Bug #12402: Inline rescue behavior inconsistent for method calls with arguments and assignment

• Status: Closed
• Assignee: nobu (Nobuyoshi Nakada)
• Target version: -
• ruby -v: 2.4.0
• Backport: 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: WONTFIX

[bklang (Ben Klang)]

In this example l'm intentionally passing bad data to Date.parse to trigger an exception. Depending on whether I use parenthesis in the method call, and whether I assign the result, I get different behavior for the inline rescue.

Code:

var1 = "apple"
var1 = Date.parse var1 rescue nil

var2 = "apple"
var2 = Date.parse(var2) rescue nil

def example1(bar)
  Date.parse bar rescue nil
end

def example2(bar)
  bar = Date.parse bar rescue nil
  bar
end

def example3(bar)
  bar = Date.parse(bar) rescue nil
  bar
end

puts "Variable 1: #{var1.nil?}"
puts "Variable 2: #{var2.nil?}"
puts "Example method 1: #{example1("apple").nil?}"
puts "Example method 2: #{example2("apple").nil?}"
puts "Example method 3: #{example3("apple").nil?}"

I would expect all 5 outputs from the above script to return true.

Example:

Variable 1: false
Variable 2: true
Example method 1: true
Example method 2: false
Example method 3: true

[noahgibbs (Noah Gibbs)]

Hm. Yup, that definitely seems to bind funny, specifically during the parsing. Here's the Ripper output for "var1 = Date.parse var1 rescue nil" versus "var1 = Date.parse(var1) rescue nil":

2.3.1 :006 > pp Ripper.sexp("var1 = Date.parse var1 rescue nil")
[:program,
 [[:rescue_mod,
   [:assign,
    [:var_field, [:@ident, "var1", [1, 0]]],
    [:command_call,
     [:var_ref, [:@const, "Date", [1, 7]]],
     :".",
     [:@ident, "parse", [1, 12]],
     [:args_add_block, [[:var_ref, [:@ident, "var1", [1, 18]]]], false]]],
   [:var_ref, [:@kw, "nil", [1, 30]]]]]]

Versus:

2.3.1 :007 > pp Ripper.sexp("var1 = Date.parse(var1) rescue nil")
[:program,
 [[:assign,
   [:var_field, [:@ident, "var1", [1, 0]]],
   [:rescue_mod,
    [:method_add_arg,
     [:call,
      [:var_ref, [:@const, "Date", [1, 7]]],
      :".",
      [:@ident, "parse", [1, 12]]],
     [:arg_paren,
      [:args_add_block, [[:var_ref, [:@ident, "var1", [1, 18]]]], false]]],
    [:var_ref, [:@kw, "nil", [1, 31]]]]]]]

That shows the rescue surrounding the whole assignment in the first case (in case of failure, no assignment happens) and around just the Date.parse() call for the second one.

Not sure why it binds at a different level for one versus the other, but that's one more level of "what's happening here?" I think that means that the answer would be in parse.y.

[matz (Yukihiro Matsumoto)]

I consider this is a bug. They should be consistent.
But I am not sure we can fix the issue soon. When in doubt, put parens.

Matz.

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)
• Status changed from Open to Assigned
• Assignee set to nobu (Nobuyoshi Nakada)

[nobu (Nobuyoshi Nakada)]

• Status changed from Assigned to Closed

Applied in changeset r55851.

---

parse.y: rescue modifier in rhs

• parse.y (command_asgn): rescue modifier in command assignment
  should be limited to rhs only. [ruby-core:75621] [Bug #12402]

[whitequark (whitequark *)]

Will this be backported?

[nagachika (Tomoyuki Chikanaga)]

• Backport changed from 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: UNKNOWN to 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: WONTFIX

Even though it's a bug, the behavior change could break existing applications. I won't backport into ruby_2_3 branch.

[shyouhei (Shyouhei Urabe)]

• Related to Bug #13005: Inline rescue is inconsistent when rescuing NoMethodError added