Prism::Source.for(code) returns an incomplete Prism::Source object

[eregon]

irb -Ilib -rprism
irb(main):001> src = Prism::Source.for("1 + 2")
=> #<Prism::ASCIISource:0x00007f59cab9f840 @offsets=[], @source="1 + 2", @start_line=1>
irb(main):002> src.line_start(3)
=> nil

The problem is @offsets is [] which is never valid.
(@start_line defaults to 1 which is probably fine)

Another way to see it:

irb(main):004> Prism.parse("1\n2").source
=> #<Prism::ASCIISource:0x00007f59c9affc60 @offsets=[0, 2], @source="1\n2", @start_line=1>
irb(main):005> Prism::Source.for("1\n2")
=> #<Prism::ASCIISource:0x00007f59c998cf68 @offsets=[], @source="1\n2", @start_line=1>

Prism::Source.for is notably used by Prism::DSL#source and might be used by code outside of Prism (I don't know).
It's also used in the Ripper translation layer, I'm fixing that usage in #3859.

The method takes arguments for the offsets and start_line:

prism/lib/prism/parse_result.rb

Line 13 in d3d3de9

def self.for(source, start_line = 1, offsets = [])

But because they are optional and Prism doesn't provide a way to compute the offsets (AFAIK) I don't think that's good enough.

Some solutions I can imagine:

• Remove Prism::Source.for, only allow to create complete Source objects through Prism.{lex,parse}*. This is what the Java code does, the Source constructor is private, and the only way to get a Source is through lexing/parsing and get it from the ParseResult.
• Have the default compute the offsets itself, probably most convenient but a bit of a performance trap (at least we should document it on the method). In the past we have seen that computing line offsets from Ruby is pretty slow.
• Have the default be nil and automatically compute the offsets when needed.