Ivo Wever created PROTON-2241:
---------------------------------
Summary: Cannot pass integer values between 2**31 - 1 and 2**62 -
1 for pn_atom_t variables
Key: PROTON-2241
URL: https://issues.apache.org/jira/browse/PROTON-2241
Project: Qpid Proton
Issue Type: Bug
Components: ruby-binding
Affects Versions: proton-c-0.30.0
Reporter: Ivo Wever
Assignee: Alan Conway
h4. Observation
Attempting to set a number between 2**31 - 1 and 2**62 - 1 as e.g. the
message-id in Ruby 2.6.5 (and probably any other recent Ruby) results in the
number mod 2**32.
Example (pry is a Ruby repl):
{code}
[1] pry(main)> m = Qpid::Proton::Message.new('body')
=> Message{body="body"}
[2] pry(main)> m.id = 2**61 + 2**30
=> 2305843010287435776
[3] pry(main)> m.id
=> 1073741824
{code}
h4. Analysis
The SWIG bindings contain the following snippet (lines 115-123)
{code}
case T_FIXNUM:
$1.type = PN_INT;
$1.u.as_int = FIX2LONG($input);
break;
case T_BIGNUM:
$1.type = PN_LONG;
$1.u.as_long = NUM2LL($input);
break;
{code}
Integer instances <= 2**62 - 1 will hit the T_FIXNUM case, as Ruby fixnums are
63 bits (1 bit is used to indicate they are fixnums to speed up arithmetic).
Subsequently the number is converted into a 32 bit C long, discarding the 32
highest bits.
h4. Further notes
The T_BIGNUM case takes care of correctly setting the value for numbers between
2**62 and 2**63-1, as thos numbers can be converted into a 64 bit C long long
The T_BIGNUM case will fail for all larger numbers, with
{code}
RangeError: bignum too big to convert into `long long'
{code}
This is probably OK.
Note that the autoconversion of a pn_atom_t to an int or long is not useful for
e.g. the message-id property, as it needs to be passed as a ulong (when
numeric, disregarding the uuid, string, binary alternatives). Proton doesn't
enforce this and naively setting {{message.id = 1}} results in sending a
message that is invalid with respect to the specification, as the type of the
message-id will be an int.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
