Implement zero copy I/O reads into a FFI AbstractMemory destination buffer
---------------------------------------------------------------------------
Key: JRUBY-3636
URL: http://jira.codehaus.org/browse/JRUBY-3636
Project: JRuby
Issue Type: Improvement
Reporter: Wayne Meissner
Assignee: Thomas E Enebo
Fix For: JRuby 1.x+
Currently, to read data from an IO stream into a FFI memory buffer, the data
must be first read into a ruby string instance, then copied into the FFI buffer.
The number of copies is actually more than that, since NIO will first read into
a direct buffer (native memory), then copied to the ChannelStream buffer (if
the read size is less than 4K), then copied again to the String byte array.
In the best case, three copies could be eliminated from the total sequence, if
the FFI buffer is also native memory. Worst case, only two copies are
eliminated.
For now, a simple API that mimics the C read(2) interface can be supplied in
the FFI namespace to abstract the read process, which we can replace later with
one that eliminates the extra copying.
e.g.
{code:title=io.rb}
def self.read(io, buf, len)
tmp = io.read(len)
return -1 unless tmp
buf.put_bytes(0, tmp)
tmp.length
end
{code}
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://jira.codehaus.org/secure/Administrators.jspa
-
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
