Prevent duplication of include_bytes! statements

[GuillaumeGomez]

Fixes #72.

[Kijewski]

Thinking about it, I don't think using a cache this why is sane:

• Given the file layout.html is used in compiling units a.rs and b.rs, then the include might wind up only in one of the generated objects, say, a.rs.
• Afterwards, we remove a.rs and any mentions to it.
• Then we alter layout.html: b.rs does not mention the dependency, so the change won't cause a recompilation.

[GuillaumeGomez]

If we remove a.rs, the crate will be recompiled and layout.html will be used in b.rs. Or did I miss something?

[Kijewski]

You cannot assume that the whole crate is recompiled when incremental compilation is enabled, can you?

[GuillaumeGomez]

Hum... It's a good question. But since you remove a.rs, it also means you remove mod a somewhere else, so in this case, I wonder if the whole crate isn't simply rebuilt...

[Kijewski]

But what if the add more indirection? (Assume the usual folder structure):

mod x1 {
    mod x2 {
        mod a;
    }
    mod x3 {
        mob b;
    }
}

Then removing mod a will cause a recompilation of mod x2, but not x1. If the need to rebuild would bubble up, then there could be no incremental compilation.

[GuillaumeGomez]

The fact that we're debating over this is already a pretty good sign that we should not do it. Closing then! :)

[Kijewski]

Maybe once upon a time rust-lang/rust#99515 gets implemented. :) If I understand the proposal correctly, then this could be a game changer, and we could even use spans inside the template file to show error messages, etc.