[ https://issues.apache.org/jira/browse/MESOS-3345?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14724630#comment-14724630 ]
Benjamin Mahler commented on MESOS-3345: ---------------------------------------- Yes, we use proto2, I'm suggesting we mimic the proto3 <-> JSON conversion (much like we did with base64 for bytes fields). Yes, picojson will parse it as a string, that's the point :) Within our JSON->Protobuf conversion, we have to numify accordingly. > Expand the range of integer precision when converting into/out of json. > ----------------------------------------------------------------------- > > Key: MESOS-3345 > URL: https://issues.apache.org/jira/browse/MESOS-3345 > Project: Mesos > Issue Type: Task > Components: stout > Reporter: Joseph Wu > Assignee: Joseph Wu > Priority: Minor > Labels: json, mesosphere, protobuf > > For [MESOS-3299], we added some protobufs to represent time with integer > precision. However, this precision is not maintained through protobuf <-> > JSON conversion, because of how our JSON encoders/decoders convert numbers to > floating point. > To maintain precision, we can try one of the following: > * Try using a {{long double}} to represent a number. > * Add logic to stringify/parse numbers without loss when possible. > * Try representing {{int64_t}} as a string and parse it as such? > * Update PicoJson and add a compiler flag, i.e. {{-DPICOJSON_USE_INT64}} > In all cases, we'll need to make sure that: > * Integers are properly stringified without loss. > * The JSON decoder parses the integer without loss. > * We have some unit tests for big (close to {{INT32_MAX}}/{{INT64_MAX}}) and > small integers. -- This message was sent by Atlassian JIRA (v6.3.4#6332)