Issue 1673 - Implement the isTemplate trait.

[ghost]

http://d.puremagic.com/issues/show_bug.cgi?id=1673

A rework of #1239, this time implementing an isTemplate trait.

@9rnsr @yebblies: Are the two checks in bool isTemplate(Dsymbol *s) good enough?

[yebblies]

Are the two checks in bool isTemplate(Dsymbol *s) good enough?

No idea, sorry. Kenji will know.

[yebblies]

I'm not sure this is worth putting in a function for only one use.

[ghost]

I was thinking maybe future code might have a use for it. I'm fine either way. Should I remove it?

[yebblies]

Nah I don't feel that strongly about it.

[yebblies]

I can't wait to delegatize this damn file.

[9rnsr]

Are the two checks in bool isTemplate(Dsymbol *s) good enough?

Template definitions can be overloaded with functions. So do you think how __traits(isTemplate) should work for the cases?

void foo() {}
void foo(T)(T) {}
enum bool x = __traits(isTemplate, foo);

Current code will return false with the case. To fix it, you can use overloadApply function.

[ghost]

@9rnsr: Thanks! Updated the code and the test-case, let me know if I've missed something.

[9rnsr]

toAlias case is unnecessary because it is already handled in overloadApply

[ghost]

Thanks! Btw, is there any way to make fpisTemplate stop the overloadApply early? I'm using a if (p->found) check to return 0 early, but this doesn't stop the overloadApply.

[ghost]

I guess I could return 1?

[ghost]

Ah I think I can implement this in a simpler way:

static int fpisTemplate(void *param, Dsymbol *s)
{
    if (s->isTemplateDeclaration())
        return 1;

    return 0;
}

// edited
bool isTemplate(Dsymbol *s)
{
    if (s->isTemplateDeclaration() || s->toAlias()->isTemplateDeclaration())
        return true;

    return overloadApply(s, NULL, &fpisTemplate);
}

/*
bool isTemplate(Dsymbol *s)
{
    if (s->isTemplateDeclaration() || s->toAlias()->isTemplateDeclaration())
        return true;

    FuncDeclaration *f = s->isFuncDeclaration();
    if (!f) return false;

    return overloadApply(f, NULL, &fpisTemplate);
}
*/

This seems to work locally. @9rnsr what do you think?

Edit: Updated with your latest comment.

[ghost]

It looks like I can even use this:

bool isTemplate(Dsymbol *s)
{
    /*
    if (s->isTemplateDeclaration() || s->toAlias()->isTemplateDeclaration())
        return true;
    */

    return overloadApply(s, NULL, &fpisTemplate);
}

[9rnsr]

I don't like to use wrapper template in the compiler test suite. This test suite tests only the case that __traits(isTemplate) gets an aliased symbol T.

[ghost]

Ok, will fix.

[9rnsr]

More better is:

bool isTemplate(Dsymbol *s)
{
    return overloadApply(s, NULL, &fpisTemplate) != 0;
}

!= 0 would be necessary for DDMD conversion.

[ghost]

More better is:

If I remove those calls I get errors:

static int fpisTemplate(void *param, Dsymbol *s)
{
    if (s->isTemplateDeclaration())
        return 1;

    return 0;
}

bool isTemplate(Dsymbol *s)
{
    /* if (s->isTemplateDeclaration() || s->toAlias()->isTemplateDeclaration())
        return true;

    FuncDeclaration *f = s->isFuncDeclaration();
    if (!f) return false; */

    return overloadApply(s, NULL, &fpisTemplate) != 0;
}

Output:

test.d(17): Error: template instance test1673.Foo!int is aliased to a function
test.d(21): Error: template instance test1673.Bar!int is aliased to a function
test.d(23): Error: template instance test1673.Bar!int.Doo!int is aliased to a function
test.d(17): Error: template instance test1673.Foo!int is aliased to a function
test.d(21): Error: template instance test1673.Bar!int is aliased to a function
test.d(23): Error: template instance test1673.Bar!int.Doo!int is aliased to a function

I seem to need both commented-out sections in order for the test-suite to pass. Test-suite:

module test1673;

template Foo(T...) { }

template Bar(T...)
{
    template Doo(T...)
    {
    }
}

template Tuple(T...) { alias Tuple = T; }

void main()
{
    static assert( __traits(isTemplate, Foo));
    static assert(!__traits(isTemplate, Foo!int));
    static assert(!__traits(isTemplate, main));

    static assert( __traits(isTemplate, Bar));
    static assert(!__traits(isTemplate, Bar!int));
    static assert( __traits(isTemplate, Bar!(int).Doo));
    static assert(!__traits(isTemplate, Bar!(int).Doo!int));

    alias Tup = Tuple!(Foo, Foo!int, Bar, Bar!int, Bar!(int).Doo, Bar!(int).Doo!int);

    static assert( __traits(isTemplate, Tup[0]));
    static assert(!__traits(isTemplate, Tup[1]));
    static assert( __traits(isTemplate, Tup[2]));
    static assert(!__traits(isTemplate, Tup[3]));
    static assert( __traits(isTemplate, Tup[4]));
    static assert(!__traits(isTemplate, Tup[5]));
}

/// test overloads
void foo_over() { }
void foo_over(T : int)(T) { }
void foo_over(T : float)(T) { }
static assert(__traits(isTemplate, foo_over));

/// ditto
void bar_over() { }
void bar_over(int) { }
static assert(!__traits(isTemplate, bar_over));

/// alias to overloads
alias a_foo_over = foo_over;
static assert(__traits(isTemplate, a_foo_over));

/// ditto
alias a_bar_over = bar_over;
static assert(!__traits(isTemplate, a_bar_over));

[9rnsr]

Add this code at the start of isTemplate function would fix the issue:

if (!s->toAlias()->isOverloadable())
    return false;

[ghost]

@9rnsr: Excellent, thanks! Updated, let's wait for green.

[ghost]

It's green now. Hopefully this should do it? I'll write a spec update soon if this is a go.

[mihails-strasuns]

Please add test case with checking templated function symbol (eponymous templates are distinct enough)

[ghost]

@Dicebot: I thought foo_over covered it? What did you have in mind?

[mihails-strasuns]

Nevermind, just being blind :)

[ghost]

It's been green for a while. Ping @9rnsr

[ghost]

Ping-pong. If it's good to go auto-merge should be set. It's a top-wanted feature after all. :>

[ghost]

@WalterBright ping for review / approval (Andrei approved already). I'll write a spec pull shortly.

[ghost]

Rebased.

[quickfur]

ping
Autotester has been passing for a while now. Are we just waiting for Walter to approve this?

[andralex]

ping - what's the status of this?

[andralex]

Well I think I'll just merge this. "What's the worst that could happen?"

[ghost]

Thanks!!