postfix completion on a closure could probably work better.

consider the following code:

let x = || {};

This defines x as a closure that returns nothing. However, when doing:

let x = || {}.box;

and selecting the box postfix completion, currently, rust-analyzer does the following:

let x = || Box::new({});

However, what we probably want to do in this case is actually:

let x = Box::new(|| {});

Wonder if we should special case this. We could probably see if the direct parent to the BLOCK_EXPR is a CLOSURE_EXPR and pass that as the receiver instead?

I tried to take a stab at this, but ran into the issue where the AST when the . exists ends up being:

            CLOSURE_EXPR@12..18
              PARAM_LIST@12..14
                PIPE@12..13 "|"
                PIPE@13..14 "|"
              WHITESPACE@14..15 " "
              FIELD_EXPR@15..18
                BLOCK_EXPR@15..17
                  STMT_LIST@15..17
                    L_CURLY@15..16 "{"
                    R_CURLY@16..17 "}"
                DOT@17..18 "."

We would need to to fix this up a bit by creating some fake syntax so that it looks like this for the receiver:

            CLOSURE_EXPR@12..18
              PARAM_LIST@12..14
                PIPE@12..13 "|"
                PIPE@13..14 "|"
              WHITESPACE@14..15 " "
              BLOCK_EXPR@15..17
                STMT_LIST@15..17
                  L_CURLY@15..16 "{"
                  R_CURLY@16..17 "}"

Not quite sure how to do that yet, though 🤔 - any pointers for where to look?

Start in rust-analyzer's postfix completion handling and inspect how the CLOSURE_EXPR, BLOCK_EXPR, and FIELD_EXPR AST shape is used when the dot is present. Determine how the receiver syntax can be adjusted so the closure, rather than its block, receives the box completion; done means the example produces Box::new(|| {}) and the behavior is covered by a regression test.