fix(parse): make template-argument-list optional

[JohelEGP]

Resolves #469.

Testing summary.

100% tests passed, 0 tests failed out of 508

Total Test time (real) = 120.76 sec

Acknowledgements.

• Easily tested thanks to https://github.com/modern-cmake/cppfront.
• The commit message uses Conventional Commits 1.0.0.

[hsutter]

Looks pretty good, thanks! Did you also check the other places template_arg gets used to ensure that the code will handle an empty list correctly?

For example, unqualified_id_node::visit should probably still emit the template_arms_tag even though no arguments will follow, so that the visit doesn't lose the empty list.

Also, does lowering emit the empty list? I don't see a .cpp result for the regression test, and I wouldn't expect it seeing cppfront.cpp line 1669. These probably should take the cue that a non-default .open_angle means there's a list, and allow an empty list?

[JohelEGP]

Did you also check the other places template_arg gets used to ensure that the code will handle an empty list correctly?

No.
Testing didn't catch anything.
I'll have to investigate that.

Also, does lowering emit the empty list?

Seems like it doesn't.
The added test happens to work without it.

For example, unqualified_id_node::visit should probably still emit the template_arms_tag even though no arguments will follow, so that the visit doesn't lose the empty list.

I don't see a .cpp result for the regression test, and I wouldn't expect it seeing cppfront.cpp line 1669. These probably should take the cue that a non-default .open_angle means there's a list, and allow an empty list?

I'll have to investigate those places.

I don't see a .cpp result for the regression test,

I have to investigate why https://github.com/modern-cmake/cppfront
isn't picking up the tests I add
(there's always 250 tests on main and my branches).

I do test it manually in my developer environment.
I suppose I could just copy the generated Cpp1 source back to regression-tests.

[hsutter]

I suppose I could just copy the generated Cpp1 source back to regression-tests.

Yes, to regression-tests/test-results please. The generated Cpp1 should be checked in with each new test. That should happen automatically if you re-run all of the regression tests to make sure there are no diffs, using the test-results/run-tests.bat or equivalent .sh.

[JohelEGP]

That should happen automatically if you re-run all of the regression tests to make sure there are no diffs, using the test-results/run-tests.bat or equivalent .sh.

Unfortunately, I'm on Linux.
https://github.com/modern-cmake/cppfront does it platform-agnostic.
As I suspected, it only picks up tests that have an .output in test-results.
I'll see about enhancing it to generate all those, as it's convenient for developers.

[JohelEGP]

Testing didn't catch anything.

Of course, silly me. That's why I made this PR.

The added test happens to work without it.

Looks like I didn't actually run it.

I checked P2601 To make redundant empty angle brackets optional for class template argument lists
for a context that requires <> and enhanced the test with it.

[JohelEGP]

These probably should take the cue that a non-default .open_angle means there's a list, and allow an empty list?

Applied.

[JohelEGP]

100% tests passed, 0 tests failed out of 508

FYI, these are the tests under regression-tests.
Currently, each one split into codegen and build tests.
codegen (invokes cppfront),
codegen/check (checks the outputs with that under test-results), and
codegen/update (copies the outputs to under test-resutls in developer mode).
Excluding the *-error tests,
build (invokes the Cpp1 compiler),
build/execute (executes the compiled Cpp2 program), and
build/update (copies the outputs to under test-resutls/compiler-major_version in developer mode).

It has become evident that I'm missing build/check analogous to codegen/check.

[hsutter]

Looks good, thanks!