I have a Postgres script that I have that I intend to run against multiple databases. The sample SQL has functions and tables like
CREATE FUNCTION point() RETURNS trigger
LANGUAGE plpgsql
AS $$
BEGIN
NEW.the_geom:=ST_SetSRID(geom, :CRS) ;
RETURN NEW;
END
$$;
CREATE TABLE admin (
gid integer NOT NULL,
geom geometry(Polygon,:CRS)
);
I have put a variable inside so that I can substitute it on runtime as
psql -d db -p 5432 -U username -h localhost -f test.sql --variable=CRS=3857
Why does the variable only get properly substituted in table definitions and not function definition
https://www.postgresql.org/docs/current/static/app-psql.html#APP-PSQL-INTERPOLATION
Therefore, a construction such as ':foo' doesn't work to produce a
quoted literal from a variable's value (and it would be unsafe if it
did work, since it wouldn't correctly handle quotes embedded in the
value).
a function definition is put between quotes - in your case $$ instead of ', so it fails. You can change ST_SetSRID(geom, :CRS) to ST_SetSRID(geom, $$:CRS) to see it start being handled again (of course ruining the function body)
As a workaround you can try using bash variable instead:
vao#vao-VirtualBox:~$ export CRS=33
vao#vao-VirtualBox:~$ psql t << EOF
> CREATE OR REPLACE FUNCTION point() RETURNS trigger
> LANGUAGE plpgsql
> AS \$\$
> BEGIN
> NEW.the_geom:=ST_SetSRID(geom, $CRS) ;
> RETURN NEW;
> END
> \$\$;
> \sf point()
> EOF
CREATE FUNCTION
CREATE OR REPLACE FUNCTION public.point()
RETURNS trigger
LANGUAGE plpgsql
AS $function$
BEGIN
NEW.the_geom:=ST_SetSRID(geom, 33) ;
RETURN NEW;
END
$function$
Of course it can be scripted to a file that you can call as bash with argument
Shell substitution, as Vao Tsun suggests, might work fine for small psql scripts, but it's still not a general solution. For instance, shells are not aware of psql and pgSQL syntaxes and may substitute sub-strings that are not meant to be substituted in the psql script:
-- If CA and AU are undefined environment variable (likely),
-- the following SQL, after shell substitution, will result in:
-- ... VALUES ('Australian Dollar', ''), (Canadian Dollar, '');
INSERT (name, symbol) INTO currencies VALUES
('Australian Dollar', '$AU'),
('Canadian Dollar', '$CA');
This will become unmanageable with large scripts without additional measures. A better approach is to use string concatenation. Something like this would be ideal:
-- This won't work; syntax error
CREATE FUNCTION point() RETURNS trigger
LANGUAGE plpgsql
AS $$
BEGIN
NEW.the_geom := ST_SetSRID(geom, $$ || :'CRS' || $$);
RETURN NEW;
END
$$;
Unfortunately, pgSQL syntax expects literal strings as function bodies, not arbitrary string expressions. Fortunately, however, PostgreSQL implements as part of its SQL syntax implicit concatenation between single-quoted strings separated by only whitespaces and at least one line break:
-- This gets parsed as "SELECT 'mystring';"
SELECT 'my'
'string';
The catch though is that it only works with single-quoted strings, not dollar-quoted strings. The following will work fine, but it's an eye-sore:
-- This will work, but it's ugly
CREATE FUNCTION point() RETURNS trigger
LANGUAGE plpgsql
AS
-- Mind the trailing space below
'BEGIN '
'NEW.the_geom := ST_SetSRID(geom, $$'
:'CRS'
'$$);'
'RETURN NEW;'
'END';
This is perhaps a more eye-pleasing variant:
-- This will work
-- (assign appropriate ownership and permissions if CRS contains sentitive data)
CREATE FUNCTION _get_crs() RETURNS text LANGUAGE sql AS 'SELECT $$'
:'crs'
'$$';
CREATE FUNCTION point() RETURNS trigger
LANGUAGE plpgsql
AS $$
BEGIN
NEW.the_geom := ST_SetSRID(geom, _get_crs());
RETURN NEW;
END
$$;
And here's another variant:
-- This will also work
-- (may expose sensitive data in the catalog)
SET "app.crs" = :'crs';
CREATE FUNCTION point() RETURNS trigger
