Well, I would agree that
local is to be avoided where possible, but it's rather hard in this case:
sub main_loop {
local($dbh->{AutoCommit}) = 0;
... lots of code here, with multiple returns
}
To avoid a problem with DBD::Pg, I have all my database connection handles set to "AutoCommit" (otherwise, they start a new transaction after every COMMIT; daemons thus hold open transactions for hours at a time, waiting for the next request).
I turn the AutoCommit setting off just before starting to process something. I could turn it back on, but I'd have to put that before each and every
return statement.
Making a lexical copy of the object doesn't help:
my $tmp_dbh = $dbh;
$tmp_dbh->{AutoCommit} = 0; #BZZT!
because they're handles, what you do to one affects all the copies.
What I really need is a
continue block for subroutines ...
# Wishful thinking
sub main_loop {
... stuff happens ...
} continue { ... oops! I must have returned! }
Perhaps I could do this with a naked block?
sub main_loop {
{
... stuff happens ...
... use 'last' instead of 'return'?
}
... restore settings here
}
but
local for all its shortcomings, is much easier to understand by the next guy (or is it?).
As to the idea of a container: no, but not applicable in this case. One object is the return value of a subroutine (and is a package variable from elsewhere), the other is a package variable in the current package.
| -- |
| Jeff Boes |
| Database Engineer |
| Nexcerpt, Inc. |
|
|
|
...Nexcerpt...Connecting People With Expertise
|