Re: [HACKERS] proposal psql \gdesc - Mailing list pgsql-hackers

From Fabien COELHO
Subject Re: [HACKERS] proposal psql \gdesc
Date
Msg-id alpine.DEB.2.20.1705072201120.3983@lancre
Whole thread Raw
In response to Re: [HACKERS] proposal psql \gdesc  (Pavel Stehule <pavel.stehule@gmail.com>)
Responses Re: [HACKERS] proposal psql \gdesc
Re: [HACKERS] proposal psql \gdesc
List pgsql-hackers
Hello Pavel,

>> Sometimes I have to solve the result types of some query. It is invisible
>> in psql.

Indeed. My 0.02€ about this patch:


About the feature:

It looks useful to allow to show the resulting types of a query.


About the code:

Patch applies cleanly and compiles.

I'm afraid that re-using the "results" variable multiple times results in 
memory leaks... ISTM that new variables should be used when going down the 
nested conditions, and all cleanup should be done where and when 
necessary.

Also, maybe it would be better if the statement is cleaned up server side 
at the end of the execution. Not sure how to achieve that, though, libpq 
seems to lack the relevant function:-(
  """although there is no libpq function for deleting a prepared  statement, the SQL DEALLOCATE statement can be used
forthat purpose."""
 

Hmmm... I have not found how to use DEALLOCATE to cleanup an unnamed 
statement, it does not allow a "zero-length" name. Maybe it could be 
extended somehow, or a function could be provided for the purpose, eg
by passing a NULL query to PQprepare...

Resetting "gdesc flag" could be done only when the flag was true, at the 
end of the if, rather than on each query.

I understand that the second level query is to retrieve the type names in 
place of the Oid returned by QPftype?

The pg_malloc length computation looks a little bit arbitrary. Would it 
make sense to use PQescapeLiteral instead?


About the documentation:

I would suggest some rewording, maybe:

"Show the description of the result of the current query buffer without 
actually executing it, by considering it a prepared statement."


About tests:

There should be some non-regression tests added, maybe something like:
  SELECT    NULL AS zero,    1 AS one,    2.0 AS two,    'three' AS three,    $1 AS four,    CURRENT_DATE AS now    --
...   \gdesc
 

And also case which trigger an error, eg:
  SELECT $1 AS unknown_type \gdesc  SELECT 1 + \gdesc

Some fun:
  PREPARE test AS SELECT 1;  EXECUTE test \gdesc  ...

I'm unsure about some error messages, but it may be unrelated to this 
patch:
 calvin=# SELECT '1 km'::unit AS foo, $2 as boo \gdesc ERROR:  could not determine data type of parameter $1

-- 
Fabien.

pgsql-hackers by date:

Previous
From: Mat Arye
Date:
Subject: Re: [HACKERS] Question about toasting code
Next
From: Daniele Varrazzo
Date:
Subject: Re: [HACKERS] Detecting schema changes during logical replication