Range#step with Float arguments differs from MRI
------------------------------------------------
Key: JRUBY-6133
URL: https://jira.codehaus.org/browse/JRUBY-6133
Project: JRuby
Issue Type: Bug
Environment: --1.9
Reporter: Martin Harriman
Assignee: Thomas E Enebo
Range#step in MRI uses multiplication to step a floating-point value.
Range#step in jruby uses repeated addition, which has significantly poorer
error behavior and results in different values and even different numbers of
iterations when compared with MRI.
For example:
{noformat}
1.0...55.6.step(18.2).to_a
{noformat}
In MRI, the result is a four-element array, with the final element just an
epsilon less than 55.6. In jruby 1.7.0, the result is a three element array,
terminating at 37.4.
As a rule of thumb, strength-reducing floating-point multiplication into
repeated floating-point addition is a bad idea, as you will get much larger
quantization error with the repeated addition (and it is not usually faster
than floating-point multiplication).
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe from this list, please visit:
http://xircles.codehaus.org/manage_email
