Fix block parameter unpacking inside macros

[HertzDevil]

Fixes #13742 (regression on master).

This effectively reverts #13709; it seems the compiler doesn't need normalizations on all those kinds of macro nodes every time.

[straight-shoota]

FTR: The regression was introduced in #11597 (not yet released)

[Blacksmoke16]

I applied this patch on latest commit on master and tried the Athena specs again and now got:

In src/components/framework/src/ext/routing/annotation_route_loader.cr:231:128

 231 | [{ATHA::QueryParam, "ATH::Params::QueryParam".id}, {ATHA::RequestParam, "ATH::Params::RequestParam".id}].each do |(param_ann, param_class)|
                                                                                                                          ^---------
Error: undefined macro variable '__temp_113'

So there still seems to be another regression. I'll see about reducing it and will create another ticket.

[Blacksmoke16]

Actually no, reproduction for this is the same as #13742, but just call test macro twice. Because of that I don't think it needs its own ticket?

macro test
  {%
    data = [{1, 2}, {3, 4}]

    data.each do |(k, v)|
      pp k, v
    end
  %}
end

test
test

$ ccrystal test.cr --error-trace
Using compiled compiler at /Users/georgedietrich/bsmoke/crystal/.build/crystal
1
2
3
4
Showing last frame. Use --error-trace for full trace.

In test.cr:5:20

 5 | data.each do |(k, v)|
                    ^--------
Error: undefined macro variable '__temp_49'

[HertzDevil]

Apparently the normalizer mutates Block nodes in-place if they contain unpacked parameters, without emptying Block#unpacks, so calling the same macro twice will also create the multi-assign expression twice, but overwrite the previous temporary packed parameter:

do |__temp_4|
  k, v = __temp_4
  k, v = __temp_1
  k
end

[Blacksmoke16]

Awesome! Testing again with that latest commit and all Athena specs pass. Thanks!

[straight-shoota]

Specs are failing though.

[HertzDevil]

So Block#unpacks is needed by the main visitor it seems. I have replaced that with a different check