[tannevaled]

What

A Block had no line of its own. Program.Lines stamps statement nodes,
and a Block is not one — it is not even a Node, being a field of Call —
so a consumer had to take the line of whatever statement enclosed the block.
That is the same line only when the block opens on that statement's first line.

Measured

ruby 4.0.5 reports the opener's line as a block's
location.first_lineno:

h = {                      # line 1
  :a => proc {             # line 2
    1
  },
  :b => proc { 2 },        # line 5
  :c => proc {             # line 6
  },
}
h[:a].source_location[1]   # ruby 2    from the enclosing statement: 1
h[:b].source_location[1]   # ruby 5    from the enclosing statement: 1
h[:c].source_location[1]   # ruby 6    from the enclosing statement: 1

The body's first line is not a substitute, which is why this needs the
opener rather than a cheaper approximation: ruby answers 2 for :a, whose
first statement is on line 3, and :c has no body statement at all.

Shape

Block.Line, a field — not a position field on all 55 node types, and not
Program.Lines stretched to keys that are not Nodes (it cannot be: Block
does not implement the interface). 0 means "not recorded".

Three sites set it: the { of parseBraceBlock, the do of parseDoBlock,
and the -> of parseLambda — the arrow itself, since that is the line ruby
reports for an arrow lambda (measured; they coincide in all but deliberately
odd layouts).

The Lines doc comment now records this as its one deliberate exception, with
the measurement that forced it.

Why it matters downstream

It is observable as Proc#source_location, Method#source_location for a
define_method, and the line a Proc prints when it inspects itself. In rbgo
the block ISeq's FirstLine was taken from the enclosing scope's current line,
so proc { } inside any multi-line literal reported the literal's opening line
— 1, 8, 13 where ruby says 4, 10, 15. That cannot be fixed
downstream without this, because the information was not in the tree.

Green

go test ./... -count=1 exit 0 (4 packages), go vet clean, gofmt clean,
and the parser package stays at 100.0% of statements.

Five new cases, each want being ruby 4.0.5's own Proc#source_location: the
multi-line hash (the discriminating one — body-first-line would give 3, 6, 7),
a multi-line array with do…end, the arrow lambda, a single-line hash as the
control that nothing moved, and a method call whose do lands on a later line
than its receiver.

Also gofmt'd paren_singleton_def_test.go, which was unformatted on main.

🤖 Generated with Claude Code