Fix Issue 21861 - ctfe fails when a nested enum or function has a UDA

[BorisCarvajal]

If declaration is an UDA just eval the symbol it contains.

[dlang-bot]

Thanks for your pull request, @BorisCarvajal!

Bugzilla references

Auto-close	Bugzilla	Severity	Description
✓	21861	normal	ctfe fails when a nested enum or function has a UDA

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub run digger -- build "master + dmd#12472"

[MoonlightSentinel]

Why only one declaration?

[BorisCarvajal]

Currently, at function scope the parser only allows one declaration after UDAs.

[MoonlightSentinel]

Fair enough.
But the issue should rather be fixed by changing the AttribDeclaration branch below (which rejects enums ATM and hence causes the bug). The current implementation doesn't care about the UDA and only considers the AttribDeclaration part.

[MoonlightSentinel]

Possible solution (unless im missing something):

diff --git a/src/dmd/dinterpret.d b/src/dmd/dinterpret.d
index 8676f3b59..3d383b33d 100644
--- a/src/dmd/dinterpret.d
+++ b/src/dmd/dinterpret.d
@@ -2313,12 +2313,8 @@ public:
             AttribDeclaration ad = e.declaration.isAttribDeclaration();
             if (ad && ad.decl && ad.decl.dim == 1)
             {
-                Dsymbol sparent = (*ad.decl)[0];
-                if (sparent.isAggregateDeclaration() || sparent.isTemplateDeclaration() || sparent.isAliasDeclaration())
-                {
-                    result = null;
-                    return; // static (template) struct declaration. Nothing to do.
-                }
+                result = null;
+                return; // static declaration. Nothing to do.
             }

             // These can be made to work, too lazy now

[BorisCarvajal]

That branch just returns null, but if there is a variable declaration inside the UDA it needs to be evaluated.
(I modified the test to check for that)
I kept that code untouched, although I think it should be revisited some day.

[MoonlightSentinel]

Ahh, good catch. Probably still better to simply check for any AttribDeclaration instead of UDAs:

diff --git a/src/dmd/dinterpret.d b/src/dmd/dinterpret.d
index 8676f3b59..85c1858a2 100644
--- a/src/dmd/dinterpret.d
+++ b/src/dmd/dinterpret.d
@@ -2225,6 +2225,11 @@ public:
             printf("%s DeclarationExp::interpret() %s\n", e.loc.toChars(), e.toChars());
         }
         Dsymbol s = e.declaration;
+        if (auto ad = s.isAttribDeclaration())
+        {
+            assert(ad.decl && ad.decl.dim == 1);
+            s = (*ad.decl)[0];
+        }
         if (VarDeclaration v = s.isVarDeclaration())
         {
             if (TupleDeclaration td = v.toAlias().isTupleDeclaration())

Unless we know for sure that other AD's can never reach this code.

EDIT: Albeit that might require a loop instead of a plain if (if nesting can occur)