Endilll wrote:

> Do NOT drop the variant tests. Even the AST ones. We could move the AST ones 
> to codegen ones, but that just shifts the issue. We can make the AST checks 
> much more relaxed though. We need to check that the right function was 
> selected in the AST. For that we need the pseudo expression with the two 
> calls, and the declarations, preferably with their return statements. All 
> other fluff can go.

Sure, I'll preserve test coverage that variant AST dump tests provide.

> Generally, do not drop message tests, they are the only error emmiting tests 
> we have.

I guess there is misunderstanding here. The only reason message tests were 
brought up in this discussion is that they extensively use `variant` clause, so 
I wondered if their coverage overlaps with the coverage of variant AST dump 
tests. I was told this is not the case. Message tests I saw do not do AST dump 
matching, so there is no reason for me to touch them.

> I think the reduced AST checks still capture the essence, namely the clause 
> variables and the captured variables.

Does it mean you disagree with Alexey and Joseph that non-variant AST dump 
tests can be dropped entirely? For instance, it seems that `target` clause is 
extensively tested by CodeGen tests under `test/OpenMP`.

If you disagree with them, are all CHECK lines I preserved really needed? As an 
example, there is a lot of `ImplicitParamDecl` nodes for which I do not capture 
any details, but if someone ever adds another one, it will be painful to update 
the tests even after the reduction I did so far.

https://github.com/llvm/llvm-project/pull/200077
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to