SmallDoubles replicated incorrectly to big endian clients
When a GemStone server SmallDouble is replicated from a big-endian server
to a big-endian GBS/VW client, the bytes are read incorrectly from the
buffer, resulting in a wrong SmallDouble on the client. For example, 0.995
on the server is replicated as -5.0024987722276d-5.
This bug requires both a big-endian server platform (Solaris/Sparc, AIX,
or HP-UX) and a big-endian client (HP-UX or Solaris/Sparc). Note that
GBS 7.2 and later clients were not supported on Solaris/Sparc for server
versions earlier than 2.4.1, due to an unrelated issue.
Workaround
File in the following to the VW image:
<?xml version="1.0"?> <st-source> <methods> <class-id>GemStone.Gbs.GbxB8BufferOop</class-id> <category>oop access</category> <body package="GbsServerInterfaceVW" selector="oopShortAt:">oopShortAt: shortOffset "Shouldn't really be necessary -- the senders were optimized for C memory access, which is slow, so accessing two at a time was faster." "Return the unsigned 16-bit chunk of the oop at the given shortOffset, starting from the high order end. For example, (aDelegate oopShortAt: 1) returns the most significant 16 bits of the receiver's oop." | byteOffset | byteOffset := 2 * (shortOffset - 1). ^(buffer basicAt: offset + byteOffset) * 16r100 + (buffer basicAt: offset + byteOffset + 1)</body> </methods> </st-source>
Last updated: 9/22/10