Re: [PATCH for Dlang support] d: make enum SymbolKind idiomatic D

[H. S. Teoh]

On Thu, Oct 01, 2020 at 02:40:25AM +0300, Adela Vais wrote:
> Hello,
> 
> I realized that I was still triggering the GC within the toString
> function.  Fixed now.
[...]

Here's an idiom you might find useful for catching things like hidden
unwanted GC allocations:

        struct S {
                string x;
                void toString(W)(W sink) {
                        sink(x ~ "a");
                }

                @nogc unittest {
                        S s;
                        static void sink(const(char)[]) {}
                        s.toString(&sink);
                }
        }

The @nogc on the unittest ensures that it will not compile if toString
has any hidden GC allocations. But it will not stop the user from
calling toString with an allocating sink, as it would if we had put
@nogc on toString itself.  By not putting @nogc on toString, we allow
the user to choose whether or not they want to use an allocating sink or
not; the @nogc on the unittest ensures that toString itself will not
introduce GC dependency, thus making toString compatible with both.

(Note: for this to work, the code must be compiled with -unittest. I
highly, highly recommend building with -unittest as part of the test
suite.)


T

-- 
Give me some fresh salted fish, please.