[ruby-core:86358] [Ruby trunk Bug#5304][Rejected] Array#pack handles objects for eg format 'E' differently than 1.8

Issue #5304 has been updated by mrkn (Kenta Murata).

Status changed from Assigned to Rejected

Rejected because this had already been meaningless now.

----------------------------------------
Bug #5304: Array#pack handles objects for eg format 'E' differently than 1.8
https://bugs.ruby-lang.org/issues/5304#change-71286

* Author: brixen (Brian Shirai)
* Status: Rejected
* Priority: Normal
* Assignee: mrkn (Kenta Murata)
* Target version: 
* ruby -v: ruby 1.9.2p290 (2011-07-09 revision 32553) [x86_64-darwin10.8.0]
* Backport: 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: UNKNOWN
----------------------------------------
In 1.9, Array#pack when passed an object that defines #to_f and one of the Float formats raises a TypeError. This is a change from 1.8. The code in 1.9 calls rb_to_float()

# 1.9

VALUE
rb_to_float(VALUE val)
{
    if (TYPE(val) == T_FLOAT) return val;
    if (!rb_obj_is_kind_of(val, rb_cNumeric)) {
	rb_raise(rb_eTypeError, "can't convert %s into Float",
		 NIL_P(val) ? "nil" :
		 val == Qtrue ? "true" :
		 val == Qfalse ? "false" :
		 rb_obj_classname(val));
    }
    return rb_convert_type(val, T_FLOAT, "Float", "to_f");
}

The rb_to_float() imposes the (arbitrary) restriction that the object be kind of Numeric. The 1.8 pack code used rb_Float(), which merely requires the object respond to #to_f and return a Float.

The treatment of objects in 1.9 for the float formats is thus different than for eg the integer formats, where the object simply must respond to #to_int. The requirement that the object be a kind of Numeric breaks the uniform ducktyping and imposes an arbitrary and unnecessary type/class requirement.

Was this change intended, and if so, why? Apologies if this was discussed on ruby-core prior to the change; I could not find any discussion through searching.

The following code illustrates the issue:

o = Object.new

def o.to_int
  puts "to_int called"
  1
end

def o.to_f
  puts "to_f called"
  1.1
end

p [o].pack("I") # => "\001..." (depends on endianness)
p [o].pack("E") # => 1.8 - "<numbers>"; 1.9 - TypeError

$ ruby1.8.7 -v pack_test.rb 
ruby 1.8.7 (2011-02-18 patchlevel 334) [i686-darwin10.8.0]
to_int called
"\001\000\000\000"
to_f called
"\232\231\231\231\231\231\361?"

$ ruby1.9.2 -v pack_test.rb 
ruby 1.9.2p290 (2011-07-09 revision 32553) [x86_64-darwin10.8.0]
to_int called
"\x01\x00\x00\x00"
pack_test.rb:14:in `pack': can't convert Object into Float (TypeError)
	from pack_test.rb:14:in `<main>'




-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>