Templates are way lazy. Is this expected behaviour?

April 25, 2008
Posted by Ary Borenszweig
Permalink
Ary Borenszweig
Permalink
First I'll post the new thing I learned today about D, then I'll explain why I made this experiment.

I have a module named "other":

---
module other;

class Something {
	char[] name = "other";
}

template Foo() {
	Something something;
	static this() {
		something = new Something();
	}
}
---

Then I have a module named "main":

---
module main;

import std.stdio;
import other;

mixin Foo!();

void main() {
	writefln("%s", something.name);
}
---

The output of compiling and running this is: other

Now... I modify "main" like this:

---
module main;

import std.stdio;
import other;

// Here is the change, I added this
class Something {
	char[] name = "main";
}

mixin Foo!();

void main() {
	writefln("%s", something.name);
}
---

The output of compiling and running this is: main

Not only that. If I change it to this:

---
module main;

import std.stdio;
import other;

class Something {
	// Here is the change this time
	char[] nameX = "main";
}

mixin Foo!();

void main() {
	writefln("%s", something.name);
}
---

I can't compile main anymore:
main.d(15): Error: no property 'name' for type 'main.Something'

So, basically, I can change a template's instantiation by overriding the symbols it uses. Is that expected behaviour?

I did the experiment because you currently don't get autocompletion or any cool stuff in Descent inside templates (as well as inside templated classes and functions), and won't probably in the next release. Why? Because these features rely heavily in DMD's semantic, but semantic isn't done for templates, not even checking that an identifier resolves correctly, until they are instantiated. So in the above example:

---
module other;

class Something {
	char[] name = "other";
}

template Foo() {
	Something something;
	static this() {
		something. // I'd like to suggest you the field "name",
                           // but who knows if it even exists!
	}
}
---

I know, no one will do that, but because template logic is evaluated at the location of the instantiation, I'd like to ask: could templates have at least some kind of semantic analysis done? I'd like the compiler to show an error if "Something" is not defined in the module I defined the template. What I do with template parameters doesn't care, but at least some kind of logic...
Forums
