On Thu, Mar 6, 2014 at 12:09 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: > Hm ... the code in psql's describe.c is not terribly conducive to that. > Parsing of the backslash command, execution of the query/queries, and > presentation of the results is all rather tightly bound up; you'd have > to think about how to decouple those.
I just looked into describe.c, but it was virtually my first look at C in 15 years and I'm pretty intimidated.
Come to think of it, I really like the idea of moving the query execution and presentation of results out into plpgsql functions in template0; then it would be super-easy for applications to utilize them. Parsing the command could stay with psql, because the applications would probably want to do that parsing themselves, anyway.
I don't think so using plpgsql can be good idea:
a) client side is better for implementation version independent code
b) possible bugs in plpgsql code should not be fixed without living database - so any bug release can be significantly more dangerous than now.
Regards
Pavel
The trouble is that psql is written to work against older databases, so if the code were stripped from describe.c in favor of calling stored functions, those commands would stop working against older PostgreSQL installations; and if the code were left in describe.c, then the same function would exist in two different forms in two different places, with all the maintanability problems that implies.
Is there a way around this I'm not thinking of?
Alternately, I may just give up and write my own, freestanding, approximate implementation of the meta-commands with no effort to actually replace the native one. It could still be useful to application developers if they find out about it. -- - Catherine http://catherinedevlin.blogspot.com