Lightweight getMember resolution for aggregate types

[CyberShadow]

Generated by Claude Opus 4.6, usual disclaimers apply.

Implements the idea in #22541 (comment) .

Supersedes #22541, #22542, and #22545.

---

When __traits(getMember, T, name) is used inside allMembers loops on
template struct types, the normal path through expressionSemantic
triggers functionSemantic → functionSemantic3 (eager body compilation)
on member functions, causing circular dependency errors.

Add a lightweight fast path in getMember that resolves via sym.search()
and returns a DsymbolExp with the type pre-set, bypassing the cascade.
This covers both auto-return functions (inferRetType) and instantiated
functions in template structs (isInstantiated).

Also handle DsymbolExp in TemplateInstance_semanticTiargs so that
function symbols are stored directly as Dsymbols rather than being
passed to ctfeInterpret, which would fail.

[dlang-bot]

Thanks for your pull request, @CyberShadow!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

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#22546"

[CyberShadow]

@dkorpel What do you think about this one? I don't fully understand it, but it's green and the diff is small 😛

[tim-dlang]

I like the approach, but this can cause regressions, when the result of __traits(getMember, ...) is used in different ways, for example:

struct S(T)
{
    static void f()
    {
    }
}
pragma(msg, __traits(getMember, S!int, "f").mangleof);

With DMD v2.112.0 this prints _D17testlazygetmember__T1STiZQf1fFNaNbNiNfZv, but with this PR it only prints FZv.

CTFE also does not work anymore:

struct S(T)
{
    static int f()
    {
        return 42;
    }
}
pragma(msg, __traits(getMember, S!int, "f")());

DMD with this PR prints the following:

testlazygetmember.d(51): Error: cannot interpret `f` at compile time
pragma(msg, __traits(getMember, S!int, "f")());
            ^
testlazygetmember.d(51):        while evaluating `pragma(msg, f())`
pragma(msg, __traits(getMember, S!int, "f")());
^

[CyberShadow]

Argh, good catch. Let me have a look.

[CyberShadow]

@tim-dlang Latest revision adds more isDsymbolExp checks which fixes those two issues. Does this direction make sense?

[tim-dlang]

Yes, I think the direction makes sense. I worry that other traits, properties or is expressions could still have problems, but maybe the traits work, because they use the common function getDsymbolWithoutExpCtx, which is already modified by this PR.

Here is another failing test case:

struct S5(int line)
{
    static void badfunc()
    {
        undefined;
    }
}

static assert(!__traits(compiles,__traits(getMember, S5!__LINE__, "badfunc")));
static assert(!is(typeof(__traits(getMember, S5!__LINE__, "badfunc"))));