LANGUAGE plpgsql
AS $$
BEGIN
NEW.the_geom := ST_SetSRID(geom, current_setting('app.crs'));
RETURN NEW;
END
$$;
The function body is a sting literal. psql has hardly any string modification functionality. However, PostgreSQL has plenty of it. With psql variables and \gset (since PostgreSQL 9.3) you can utilize this to first build your command and then execute it:
--for demo purpose set CRS directly. The OP did pass it from outside to psql
\set CRS 1234
--here we wrap the whole CREATE FUNCTION call in a literal separated by $_$
--this improves the readabilty and covers the inner $$. The result will be
--stored in psql variable func. This is workaround for a multi-line \set
SELECT $_$
CREATE FUNCTION point() RETURNS trigger
LANGUAGE plpgsql
AS $$
BEGIN
NEW.the_geom:=ST_SetSRID(geom, :CRS) ;
RETURN NEW;
END
$$;
$_$ AS func \gset
--replace the substring ':CRS' with content of variable CRS
--and store the result in func again
SELECT replace(:'func', ':CRS', :'CRS') AS func \gset
--we can now have a look to the resulting DDL
\echo :func
--and we can execute it:
:func
--clean up psql variables
\unset func
\unset CRS
with the help of \gexec (since PostgreSQL 9.6) we can also combine the last two steps:
--replace the substring ':CRS' with content of variable CRS
--and execute the result
SELECT replace(:'func', ':CRS', :'CRS') \gexec
or even do all in one command:
SELECT replace($_$
CREATE FUNCTION point() RETURNS trigger
LANGUAGE plpgsql
AS $$
BEGIN
NEW.the_geom:=ST_SetSRID(geom, :CRS) ;
RETURN NEW;
END
$$;
$_$, ':CRS', :'CRS') \gexec
Related
I'm trying to get FULL definition text of a function, but the query that everyone suggests returns the functions script without the header and footer parts.
Here's what I'm using:
select prosrc from pg_proc where proname='my_func';
Here's what is returned:
DECLARE
some_var1 TEXT;
some_var2 TEXT;
Status character varying;
BEGIN
--SCRIPT CODE HERE
RETURN retval;
END;
But what I want is this:
CREATE OR REPLACE FUNCTION my_func(logdate_utc date)
RETURNS character varying
LANGUAGE 'plpgsql'
COST 100
VOLATILE
AS $BODY$
DECLARE
some_var1 TEXT;
some_var2 TEXT;
Status character varying;
BEGIN
--SCRIPT CODE HERE
RETURN retval;
END;
$BODY$;
ALTER FUNCTION my_func(date)
OWNER TO some_owner;
Searched around and haven't found much. How can I get that result?
Using SQL:
select pg_catalog.pg_get_functiondef('my_func(date)'::regprocedure::oid);
Doc
Note that you should to use regprocedure type if you specify the parameters list and regproc type if you specify only function name (like 'my_func'::regproc::oid)
And using command line psql with \sf meta-command:
This command fetches and shows the definition of the named function, in the form of a CREATE OR REPLACE FUNCTION command. The definition is printed to the current query output channel, as set by \o.
The target function can be specified by name alone, or by name and arguments, for example foo(integer, text). The argument types must be given if there is more than one function of the same name.
from the documentation.
For example:
psql -c '\sf my_func(date)'
And to retrieve function's owner name you can use next query:
select proowner::regrole
from pg_proc
where oid = 'my_func(date)'::regprocedure;
Use pg_get_functiondef(func_oid):
SELECT pg_get_functiondef('my_func'::regproc);
SELECT pg_get_functiondef(( SELECT oid FROM pg_proc where proname= 'my_func'));
I'm writing my own regression test, and I'm using \o command to output my query result to a temporary file. For example
create or replace function aaaa()
returns integer as
$$
BEGIN
RAISE NOTICE 'ssss';
RETURN 1;
END;
$$ language plpgsql;
\o a.tmp
select aaaa();
With those commands, I'm able to output the return value (1 in this case) to file a.tmp. However, the message NOTICE: ssss is print directly to my psql client.
