Chad - This is an interesting observation. Do you have a sense of how much of a difference it makes?
I think you should open a ticket for this regression. Your proposed solutions sound viable and easy, so I can probably whip something up prett quickly. -Bryan On Tue, Jan 12, 2010 at 10:54 PM, Chad Walters <chad.walt...@microsoft.com>wrote: > I was playing around with upgrading some benchmarks to Thrift 0.2 and I > think that THRIFT-623 <https://issues.apache.org/jira/browse/THRIFT-623> > may have introduced a performance regression. > > Struct deserialization now does a map lookup to go from an integer to an > enumeration, checks for null in case the integer was not known, and then > does a switch on the value of the enumeration. > > But it appears to me that this mapping is entirely fixed at compile time. > Furthermore the enumeration does not appear to be used for anything other > than performing the switch (in the struct reading routine at least). > > This leads me to ask: > 1. Why not just switch on the integer and have a default case for unknown > values? > 2. Why have a dynamic map for a mapping that is known at compile time? > Wouldn't it be faster to just provide the mapping via a switch (when it is > needed at all)? > > Perhaps I am missing some subtlety but I thought I'd ask... > > Chad > > >
