Re^2: Regex: Identifying comments

[pvaldes]

in reply to Re: Regex: Identifying comments
in thread Regex: Identifying comments

IMHO the basic problem with your sql files is that don't look really like sql files, thus anything can be a comment and anything can be wrong, even if perl is acting fine.

I don't know all databases in the whole world so don't take this advice as immutable, of course, but sql sentences normally aren't complete (and don't run) until you find a semicolon. Is also surprising that you could select a field named '--foo' when you can't even create a table like this:

create table (--foo int);  # error
create table ('--foo' int);  # error
create table ("--foo" int);  # error
create table (\-\-foo int); # big error

So you are probably trying to solve a nonexistent problem. Comments are always after a semicolon or in your own line. And it you have i.e this:

-- ; select * from mytable where foo = "--34"; -- hi ', hi'. "hi'

The complete line IS a comment here. I'm quoting lines that I don't want to run (temporally) all the time in sql files, it's not unusual to see something like this.

So you should be pointing your arrows to ";" instead to "'". You probably should check and sanitize first what's in your files, before to pass this files to perl

[remiah]

Thanks for reply, pvaldes.

I noticed I misunderstood how to escape single quotes in sql. I thought of the case like this

insert into foo (bar,boo) values('''a', 'b');

And this is nothing surprise.

select '--foo';

I changed test data with real SQL with your advice. SQLite and Postgres accept these commands.

drop table foo;
create table foo(   --tabel foo
    bar text,       --fld name
    boo text        --again
);
--populate
insert into foo (bar,boo) values('''a', 'b'); -- insert quote's sake
insert into foo (bar,boo) values('c', 'd');   -- another line
select 'test' || 'abc' as a ;                 -- maybe you use dual wi
+th oracle?
select 'test''s' || 'abc' as a ;              -- maybe you use dual wi
+th oracle?
select '--foo';

So far, so good.

[pvaldes]

create table foo(   --tabel foo
    bar text,       --fld name
    boo text        --again
);

mmh... this is an interesting example, Yes. Perfectly legitim third type of comments, and difficult to debug with a regex... maybe you can profit that the fact that the type of data are limited with something like:

m/(\(|text|int|integer|char \(\d+\)),*\s*--(.*)$/

select '--foo';

This is not a very probable situation, but certainly is possible too. In any case this false comment is not after a semicolon, nor at the beginning of the line or inside a table, so if there are a ^\s*select in the same line you probably could safely ignore it. But then you could have something like this:

select field from mytable where field = 'text, --foo important information here about to be lost';

The safest actitude (although maybe a little paranoic) should be to isolate and examinate personally any case so special, the idea is: "if you found two - after a ' or a " and before a ";" in a line having the string "select" I want to see it personally"

You can improve your regex if you check previously for troublesome data:

select * from mitable where field1 like '%--%' or field2 like '%--%' or field3 like '%--%'... etc ;

[remiah]

umm..
No.

You will not see my regex fault with your examples. My regex will stumble with this sql.

update set bar = bar - 1 ; -- subtraction symbol may disappear.

I expected to see SQL parser solution in this thread, like this

my $p = SQLParser->new(type=>'mysql', sql=>$sql)
  or die SQLParser->error();

$p->prettyprint(1);
$p->without_comment(1);
print $p->sql;

At first I looked SQL::Parser. It seems quite near for such tasks, but I couldn't find good solution to rip off comments. Do you know such module?