[pulseaudio-discuss] RFC: pa_database to adopt hdf5?

[Jason Newton]

On Mon, Jun 24, 2013 at 11:59 PM, Ian Malone <ibmalone at gmail.com> wrote:

> Having followed this thread for a while I'm wondering why HDF5, which
> seems geared towards large numerical datasets and rather specialist,
> would be preferable to sqlite if a dedicated database library is
> needed for application settings.
>
>
> --
> imalone

Having used sqlite in the past, I would say that sqlite would be a
more clunkier fit for these purposes, although it can and has been
used for those purposes in other contexts.  I would say HDF5 was
designed to take care of numerical datasets but it isn't necessarily
so specialized as to do that alone, much like ext4 can handle an
exabyte but is still quite happy being a root fs on common linux
distros.  A key feature of HDF5 is that it allows storing compound
types aka POD structs and arrays and the library takes care of getting
to/from disk via the description in the compound type the developer
builds and which is present in an existing file.  You may consider
this an ORM layer that is built into an FS.  Unlike sqlite however,
not everything has to be a table and any object can have any number of
attributes to store additional metadata.  One thing both would share
however is additive changes to the structures will not break anything
in terms of backwards compatibility. Both approaches also allow one to
only reference what is needed which helps minimize breakage when
non-additive changes are made to structures.

-Jsaon