Correctly infer argument types for constructor calls

[mihaibudiu]

Signed-off-by: Mihai Budiu mbudiu@vmware.com
This PR is extracted from #3520.
It fixes a couple of bugs which were discovered using the list tests, but it really has nothing to do with lists.
#3520 should be rebased on top of this one.
The attached test used to fail before this fix.

[apinski-cavium]

Just some simplifications because of C++17. structured binding and initialization of pairs via {} syntax.

[apinski-cavium]

Since p4c is now compiling as C++17 you can just do:
return {objectType, arguments};
here.

[apinski-cavium]

auto [contType, newArgs] = checkExternConstructor(expression, simpleType->to<IR::Type_Extern>(),
                                                  expression->arguments);

(github did save my comment previously).

[mihaibudiu]

The bazel build system complains about your proposal:

frontends/p4/typeChecking/typeChecker.cpp:1130:14: warning: decomposition declaration only available with -std=c++1z or -std=gnu++1z
         auto [type, newArgs] = checkExternConstructor(decl, et, decl->arguments);
              ^

Anyone know where the language level is in the bazel scripts?

[jfingerh]

The test case and expected output looks reasonable to me. I have not made any attempt to review the changes to the compiler internals, so treat this approval with that qualification.

[jafingerhut]

Adding review from the account with write access, in case that helps make progress. Again, I am only able to knowledgeably review the test case and expected output files, which look good.

[rst0git]

LGTM

p4testdata/random.p4(55): [--Werror=type-error] error: AssignmentStatement

[rst0git]

It looks like this change breaks the random.p4 test in the eBPF backend:

$ p4c-ebpf -Ibackends/ebpf/runtime/p4include --target kernel -o ptf_out/random.c p4testdata/random.p4 --Wdisable=unused --max-ternary-masks 3 -DPORT0=4 -DPORT1=5 -DPORT2=6 -DPORT3=7 -DPORT4=8 -DPORT5=9 -DPSA_RECIRC=2 --xdp2tc=cpumap --arch psa;
p4testdata/random.p4(55): [--Werror=type-error] error: AssignmentStatement
        a.rand.f3 = r3.read();
                  ^
  ---- Actual error:
  /usr/local/share/p4c/p4include/psa.p4(676): Cannot cast implicitly type 'T' to type 'bit<16>'
  extern Random<T> {
                ^
  ---- Originating from:
  p4testdata/random.p4(55): Source expression 'r3.read' produces a result of type 'T' which cannot be assigned to a left-value with type 'bit<16>'
          a.rand.f3 = r3.read();
                      ^^^^^^^^^
  /usr/local/share/p4c/p4include/psa.p4(676)
  extern Random<T> {
                ^
p4testdata/random.p4(64): [--Werror=type-error] error: cast: action argument must be a compile-time constant
        default_action = do_forward((PortId_t) 5);
                                    ^^^^^^^^^^^^
p4testdata/random.p4(70): [--Werror=type-error] error: AssignmentStatement
        a.rand.f2 = r1.read();
                  ^
  ---- Actual error:
  /usr/local/share/p4c/p4include/psa.p4(676): Cannot cast implicitly type 'T' to type 'bit<16>'
  extern Random<T> {
                ^
  ---- Originating from:
  p4testdata/random.p4(70): Source expression 'r1.read' produces a result of type 'T' which cannot be assigned to a left-value with type 'bit<16>'
          a.rand.f2 = r1.read();
                      ^^^^^^^^^
  /usr/local/share/p4c/p4include/psa.p4(676)
  extern Random<T> {
                ^

[mihaibudiu]

@fruffy this fails in cpplint. However, 'make cpplint' no longer runs cpplint. Can we fix that so that is continues to work the same way it did? Also, from the current output logs I cannot understand where the failure is.

[mihaibudiu]

I see the actual error, but it's not easy to search for it, since the word "error" does not appear close to the problem.

[fruffy]

I can add a CMake target. In the meantime, tools/run-cpplint.sh should do the same thing.

[jfingerh]

@mbudiu-vmw Are you hoping for any other reviews for this? If so, it might be nice to ping them explicitly.