String Interpolation

[Licenser]

Rendered

[mfelsche]

Great RFC!

Unfortunately I think the current situation doesnt leave too much wiggle room for alternatives here:

A string like:

string::format("this should create a {{}} and say {}", 42) == "this should create a {} and say 42";

cannot really work right now. At runtime, we want to represent the string as compact as possible, a nice little continuous string would the the best option for performance reasons. In the resulting formatted string, we intend {{}} to be {} in the final result, it will be \{\} in the runtime representation, so that string::format doesnt use it as placeholder. When calling string::format we get our literal {} in the resulting formatted string. But what if we do not use string format? We only get a \{\} in the resulting string. What if we use interpolation? "{ 42 } this should work and create a {{}} and say {}" does create a runtime about mismatching count of args and placeholders, error even before we fully executed it.

Lets look at this:

string::format("this { "should work and" } create a {{}} and say {}", 42)

What this transforms into in the AST is:

string::format(string::format("this {} create a \{\} and say {}", "should work and"), 42)

This will obviously fail, as there is only additional argument to the inner string::format although the string somehow produced 2 placeholders. But to the user, this behaviour is very much not obvious.

To make this work as intended the string would need to look like:

string::format("this { "should work and" } create a {{{{}}}} and say {{}}", 42) == "this should work and create a {} and say 42"

Whereas when no interpolation is used, the above raises a runtime error as the number of placeholders and arguments does not match. It needs to be written as:

string::format("this { "should work and" } create a {{}} and say {}", 42) == "this should work and create a {} and say 42"

Having to use different escaping levels depending on whether interpolation is used or not, is not acceptable.

I think the only way to overcome this, is to use option 2. or 3. in the RFC.
While Option 4 seems nice, removing string::format means people cannot do purely dynamic formatting anymore as this cannot be replaced by string interpolation:

string::format(event.format_string, event.args)

Thus, i would vote for adding a prefix like ${} (or any other) to interpolations.

[Licenser]

Very good point. And I think you're right. I expanded on the prior art section and added a conclusion section to reflect that input.