Re: Tighten up a few overly lax regexes in pg_dump's tap tests

Moin,

On Wed, February 6, 2019 8:10 pm, Michael Paquier wrote:
> On Wed, Feb 06, 2019 at 03:41:20PM +0100, Daniel Gustafsson wrote:
>> Correct.  One could argue that the regex is still suboptimal since
>> “COMMENT ON
>> DATABASE postgres IS  ;” will be matched as well, but there I think the
>> tradeoff
>> for readability wins.
>
> Okay, that looks like an improvement anyway, so committed after going
> over the tests for similar problems, and there was one for CREATE
> DATABASE and DROP ROLE.  It is possible to have a regex which tells as
> at least one non-whitespace character, but from what I can see these
> don't really improve the readability.

Sorry for being late to the party, but it just occured to me that while
".+"  is definitely an improvement over ".*", it isn't foolproof either,
as it also matches ";".

Thus:

  regexp => qr/^COMMENT ON DATABASE postgres IS .+;/m,

matches things like:

  'COMMENT ON DATABASE postgres IS ;;'
  'COMMENT ON DATABASE postgres IS  ;'
  'COMMENT ON DATABASE postgres IS --;'

etc.

I'm not sure it is really necessary to deal with these cases, but one
possibility would be to pre-compute regexps:

  $QR_COMMENT = qr/[^ ;]+/;
  $QR_IDENTIFIER = qr/[^ ;]+/; # etc

or whataver is the thing that should actually be matched here and use them
like so:

  regexp => qr/^COMMENT ON DATABASE postgres IS $QR_COMMENT;/m,

That way it is easily changable and quite readable.

Oh, one more question. Shouldn't these regexps that start with "^" also
end with "$"? Or can there be output like:

  'COMMENT ON DATABASE postgres IS $QR_IDENTIFIER; SELECT 1;'

?

Best regards,

Tels