I'm wondering if there is a way to print the NOTICE: ssss also to a.tmp.
I'm trying to get FULL definition text of a function, but the query that everyone suggests returns the functions script without the header and footer parts.
Here's what I'm using:
select prosrc from pg_proc where proname='my_func';
Here's what is returned:
DECLARE
some_var1 TEXT;
some_var2 TEXT;
Status character varying;
BEGIN
--SCRIPT CODE HERE
RETURN retval;
END;
But what I want is this:
CREATE OR REPLACE FUNCTION my_func(logdate_utc date)
RETURNS character varying
LANGUAGE 'plpgsql'
COST 100
VOLATILE
AS $BODY$
DECLARE
some_var1 TEXT;
some_var2 TEXT;
Status character varying;
BEGIN
--SCRIPT CODE HERE
RETURN retval;
END;
$BODY$;
ALTER FUNCTION my_func(date)
OWNER TO some_owner;
Searched around and haven't found much. How can I get that result?
Using SQL:
select pg_catalog.pg_get_functiondef('my_func(date)'::regprocedure::oid);
Doc
Note that you should to use regprocedure type if you specify the parameters list and regproc type if you specify only function name (like 'my_func'::regproc::oid)
And using command line psql with \sf meta-command:
This command fetches and shows the definition of the named function, in the form of a CREATE OR REPLACE FUNCTION command. The definition is printed to the current query output channel, as set by \o.
The target function can be specified by name alone, or by name and arguments, for example foo(integer, text). The argument types must be given if there is more than one function of the same name.
from the documentation.
For example:
psql -c '\sf my_func(date)'
And to retrieve function's owner name you can use next query:
select proowner::regrole
from pg_proc
where oid = 'my_func(date)'::regprocedure;
Use pg_get_functiondef(func_oid):
SELECT pg_get_functiondef('my_func'::regproc);
SELECT pg_get_functiondef(( SELECT oid FROM pg_proc where proname= 'my_func'));
I may be contorting postgres here, but fundamentally, what I would like to do is take a string variable and pass it to an sql command (in this case COPY) using only psql.
So this is what I came up with. The commands are separated into 2 files because I want to be able to use the mydb_functions in other situations:
file one: mydb_functions--1.0.sql (in share/extension and with mydb_functions.control also setup as described in the manual. Given a filename, returns a full filepath. this is done solely to make the COPY statements in add_data.sql below, neater.
CREATE OR REPLACE FUNCTION fpn(filename text) RETURNS TEXT as '
DECLARE
mypath text := ''/path/to/my/directory/'';
BEGIN
RETURN mypath || filename;
END
' LANGUAGE plpgsql;
file two: add_data.sql . This exists solely to copy data into existing postgres tables using psql at the command line. Note: running psql with superuser privileges is required because of the CREATE EXTENSION command.
CREATE EXTENSION IF NOT EXISTS mydb_functions;
-- haven't figured out how to create a function without arguments yet.
CREATE OR REPLACE FUNCTION test (dummyvar text) RETURNS text as '
DECLARE
filepath RECORD;
BEGIN
SELECT * INTO filepath from fpn(''mydatafile.data'');
COPY tablename (columnname) FROM filepath;
END
' LANGUAGE plpgsql;
The part I am stuck on is how to extract a text from the filepath record to use in the COPY command. Any tips on an easier way to achieve this are also welcome. I think creating a table to store the variable is far easier than this. But I would like to finish the last step.
If your issue is running COPY with a dynamic target path, use EXECUTE to run a formatted SQL query.
