inconsistent exception handling

[XSven]

The new() documentation contradicts the example given in the SYNOPSIS. The example in the SYNOPSIS uses eval{} to catch exceptions but doesn't use $Test::PostgreSQL::errstr but $@ instead. The new() documentation doesn't even mention that the constructor may throw exceptions in the case of error.

[TJC]

Thanks for raising this -- if you're happy to create a pull request improving it, I'm happy to merge it.

[XSven]

Ok then let's discuss the indented design first. I have read again the documentation that explains when the $errstr is set:

If required programs (initdb and postmaster) were not found, the function returns undef and sets appropriate message to $Test::PostgreSQL::errstr.

Today I have provoked an exception (a die $DBI::errstr;) in a _create_test_database() method call. This exception does not set $Test::ProstgreSQL::errstr although it was raised when calling the constructor new(). Conclusion: To trap exceptions when calling new(), $Test::ProstgreSQL::errstr and $@ both have to be checked. From my perspective this design is inconsistent and leads to errors.

[TJC]

Most of the code dates back to 2011 or earlier, with some small changes over the years. I haven't worked with Perl much for many years now, so I'm not actively maintaining this module myself. I'm happy to review and merge patches, and push releases to CPAN.

I agree that the current design isn't ideal by modern standards. It would make more sense to only handle exceptions, and not have the $Test::PostgreSQL::errstr. Would you like to refactor the code to support that?

[XSven]

Thanks for the clarification. Most likely I will not have time short term to work on this but let's see.