TimeStamp vs java.time.Instant

[shroffk]

based on the interface description of http://epics-pvdata.sourceforge.net/docbuild/4.3.0/pvDataJava/documentation/html/org/epics/pvdata/property/TimeStamp.html

It seems like almost all the functionality and behaviour of org.epics.pvdata.property.TimeStamp is supported by https://docs.oracle.com/javase/8/docs/api/java/time/Instant.html

The only missing piece is userTag. Should we simplify this interface by using the java.time.Instant to represent time and only keeping the userTag field.

[kasemir]

I'm afraid we need to distinguish between a time stamp that you use in some programming language to hold time information, and the EPICS normative type from
http://epics-pvdata.sourceforge.net/alpha/normativeTypes/normativeTypes.html :

time_t := structure
    long secondsPastEpoch
    int  nanoseconds
    int  userTag

When you read pvdata in Java, I agree that the seconds and nanosecond information from within the time_t should somehow end up in a java Instant.
On the other hand, the pvdata needs to be somehow conserved.
I need to be able to access elements of a structure as pva://some_pv/struct/substruct/element and that includes pva://some_pv/time/secondsPastEpoch.
Not sure how that would work if the time_t received as a pvdata structure is early on converted into an Instant, loosing the original structure.

The existing Java pvdata library keeps the time_t as a structure, and the VType then parses elements of the time_t into an Instant, elements of the display_t into a Display etc.

[ralphlange]

I'm with Kay on this.
While for any specific application I couldn't care less which favorite kind of object they're turning everything into (I would opt for Polenta), enforcing it on all clients written in a specific language is clearly wrong.
E.g. how would the archive appliance (Java) write timestamps into the archive? Instantly? How would a retrieval tool (C++) get the data back?

[mrkraimer]

I agree with Kay and Ralph.
The format

time_t := structure
    long secondsPastEpoch
    int  nanoseconds
    int  userTag

was decided on several years ago.
Except for userTag it has existed since 2006.
userTag was added after conversion of pvData to CPP started (about 2010?).
It is used in both CPP and Java and has associated normative type and pvData property definitions.

Note that the C++ definition of timeStamp.h  has additional methods:
    /**
     * Set timeStamp from standard C time
     * @param time time as returned by std::time
     */
    void fromTime_t(const time_t & time);
    /**
     * Set time from timeStamp
     * @param time time as defined by std::time
     */
    void toTime_t(time_t &time) const;

Perhaps you might consider adding similar methods to the Java implementation of TimeSramp.Java

[shroffk]

Thanks for the explanation.