OK, here is my solution.
I post a more detailed answer on my Persagen.com blog.
Basically, I decided to abrogate the DO $$DECLARE ...
approach (described in SO 49950384) in favor of the simplified approach, below.
I am then able to access the BASH / PSQL shared variable, :bash_var
, thusly:
xpath('//metabolite', XMLPARSE(DOCUMENT convert_from(pg_read_binary_file(:'bash_var'))))
Here is a sample SQL script, illustrating that usage:
hmdb.sql
\c hmdb
CREATE TABLE hmdb_identifiers (
id SERIAL,
accession VARCHAR(15) NOT NULL,
name VARCHAR(300) NOT NULL,
cas_number VARCHAR(12),
pubchem_cid INT,
PRIMARY KEY (id),
UNIQUE (accession)
);
\echo '\n[hmdb.sql] bash_var:' :bash_var '\n'
-- UPDATE (2019-05-15): SEE MY COMMENTS BELOW RE: TEMP TABLE!
CREATE TEMP TABLE tmp_table AS
SELECT
(xpath('//accession/text()', x))[1]::text::varchar(15) AS accession
,(xpath('//name/text()', x))[1]::text::varchar(300) AS name
,(xpath('//cas_registry_number/text()', x))[1]::text::varchar(12) AS cas_number
,(xpath('//pubchem_compound_id/text()', x))[1]::text::int AS pubchem_cid
-- FROM unnest(xpath('//metabolite', XMLPARSE(DOCUMENT convert_from(pg_read_binary_file('hmdb/hmdb.xml'), 'UTF8')))) x
FROM unnest(xpath('//metabolite', XMLPARSE(DOCUMENT convert_from(pg_read_binary_file(:'bash_var'), 'UTF8')))) x
;
INSERT INTO hmdb_identifiers (accession, name, cas_number, pubchem_cid)
SELECT lower(accession), lower(name), lower(cas_number), pubchem_cid FROM tmp_table;
DROP TABLE tmp_table;
SQL script notes:
In the xpath statements I recast the ::text
(e.g.: ::text::varchar(15)
) per the Postgres table schema.
More significantly, if I did not recast the datatypes in the xpath statement and a field entry (e.g. name
length) exceeded the SQL varchar(300)
length limit, those data threw a PSQL error and the table did not update (i.e. a blank table results).
I uploaded the XML data files used in this answer at this Gist
https://gist.github.com/victoriastuart/d1b1959bd31e4de5ed951ff4fe3c3184
Direct links:
UPDATE (2019-05-15)
In follow-on work, detailed in my research blog post Exporting Plain Text to PostgreSQL, I directly load XML data into PostgreSQL, rather than using temp tables.
TL/DR. In that project, I observed the following improvements.
Parameter | Temp Tables | Direct Import | Reduction
Time: | 1048 min | 1.75 min | 599x
Space: | 252,000 MB | 18 MB | 14,000x