https://github.com/cc10512/p4c/blob/cc/psa/p4include/psa.p4

[mihaibudiu]

There are generic externs in v1model.p4. In particular, register is a working example.
If you want a constant value, pass it as a constructor parameter.
In our language both templates and constructor arguments are evaluated at compile-time, so the expressivity is the same.

[jnfoster]

Can we safely close this?

[hanw]

Closed by passing constant value as a constructor parameter.