Re: Similar to csvlog but not really, json logs?

On Tue, Aug 26, 2014 at 7:47 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> I think that it would be a good beginner's project to make pprint()
>> print JSON.
>
> There's something to be said for that (or, really, for any standardized
> widely-popular textual data format; but JSON is a perfectly reasonable
> candidate).

Yeah. The structures that the optimizer produces in particular are
very intricate. It would be quite nice to have a way to manipulate it
mechanically.

>> The existing infrastructure is user visible because of GUCs like
>> debug_print_parse.
>
> There is that :-(.  The fact that these strings are stored in the catalogs
> isn't a problem as long as we make the change in a major version upgrade.
> But to the extent that there is client-side code that expects to make
> sense of the strings, it could be a problem.  Is there any such code?
> If so, are we really beholden to not break it?  It's not like we don't
> change those data representations routinely anyway ...

I highly doubt that there is any such code in the wild. It seems to
only be intended for debugging user-defined rules, which are not a
popular feature. Personally, I've occasionally used tricks like
diffing two files with similar Nodes to see where and how differences
that are of interest arise.  :-)

-- 
Peter Geoghegan