[
https://issues.apache.org/jira/browse/THRIFT-964?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12926850#action_12926850
]
Arvind Jayaprakash commented on THRIFT-964:
-------------------------------------------
The important thing in the decode will be to have a fixed memory overhead i.e.
if I want an array of 10 million floats, the code should not create an array of
10 million doubles and then trim it for me. That is something that can be
performed by wrapping the thrift deserializer. The real benefit for the
consumer comes only if the deserializer guarantees that peak memory usage in
the entire deserialization process is close to final memory needed to hold the
result.
It still does not address the "wire format" overheads but I'm willing to take
one improvement at a time.
> Support for single precision floating point numbers
> ---------------------------------------------------
>
> Key: THRIFT-964
> URL: https://issues.apache.org/jira/browse/THRIFT-964
> Project: Thrift
> Issue Type: New Feature
> Components: Compiler (General)
> Reporter: Arvind Jayaprakash
>
> We have an application that uses "float", the IEEE single precision numbers.
> The reasons are the two obvious ones: the lack of need for precision as high
> as "double" and data size given that we have millions of such numbers.
> I'd like to see support for this data format.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
