Make the `verify` pass on by default

[SeanTAllen]

As we slowly move towards Pony 1.0.0, fixing issues in LLVM IR generation is an important consideration. We've turned on LLVM IR generation verification as a default in the compiler.

Hopefully this will generate reports for failures that are currently hidden in the wild.

[SeanTAllen]

Two of our runner tests currently fail verification. We'll need to fix those (and other errors that might be hidden by those failures) before this is merged.

[jemc]

Suggested change to make verification failure more friendly and make people aware that they can turn it off if they run into an error:

diff --git a/src/libponyc/codegen/genopt.cc b/src/libponyc/codegen/genopt.cc
index 970e43fc..0761075d 100644
--- a/src/libponyc/codegen/genopt.cc
+++ b/src/libponyc/codegen/genopt.cc
@@ -1446,6 +1446,8 @@ bool genopt(compile_t* c, bool pony_specific)
     if(LLVMVerifyModule(c->module, LLVMPrintMessageAction, &msg) != 0)
     {
       errorf(errors, NULL, "Module verification failed: %s", msg);
+      errorf_continue(errors, NULL,
+        "Please file an issue ticket. Use --noverify to bypass this error.");
       LLVMDisposeMessage(msg);
       return false;
     }

[SeanTAllen]

@jemc added.

[SeanTAllen]

This shouldnt be merged until we fix all the associated issues that cause CI to fail.

[SeanTAllen]

I've rebased this against main. I assume it will still fail CI, but we can at least see where we are at with this.

[jemc]

The only test failing (at least on the Linux build that I checked) is the full program test named with-return - it's getting the IR verification error that there is a "Terminator found in the middle of a basic block!".

We need to fix something about the interaction of an explicit return inside a with block. Should be fairly simple I think?

[SeanTAllen]

We need to fix something about the interaction of an explicit return inside a with block. Should be fairly simple I think?

Well, maybe? the issue is, we wouldn't want both returns if they end up together, but a return from a with is perfectly fine the issue is that it is the last statement and our "no explicit return except early doesn't cover this"

[SeanTAllen]

We are going to remove the with-return test for now as it is flawed. See #4464.

[SeanTAllen]

This is looking good, but I don't want to merge yet. I want to get a release that solves #4454