CREATE OR REPLACE FUNCTION test () RETURNS text as $$
DECLARE
filepath RECORD;
BEGIN
SELECT * INTO filepath from fpn('mydatafile.data');
EXECUTE format('COPY tablename (columnname) FROM %L', filepath);
END
$$ LANGUAGE plpgsql;
See, this fails:
DO
$$
DECLARE
somevar text;
BEGIN
somevar := '/tmp/somepath.csv';
COPY tablename(columnname) FROM somevar;
END;
$$;
but this works:
DO
$$
DECLARE
somevar text;
BEGIN
somevar := '/tmp/somepath.csv';
EXECUTE format('COPY tablename(columnname) FROM %L', somevar);
END;
$$;
See:
EXECUTE
format()
format()'s %L specifier auto-quotes literals. %I does the same for identifiers.
If as I originally thought, you're talking about getting data into psql from outside, you may find psql variables and variable interpolation useful:
$ psql -v filepath=/path/to/my/directory/mydatafile.data regress
regress=> SELECT :'filepath';
?column?
---------------------------------------
/path/to/my/directory/mydatafile.data
(1 row)
Note that the colon is unquoted, then the variable name is quoted. Odd syntax, I know. This only works in psql; it won't work in (say) PgAdmin-III.
Alternately, you can use a here-document (unix-like shell specific, won't work in Windows' cmd.exe) to do quoted text interpolation:
$ FILEPATH=/path/to/my/directory/mydatafile.data
$ psql regress <<__END__
SELECT '$FILEPATH';
__END__
?column?
---------------------------------------
/path/to/my/directory/mydatafile.data
(1 row)
Both of these show how to insert a variable from the shell level into psql. From there, it's a matter of using it in your function. Like this, say, with -v filename=myfilename.csv:
CREATE OR REPLACE FUNCTION test() RETURNS text as $$
DECLARE
filepath RECORD;
BEGIN
SELECT * INTO filepath from fpn(:'filename');
COPY tablename (columnname) FROM filepath;
END
$$ LANGUAGE plpgsql;
or just this with -v filepath=/full/path/to/myfilename.csv:
$ psql -v filepath=/path/to/my/directory/mydatafile.data regress
regress=> COPY tablename (columnname) FROM :'filepath';
I once read this entry in mailing list http://archives.postgresql.org/pgsql-hackers/2005-06/msg01481.php
SELECT *
FROM foo_func(
c => current_timestamp::timestamp with time zone,
a => 2,
b => 5
);
Now I need this kindof solution where I can pass associative array argument to a function.
Do I need to make a dummy table and then use that table as argument type ? or there is any straight forward fix for this ? or has this hack been implemented ?
or can I emulate the same using pl/python ?
Here are the steps for an answer with hstore and PG-8.4 for debian.
1) if not installed already, install the contrib package
# apt-get install postgresql-contrib-8.4
2) install hstore in the relevant database
$ psql -U postgres -d dbname
# \i /usr/share/postgresql/8.4/contrib/hstore.sql
2bis) If the plpgsql language is not installed, install it (still inside psql as postgres user)
# CREATE LANGUAGE plpgsql;
3) create the function taking hstore as input. Here's an example in plpgsql that just enumerates the keys and values:
CREATE OR REPLACE function enum_hstore(in_h hstore) returns void
as $$
declare
kv record;
begin
for kv in select * from (select (each(in_h)).*) as f(k,v) loop
raise notice 'key=%,value=%',kv.k,kv.v;
end loop;
end
$$ language plpgsql;
4) call the function. Since the keys and values are of type text, it may be necessary to cast to text the non-literal entries, as the current_timestamp call in the question. Example:
select enum_hstore(
hstore('c',current_timestamp::text) ||
'a=>2,b=>5'::hstore
);
The result to expect from the above function:
NOTICE: key=a,value=2
NOTICE: key=b,value=5
NOTICE: key=c,value=2012-04-08 16:12:59.410056+02
This is implemented in version 9.0:
4.3.2. Using named notation
In named notation, each argument's name is specified using := to
separate it from the argument expression. For example:
SELECT concat_lower_or_upper(a := 'Hello', b := 'World');
concat_lower_or_upper
-----------------------
hello world
(1 row)