SQL compound statement awareness in ScriptUtils

[rnorth]

This came out of #570. In brief, for DB init script handling, ScriptUtils is not capable of handling scripts that contain compound statements in stored procedures (BEGIN...END sections). It will simply split the script on semicolons, mid-procedure, thus breaking the script.

An example would be:

CREATE PROCEDURE count_foo()
    BEGIN
      SELECT *
        FROM bar;   -- <-- ScriptUtils will split here, which is wrong
      SELECT 1
        FROM dual;
    END; -- <-- statement should end here

This prevents users from having stored procedures in their init scripts, which feels like an unpleasant limitation.

The Spring developers behind ScriptUtils chose not to go down the path of parsing the SQL, and I can see why they would do this. For arbitrary scripts and many SQL dialects, this could be quite open ended.

However, I wonder if we can still improve on this just enough to be useful with a small change, so have created this WIP PR. I'm not intending for this to be the solution to all possible problems, but hope it unlocks us in some ways.

As-is, this:

• won't split statements in between BEGIN/END tokens
• can deal with nested blocks
• seems to be OK with BEGIN/END appearing in string literals or comments

I don't believe this could create any regressions in its current form, and I believe BEGIN/END support is all we need to improve things greatly.

[jeacott1]

I don't think this fixes the problem, see my comment here:
#608 (comment)

[rnorth]

@bsideup @kiview I'd like to resurrect this PR, as it is a quick win to enable better SQL init scripts for JDBC containers. Issues such as #946 and #570 will continue to catch people out, which is silly when we have a solution that works well enough for typical usage.

Please could you review?

For reference init_mysql.sql shows the kinds of syntax that this PR enables, and is included in this PR for testing.

[bsideup]

@rnorth will do!

[kiview]

I just had some style remarks, actual logic seems fine for the specified case.

[rnorth]

Released in 1.10.4!

[manikmagar]

Thanks @rnorth, we ran into this issue with postgres a few weeks back. We will try it with latest version!

[johnhenaot]

Hi guys, I'm having some issues when using IF statement, it seems that it is assuming the word END in the "END IF" as the END of the BEGIN statement.

CREATE OR REPLACE PROCEDURE updatealladvertisers() LANGUAGE plpgsql AS $$ BEGIN IF (SELECT count(1) FROM information_schema.tables WHERE table_schema = 'public' AND table_name='all_advertisers') = 0 THEN CREATE TABLE IF NOT EXISTS public.all_advertisers ( partnerid BIGINT, partnername VARCHAR(255), partner VARCHAR(255), gsm_id BIGINT ); END IF; END $$;

Instead of using the BEGIN - END block to ignore ;, I think the solution is to ignore ; whenever is inside $$-$$