_ (underscore) is not ignored inside templates

[metagn]

Example

1. _ gets gensym'd if used as a variable name:

template foo() =
  let _ = 3

foo()

2. _ is substituted if used as a parameter name for the template

template foo(_: int) =
  for _ in 1..3: # Error: identifier expected, but found '4'
    echo "hello"

foo(4)

Current Output

(2, 7) Hint: '_`gensym133006' is declared but not used [XDeclaredButNotUsed]

(5, 5) Error: identifier expected, but found '4'

Expected Output

doesn't warn or error, correctly does not gensym in the first case, and does not try to substitute in the second case

Workaround

template foo() =
  let _ {.inject.} = 3

foo()

$ nim -v
Nim Compiler Version 1.2.6

[Araq]

What's the use case though? Your example should use discard.

[SolitudeSF]

template foo =
  let (_, a, b) = (3, 4, 5)

warns when unpacking a tuple

[demotomohiro]

When 2 "_" are used in tuple unpacking in a template, it generate compile error.

#No compile error
let (_, _, x) = (1, 2, 3)
echo x

template test() =
  #Error: redefinition of '_`gensym3387051'; previous declaration here: f:\temp\testTU.nim(5, 8)
  let (_, _, x) = (1, 2, 3)
  echo x

test()

[metagn]

Just ran into this bug again, but in a different way. Not only do templates not ignore _, they also try to substitute with it if it's the name of a parameter.

template foo(_: int) =
  for _ in 1..3: # Error: identifier expected, but found '4'
    echo "hello"

foo(4)

[ringabout]

It can be used inside procs too.

proc main(_: int) = echo _

#13443 (comment)

[metagn]

Fixing #13443 should fix second part (although should still be tested) so don't need to specifically fix it yet.

Edit: This might not be true, semtempl might still need to be patched