Expression on same line of `if` condition is part of the true branch

[giordano]

julia> if true println("hello")
           1
       end
hello
1

julia> Meta.@lower if true println("hello")
           1
       end
:($(Expr(:thunk, CodeInfo(
    @ none within `top-level scope`
1 ─     goto #3 if not true
    @ REPL[13]:1 within `top-level scope`
2 ─     println("hello")
│   @ REPL[13]:2 within `top-level scope`
└──     return 1
3 ─     return nothing
))))

The println("hello") expression becomes part of the body of the true branch. This "feature" is even more nasty when you do by mistake something like

if condition1 && condition2 condition3 && condition4

condition3 && condition4 aren't evaluated at all as part of the conditions of the if. This should arguably be a parser/syntax error. For example

julia> if true 1 2 3
           1
       end
ERROR: syntax: unexpected "2"
Stacktrace:
 [1] top-level scope
   @ none:1

2 is a syntax error, I don't see why 1 shouldn't.

Other weird interpretations of this "feature":

julia> if true :foo
           1
       end
ERROR: UndefVarError: `foo` not defined
Stacktrace:
 [1] top-level scope
   @ REPL[15]:1

julia> if true :2
           1
       end
ERROR: TypeError: non-boolean (UnitRange{Int64}) used in boolean context
Stacktrace:
 [1] top-level scope
   @ REPL[17]:1

[barucden]

Would it mean to also disallow one-liners in the form

if cond f() end

?

These one-liners are quite common, e.g.,:

julia/stdlib/LinearAlgebra/test/generic.jl

Line 81 in f2a5fb9

chksign(s1, s2) = if elty <: Real s1 == s2 else s1 ≈ s2 end

julia/stdlib/LinearAlgebra/src/dense.jl

Line 1346 in f2a5fb9

if m == 1 return A[1] end

julia/base/strings/search.jl

Line 189 in f2a5fb9

if i === nothing return 0 end

[adienes]

this is for sure a footgun I have both shot myself with and complained about, but strictly speaking I'm not sure it's a bug. maybe a better job for a linter

[gbaraldi]

I guess we should've disallowed this a while ago, and told users to use the ternary, which is explicit

[Keno]

I'd be in favor of requiring a semicolon, but it's probably technically breaking

[JeffBezanson]

This was intentional (many years ago) but I agree it's highly debatable. Would be breaking to change now.

[jariji]

I use

if p a end

but I'd rather use

if p then a

from #16389.

[KristofferC]

p && a is short and you learn to just pattern match what it means pretty quickly.

[jariji]

Yeah, it's just more symbols with weird special meanings. And if p then a also supports if p then a else b (common in functional languages) rather than needing to overload ? and : which isn't necessary either.