Thread: Hstore array implementation.

On Thu, Oct 6, 2011 at 8:00 AM, Ronan Dunklau <rdunklau@gmail.com> wrote:
> Hello.
> I have implemented an Hstore array type converter for psycopg2.
> I thought it may be of interest to you, even if I'm afraid I may have
> forgotten things in my implementation.
> You can look at the patch here:
> https://github.com/Kozea/psycopg/commit/60fd7ecf833f1e56a41cca7b5700dea2add75091


Hi Ronan,

thank you for the effort. But I've already added generic array support
for psycopg. Or, better, the support was there, but was available only
in C and was not exposed to Python.

I've added the function new_array_type(), which is similar to
new_type: it takes a type caster for a single item and a list of oids
and creates a new type caster to parse array of the same types: see
<http://initd.org/psycopg/docs/extensions.html#database-types-casting-functions>.
With this extension, now hstore and composite types casters support
arrays too. I think fog hasn't reviewed these patched yet so things
are not set in stone, but exposing the internal array handling system
to python seems a good idea. They should be released in the upcoming
2.4.3.

I'm sorry for having had your time wasted: we had a ticket open on the
feature (http://psycopg.lighthouseapp.com/projects/62710/tickets/66)
and I'd closed it two weeks ago. I was probably not enough as
communication (but then, I don't know if you approached the problem
because you found the ticket or you needed it for yourself).

Thank you anyway!

-- Daniele

On 06/10/11 09:00, Ronan Dunklau wrote:
> Hello.
>
> I have implemented an Hstore array type converter for psycopg2.
>
> I thought it may be of interest to you, even if I'm afraid I may have
> forgotten things in my implementation.

Hi Ronan,

thank you very much for the patch. You'll probably want to refactor it a
little bit after next release that will add generic array support
(written by Daniele).

federico

--
Federico Di Gregorio                                       fog@initd.org
                  Beh un bacio, se ben dato, non si rifiuta. --  <laura>

On Thu, Oct 6, 2011 at 8:51 AM, Federico Di Gregorio <fog@dndg.it> wrote:

[ sorry fog: your message was in moderation: you were not using the
usual email address. I've added it to the list - nopost]

-- Daniele

On Thu, Oct 6, 2011 at 12:06 PM, Ronan Dunklau <rdunklau@gmail.com> wrote:
> Sorry, I didn't see the bug report: last time I tried to access the bug
> tracker it was down.
> Don't be sorry, it did not take much time to implement, and I needed it
> right now.
> Generic array support is great !

Yes, it's neat, and it's the way all builtin array casters are
implemented. it's been buried in the C code since ever, it just had to
surface :)

> I've been waiting for it since a while (hstore arrays, record arrays...)
> Is your devel branch available publicly ?

It is in the devel branch of my clone, at
<https://github.com/dvarrazzo/psycopg/>, see commits of Sep 22.

> I tried to look at the devel branch from
>  git://luna.dndg.it/public/psycopg2.git, and did not find anything regarding
> that.
> Regarding new_type, does it automatically register an array type for every
> type, or does the developer have to manually register the type and array
> type ?

Registration has to be separate: for instance register_hstore() hast
something like that in its implementation:

    HSTORE = _ext.new_type(oid, "HSTORE", cast)
    _ext.register_type(HSTORE, [where])
    HSTOREARRAY = _ext.new_array_type(array_oid, "HSTOREARRAY", HSTORE)
    _ext.register_type(HSTOREARRAY, [where])

The hstore/composite casters think themselves to register the array
casters too; authors of new typecasters should do themselves in a
similar way.

> Thanks for the good work !

Cheers :)