Re: proposal: lob conversion functionality

On Mon, Sep 30, 2013 at 01:06:15PM +0300, Heikki Linnakangas wrote:
> On 12.08.2013 21:08, Pavel Stehule wrote:
>> 2013/8/10 Tom Lane<tgl(at)sss(dot)pgh(dot)pa(dot)us>:
>>> Pavel Stehule<pavel(dot)stehule(at)gmail(dot)com> writes:
>>>> I found so there are no simple API for working with LO from PL without
>>>> access to file system.
>>>
>>> What? See lo_open(), loread(), lowrite(), etc.
>>
>> yes, so there are three problems with these functions:
>>
>> a) probably (I didn't find) undocumented
>
> It's there, although it's a bit difficult to find by searching. See:
> http://www.postgresql.org/docs/devel/static/lo-funcs.html.
>
> I don't actually agree with this phrase on that page:
>
>> The ones that are actually useful to call via SQL commands are
>> lo_creat, lo_create, lo_unlink, lo_import, and lo_export
>
> Calling lo_open, loread and lowrite seems equally useful to me.
>
>> b) design with lo handler is little bit PL/pgSQL unfriendly.
>
> It's a bit awkward, I agree.

All the other large object functions are named like lo*, so I think new ones
should also be lo* rather than *_lo. One of the key benefits of large
objects, compared to a bytea column in some table, is their efficiency when
reading or writing only a subset of the object. However, the proposed
functions only deal with the large object as a whole. We could easily fix
that. Consider this list of new functions in their place:

lo_create(oid, bytea) RETURNS oid -- new LO with content (similar to make_lo)
lo_get(oid) RETURNS bytea -- read entire LO (same as load_lo)
lo_get(oid, bigint, int) RETURNS bytea -- read from offset for length
lo_put(oid, bigint, bytea) RETURNS void -- write data at offset

Anything we do here effectively provides wrappers around the existing
functions tailored toward the needs of libpq. A key outstanding question is
whether doing so provides a compelling increment in usability. On the plus
side, adding such functions resolves the weirdness of having a variety of
database object that is easy to access from libpq but awkward to access from
plain SQL. On the minus side, this could easily live as an extension module.
I have not used the large object facility to any significant degree, but I
generally feel this is helpful enough to justify core inclusion. Any other
opinions on the general suitability or on the specifics of the API offered?

Thanks,
nm

--
Noah Misch
EnterpriseDB http://www.enterprisedb.com