LDC sets both never-inline and always-inline on inline functions containing DMD-style inline assembly. about ldc HOT 3 CLOSED

A proposed fix: #4631

from ldc.

If a D function has pragma(inline, true) applied to it, or a C function is inline

Ideally, we'd be able to discriminate these 2 cases, and error out properly for an invalid pragma(inline, true) (upstream, in the frontend directly). But if importC translates inline to the same AST representation, then that's firstly not really correct, but probably works in most cases, and so I agree that we should just ignore it.

There's a Dsymbol::isCsymbol() method that we could use to restrict this 'hack' to functions in C files only.

from ldc.

If a D function has pragma(inline, true) applied to it, or a C function is inline

Ideally, we'd be able to discriminate these 2 cases, and error out properly for an invalid pragma(inline, true) (upstream, in the frontend directly). But if importC translates inline to the same AST representation, then that's firstly not really correct, but probably works in most cases, and so I agree that we should just ignore it.

There's a Dsymbol::isCsymbol() method that we could use to restrict this 'hack' to functions in C files only.

Mhm, ImportC just sets FuncDeclaration.inlining to PINLINE.always for inline functions (though it doesn't emit a PragmaDeclaration).

I've changed #4631 to ignore always-inline when DMD-style inline assembly is present only for C functions.

from ldc.