[prism] (no-op) multiple newline escapes collapsed in heredocs

[froydnj]

Like #771, I don't think this matters, exactly, because the string content is the same, but recording for posterity.

description = <<~EOS
  This is a very long line that we are trying to extend for as long as possible to mirror the original example which I can't completely talk about.\
  \
  For other reasons, this would point at a useful URL

  Extra rationale.
EOS

p(description)

rubyfmt --prism produces:

description = <<~EOS
  This is a very long line that we are trying to extend for as long as possible to mirror the original example which I can't completely talk about.\
  For other reasons, this would point at a useful URL

  Extra rationale.
EOS

p(description)

Running the formatted code through Ruby indicates that they both produce the same thing, though, so 🤷‍♂️ .

[reese]

TIL you can \ in heredocs like a string concat to continue lines 🙃

[reese]

Hmm, there must be some shenaniganry going on here. Either I'm bad at debugging or Prism doesn't even give us the second line of the heredoc in the content_loc() at all.

Starting with this

% cat hi.rb
p <<~EOS
  This is a very long line that we are trying to extend for as long as possible to mirror the original example which I can't completely talk about.\
  \
  For other reasons, this would point at a useful URL

  Extra rationale.
EOS

% ruby hi.rb 
"This is a very long line that we are trying to extend for as long as possible to mirror the original example which I can't completely talk about.For other reasons, this would point at a useful URL\n\nExtra rationale.\n"

I add some debug statements to just print the ProgramNode we get inside rubyfmt, and I also print out all the string parts elsewhere, and the second is skipped over.

% cargo run --release -- --prism ./hi.rb
# [...]

[librubyfmt/src/format_prism.rs:634:5] &program_node = ProgramNode([], StatementsNode([CallNode(None, None, 1, Some("p"), None, Some(ArgumentsNode([InterpolatedStringNode(Some("<<~EOS"), [
StringNode(None, "  This is a very long line that we are trying to extend for as long as possible to mirror the original example which I can\'t completely talk about.\\\n", None, [...]),
StringNode(None, "  For other reasons, this would point at a useful URL\n", None, [...]),
StringNode(None, "\n", None, [10]), 
StringNode(None, "  Extra rationale.\n", None, [...])], Some("EOS\n"))])), None, None, None)]))

[librubyfmt/src/format_prism.rs:873:9] loc_to_str(part.as_string_node().unwrap().content_loc()) = "  This is a very long line that we are trying to extend for as long as possible to mirror the original example which I can't completely talk about.\\\n"
[librubyfmt/src/format_prism.rs:873:9] loc_to_str(part.as_string_node().unwrap().content_loc()) = "  For other reasons, this would point at a useful URL\n"
[librubyfmt/src/format_prism.rs:873:9] loc_to_str(part.as_string_node().unwrap().content_loc()) = "\n"
[librubyfmt/src/format_prism.rs:873:9] loc_to_str(part.as_string_node().unwrap().content_loc()) = "  Extra rationale.\n"
p(
  <<~EOS
    This is a very long line that we are trying to extend for as long as possible to mirror the original example which I can't completely talk about.\
    For other reasons, this would point at a useful URL

    Extra rationale.
  EOS
)

but in irb it's clearly there in that content_loc:

              │       └── @ InterpolatedStringNode (location: (1,2)-(1,8))
              │           ├── opening_loc: (1,2)-(1,8) = "<<~EOS"
              │           ├── parts: (length: 3)
              │           │   ├── @ StringNode (location: (2,0)-(5,0))
              │           │   │   ├── flags: ∅
              │           │   │   ├── opening_loc: ∅
              │           │   │   ├── content_loc: (2,0)-(5,0) = "  This is a very long line that we are trying to extend for as long as possib
le to mirror the original example which I can't completely talk about.\\\n  \\\n  For other reasons, this would point at a useful URL\n"
              │           │   │   ├── closing_loc: ∅
              │           │   │   └── unescaped: "This is a very long line that we are trying to extend for as long as possible to mirror the o
riginal example which I can't completely talk about.    For other reasons, this would point at a useful URL\n"
              │           │   ├── @ StringNode (location: (5,0)-(6,0))
              │           │   │   ├── flags: ∅
              │           │   │   ├── opening_loc: ∅
              │           │   │   ├── content_loc: (5,0)-(6,0) = "\n"
              │           │   │   ├── closing_loc: ∅
              │           │   │   └── unescaped: "\n"
              │           │   └── @ StringNode (location: (6,0)-(7,0))
              │           │       ├── flags: ∅
              │           │       ├── opening_loc: ∅
              │           │       ├── content_loc: (6,0)-(7,0) = "  Extra rationale.\n"
              │           │       ├── closing_loc: ∅

It seems strange to me that we would get different results from the Rust vs Ruby bindings, though I haven't bothered to dig in enough to say anything concrete about that. The main thing that's surprising is that from the Rust bindings we're getting 4 string parts, and from Ruby (version 3.3.1 on my laptop, so maybe that's related somehow??) we're only getting 3.

[reese]

Ah, I got with the times and installed 4.0, that does show 4 parts now in irb:

              │       └── @ InterpolatedStringNode (location: (1,2)-(1,8))
              │           ├── flags: ∅
              │           ├── opening_loc: (1,2)-(1,8) = "<<~EOS"
              │           ├── parts: (length: 4)
              │           │   ├── @ StringNode (location: (2,0)-(3,0))
              │           │   │   ├── flags: static_literal, frozen
              │           │   │   ├── opening_loc: ∅
              │           │   │   ├── content_loc: (2,0)-(3,0) = "  This is a very long line that we are trying to extend for as long as possible to mirr
or the original example which I can't completely talk about.\\\n"
              │           │   │   ├── closing_loc: ∅
              │           │   │   └── unescaped: "This is a very long line that we are trying to extend for as long as possible to mirror the original ex
ample which I can't completely talk about."
              │           │   ├── @ StringNode (location: (4,0)-(5,0))
              │           │   │   ├── flags: static_literal, frozen
              │           │   │   ├── opening_loc: ∅
              │           │   │   ├── content_loc: (4,0)-(5,0) = "  For other reasons, this would point at a useful URL\n"
              │           │   │   ├── closing_loc: ∅
              │           │   │   └── unescaped: "For other reasons, this would point at a useful URL\n"
              │           │   ├── @ StringNode (location: (5,0)-(6,0))
              │           │   │   ├── flags: static_literal, frozen
              │           │   │   ├── opening_loc: ∅
              │           │   │   ├── content_loc: (5,0)-(6,0) = "\n"
              │           │   │   ├── closing_loc: ∅
              │           │   │   └── unescaped: "\n"
              │           │   └── @ StringNode (location: (6,0)-(7,0))
              │           │       ├── flags: static_literal, frozen
              │           │       ├── opening_loc: ∅
              │           │       ├── content_loc: (6,0)-(7,0) = "  Extra rationale.\n"
              │           │       ├── closing_loc: ∅
              │           │       └── unescaped: "Extra rationale.\n"
              │           └── closing_loc: (7,0)-(8,0) = "EOS\n"
              ├── closing_loc: ∅
              ├── equal_loc: ∅
              └── block: ∅,
 @warnings=[]>

so this implies that Prism essentially does just make that line disappear. Not sure if that's intended or not, but it sounds like this is likely more of a Prism thing than any specific issue in our implementation (especially since we do the right thing for the first backslash).