On 30/10/2019 16:15, Mitar wrote:
Hi!
On Wed, Oct 30, 2019 at 8:37 AM Merlin Moncure <mmoncure@xxxxxxxxx> wrote:
Check out libpqtypes: https://github.com/pgagarinov/libpqtypes
Interesting. I have looked at the code a bit but I do not find how it
determines the type for inline compound types, like the ones they
appear in my original SQL query example. Could you maybe point me to
the piece of code there handling that? Because to my
understanding/exploration that information is simply not exposed to
the client in any way. :-(
it does exactly what you want. It's a wrapper for libpq that provides
client side parsing for the binary protocol with array and composite
type parsing.
It looks to me that it does parsing of composite types only if they
are registered composite types. But not for example ones you get if
you project a subset of fields from a table in a subquery. That has no
registered composite type?
Also, how you are handling discovery of registered types, do you read
that on-demand from the database? They are not provided over the wire?
Virtually any
non-C client application really ought to be using json rather than the
custom binary structures libpqtyps would provide.
I thought that initially, too, but then found out that JSON has some
heavy limitations because the implementation in PostgreSQL is standard
based. There is also no hook to do custom encoding of non-JSON values.
So binary blobs are converted in an ugly way (base64 would be better).
You also loose a lot of meta-information, because everything non-JSON
gets converted to strings automatically. Like knowing what is a date.
I think MongoDB with BSON made much more sense here. It looks like
perfect balance between simplicity of JSON structure and adding few
more useful data types.
But yes, JSON is great also because clients often have optimized JSON
readers. Which can beat any other binary serialization format. In
node.js, it is simply the fastest there is to transfer data:
https://mitar.tnode.com/post/in-nodejs-always-query-in-json-from-postgresql/
Mitar
Then perhaps, as opposed to wedging this into the tabular paradigm, a
transition to more targeted support for hierarchical result
representation would be preferable, just done directly by the backend an
rendered by libpq... (perhaps still encapsulated as a DataRow field not
to break the traditional model. Or perhaps a special RowDescription-like
message in the backend protocol? Not my place to strongly push proposals
there).
There's a lot of room for optimisation if done natively (think label
deduplication at the source. Not sure if BSON works this way too).
There's also the problem of independent implementations of the
protocol...AFAIK the JDBC client is not a wrapper to libpq and they'd
also have to break their result surfacing paradigms to make it work...
Sounds like an enormous risk & undertaking for the hackers TBH, and I
currently see another limiting factor to idea's popularity: as it
stands, advanced SQL is daunting for the much of the industry, and IMHO
the queries to generate arbitrarily structured & lightweight inline
types/relations are relatively verbose and deeply nested (eg: last time
I checked, stripping/renaming some attributes from a relation required
subselecting them, and - prehaps due to PEBCAK - can't think of a way to
create results as associative arrays indexed by attributes).
For this to gain traction, a more streamlined syntax/functions/operators
for precision work may be necessary, or the result would only satisfy a
narrow set of users who are already intimate with the state of affairs.
Can't help thinking that the current JSON-over-field pinhole may already
be at the sweet spot between usefulness and inter-operability with
existing systems. Just the SQL side of it could be less noisy and, yes,
data type pidgeonhole problem could benefit from something like a GUC
setting to electively break standard JSON compatibility, function
arguments or else.
--
Regards
Fabio Ugo Venchiarutti
OSPCFC Network Engineering Dpt.
Ocado Technology
--
Notice:
This email is confidential and may contain copyright material of
members of the Ocado Group. Opinions and views expressed in this message
may not necessarily reflect the opinions and views of the members of the
Ocado Group.
If you are not the intended recipient, please notify us
immediately and delete all copies of this message. Please note that it is
your responsibility to scan this message for viruses.
References to the
"Ocado Group" are to Ocado Group plc (registered in England and Wales with
number 7098618) and its subsidiary undertakings (as that expression is
defined in the Companies Act 2006) from time to time. The registered office
of Ocado Group plc is Buildings One & Two, Trident Place, Mosquito Way,
Hatfield, Hertfordshire, AL10 9UL.
