Does anybody know where in the DMD source I can figure out if the return value of a function call is used or not?

I'm trying to figure out how to implement automatic detection of unused return values from calls to strictly pure functions.

The closest I have come is the function functionParameters() in expression.c

Ideas anyone?

On Thursday, 27 February 2014 at 23:38:34 UTC, Nordlöw wrote:
> Does anybody know where in the DMD source I can figure out if the return value of a function call is used or not?
>
> I'm trying to figure out how to implement automatic detection of unused return values from calls to strictly pure functions.
>
> The closest I have come is the function functionParameters() in expression.c
>
> Ideas anyone?

The return type of course also must be non-void in order for the warning to printed.

Nordlöw:

> Does anybody know where in the DMD source I can figure out if the return value of a function call is used or not?
>
> I'm trying to figure out how to implement automatic detection of unused return values from calls to strictly pure functions.
>
> The closest I have come is the function functionParameters() in expression.c
>
> Ideas anyone?

See:

https://d.puremagic.com/issues/show_bug.cgi?id=3882

Bye,
bearophile

On 02/28/2014 12:38 AM, "Nordlöw" wrote:
> Does anybody know where in the DMD source I can figure out if the return
> value of a function call is used or not?

I'm not familiar with the DMD source code, but a good way to proceed is usually to grep for existing error messages that are somehow related. In this case, DMD already complains about some cases of unused values:

$ grep "no effect" *.c
declaration.c:       error("storage class 'auto' has no effect if type is not inferred, did you mean 'scope'?");
expression.c:            e = new CastExp(loc, e, Type::tvoid);   // avoid "has no effect" error
mtype.c:        // substitution has no effect on function pointer type.
sideeffect.c:                     * Don't complain about an expression with no effect if it was cast to void
sideeffect.c:                /* Don't complain about calling functions with no effect,
sideeffect.c:                error("%s has no effect", toChars());
sideeffect.c:                error("%s has no effect in expression (%s)",

This leads directly to sideeffect.c: Expression::discardValue(), which seems to be the place where you want to insert your detection code.

> https://d.puremagic.com/issues/show_bug.cgi?id=3882
>
> Bye,
> bearophile

Did you set this as WONTFIX because of too many warnings from functions that may throw or just do asserts such as unittests?

If so does anyone see any way to restrict warnings even further for example by checking if a function will call assert or nothrow or do a debug print?

I believe I found a good solution to Issue 3882:
https://d.puremagic.com/issues/show_bug.cgi?id=3882

It works as expected and I found two bugs in my code with it.

My current solution is to add the following at line 147 in dmd/src/sideeffect.c:

      case TOKcall:
            /* Issue 3882: Don't complain about calling functions with no effect,
             * because purity and nothrow are inferred, and because some of the
             * runtime library depends on it. Needs more investigation.
             */
            if (global.params.warnings && !global.gag)
            {
                if (e->type->ty == Tvoid)
                {
                    /* TODO: Restrict this message to call hierarchies that
                     * never call assert (and or not called from inside
                     * unittest blocks) */
                    // e->warning("Call %s with void return has no effect", e->toChars());
                }
                else
                {
                    e->warning("Call %s with no effect discards return value", e->toChars());
                }
            }
            return;

Does it suffice to just do a pull request referring to Issue 3882.

/Per

Nordlöw:

> Did you set this as WONTFIX because of too many warnings from functions that may throw or just do asserts such as unittests?

It's a "RESOLVED LATER". I think it was not me to tag it like that. And it was tagged like that because of too many warnings in templated functions.

Bye,
bearophile

On Friday, 28 February 2014 at 10:40:04 UTC, bearophile wrote:
> Nordlöw:
>
>> Did you set this as WONTFIX because of too many warnings from functions that may throw or just do asserts such as unittests?
>
> It's a "RESOLVED LATER". I think it was not me to tag it like that. And it was tagged like that because of too many warnings in templated functions.
>
> Bye,
> bearophile

My tested application is massive I got not "false positives".

I can do some more tests...

/Per

> that. And it was tagged like that because of too many warnings in templated functions.

Were those warnings emitted also by non-void return functions?

/Per