Notes from a conversation with Stewart Brown of Lawrence Livermore

    phone 510 423 4889
    sabrown@llnl.gov

PDB is the data-structure I/O part of the larger PACT project.  PDB runs on the
following machines

intensive use at lab
    SUN OS4/Solaris
    SGI 4, 5, 6
    HP, HP/UX
    AIX
Also regularly used
    CRAY UNICOS 6,7,8
    Dec Ultrix
    OSF 1 Alpha
    Paragon
    Mac

No DOS, Windows, nor NT
an out-of-date port to VAX was done

idea the language does data structure and I/O part
    C structs
    F77 common blocks
    and you tell PDB about the data object and it handles I/O
        big emphasis on efficiency of I/O

no file-based DDL
    main mechanism is interactive object-creation calls describing your
    program's memory layout (done in a very c-struct-like language)
        see PFDEFS (Fortran) or PD_DEFSTR

PACT software support is four FTE's
    it is popular within the lab, and used under other packages
        emphasis on/committment to OUTSIDE users varies with political
            directives
    PDB is a low-level part of this, and thus heavily debugged
        it is stable and not under active development 
            (except for porting as needed to new machines/operating systems)
    Several 100 downloads outside of labs; 10 users regularly generate questions
    PACT is under active development:
        new rendering algorithms etc
        ntuple-like facility under design/development
        PDBVIEW: more work in direction of making an anlysis engine than of
            enhancing direct browsing
        all written in C
        C++ uses via C bindings; verified PDB compiles under C++ compiler
            light use so far (F90 also still light use)
    in-lab use is first priority
        if we wanted NT port, for example, likely model would be consultant
        coming out for a visit

Following versions of objects:
    would suggest using the directory-tree structure provided and then letting
    the user hand-copy into the "new" structure elsewhere in the tree
    this directory-tree is intended to control clashes in namespace
            (conceptually)
        /root/reco/jet.x
        /root/filt/jet.x

comments on other systems
    CDF, netCDF primitive types only and perhaps multidimensional arrays
    HDF
        central object registry a drawback
        outgrowth of natural support for raster images (rest tacked on in early
            versions)
        less emphasis on speed/efficiency than PDB
        complex types may need calls to many low-level routines to do I/O
            => lots of layering routines might need to be added for your own
            datatypes to be supported naturally?
                [I am not sure what he was saying, but would translate as be
                sure to look at example code for a nontrivial data structure
                before buying]

browsing in a debugger:
    might want a wrapper around PBDVIEW to have it do certain typical things
        it might otherwise want more typing than you'd like
    the gdb debugger is pretty good for supplying arguments to called routines

Doc:
    lab's web access in flux
    docs in framemaker 
        this is why the .ps is hard to view in ghostscript
    hoping to use html output option to make web version of docs in a few
        months
----------------------    
Jim,

This is pretty much what we discussed and the only thing I would add it
an additional contact, i.e.

        braddy@llnl.gov

Dennis Braddy is one of the PACT developers and is available for
questions.

Also there is a Majordomo based PACT mailing list by which we inform
our users of updates and PACT users can post questions.  To subscribe
send "subscribe pact" to

        Majordomo@rugrat.llnl.gov

Stewart