From: Peter Rabbitson Date: Thu, 30 Aug 2012 13:32:50 +0000 (+0200) Subject: Comments on code X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=commitdiff_plain;h=c0992bf115d0477b8345b2672d9ee0f3e6c47034;p=dbsrgits%2FDBIx-Class.git Comments on code --- diff --git a/lib/DBIx/Class/Storage/DBI.pm b/lib/DBIx/Class/Storage/DBI.pm index 3f9c2e8..80df9ca 100644 --- a/lib/DBIx/Class/Storage/DBI.pm +++ b/lib/DBIx/Class/Storage/DBI.pm @@ -28,6 +28,9 @@ __PACKAGE__->mk_group_accessors('inherited' => qw/ _unsatisfied_deferred_constraints_autorollback /); +# I would change this to something that makes sense, like +# _autorollback_on_deferred_constraint_violation +# also setting it to an explicit 0 is silly # see with_deferred_fk_checks __PACKAGE__->_unsatisfied_deferred_constraints_autorollback(0); @@ -863,6 +866,9 @@ in MySQL's case disabled entirely. sub with_deferred_fk_checks { my ($self, $sub) = @_; + # now this has a lot of logic built in, including an unconditional + # try/catch block on top. If code ends up nesting with_deferred_fk_checks + # blocks - lots of crap will go wrong. We need a test for nesting too. if ($self->can('_set_constraints_deferred') && $self->can('_set_constraints_immediate')) { @@ -871,6 +877,7 @@ sub with_deferred_fk_checks { return try { my $guard = $self->txn_scope_guard; $self->_set_constraints_deferred; + # why { $sub->() } and not just $sub ? preserve_context { $sub->() } after => sub { my $e; eval { @@ -879,10 +886,22 @@ sub with_deferred_fk_checks { if ($@) { if ($self->_unsatisfied_deferred_constraints_autorollback) { $guard->{inactivated} = 1; # DO NOT ROLLBACK + + # *POSSIBLY* a code smell + # perhaps this shouldn't be here at all, ::Storage::txn_commit + # (which is called on all commits, including guards) should know to + # check for "am I in a deferred state" and "do I auto-rollback" + # and then do all the reductions on its own $self->{transaction_depth}--; } $e = $@; } + + # some engines (e.g. Pg) auto-unset deferred txns on rollback + # there should be an extra storage flag, and a provision to + # avoid this entire eval invoking a noop + # also the semantic is not clear what happens when an autorollback + # happens in the midst of a nested savepoint. Need more tests eval { $tried_to_reset_constraints = 1; $self->_set_constraints_immediate; @@ -900,6 +919,8 @@ sub with_deferred_fk_checks { } catch { my $e = $_; + + # same commend as on the eval above if (not $tried_to_reset_constraints) { eval { $self->_set_constraints_immediate; diff --git a/lib/DBIx/Class/Storage/DBI/Informix.pm b/lib/DBIx/Class/Storage/DBI/Informix.pm index dd6ffcd..dbd0ebd 100644 --- a/lib/DBIx/Class/Storage/DBI/Informix.pm +++ b/lib/DBIx/Class/Storage/DBI/Informix.pm @@ -50,12 +50,21 @@ sub _set_constraints_immediate { # fix it up here by doing a manual $dbh->do("COMMIT WORK"), propagating the # exception, and resetting the $dbh->{AutoCommit} attribute. +# this crap is now exeuted on *every* commit, be it deferred constraints or not +# moreover there is no DBD version check - the workaround will remain here +# forever and ever until someone decides to profile it +# so far the general way of doing this was to use the current CPAN available +# version, and keep bumping it as DBDs are released with the bug outstanding. +# of course having an explicit test for this helps - we do not have one sub _exec_txn_commit { my $self = shift; my $tried_resetting_autocommit = 0; try { + # what the fuck?! what's wrong with $dbh->commit for the *general* case? + # where is the detailed comment explaining why the violation of DBI + # internals? (the 2 lines above o not count as sufficient explanation) $self->_dbh->do('COMMIT WORK'); if ($self->_dbh_autocommit && $self->transaction_depth == 1) { eval { diff --git a/lib/DBIx/Class/Storage/DBI/Pg.pm b/lib/DBIx/Class/Storage/DBI/Pg.pm index 709ec60..0028813 100644 --- a/lib/DBIx/Class/Storage/DBI/Pg.pm +++ b/lib/DBIx/Class/Storage/DBI/Pg.pm @@ -34,6 +34,8 @@ sub _set_constraints_deferred { # transaction" so that they can be checked. sub _set_constraints_immediate { + # the semantic is not clear what happens when an autorollback happens in the + # midst of a nested savepoint (txn_depth is set all the same). Moar tests $_[0]->_do_query('SET CONSTRAINTS ALL IMMEDIATE') if $_[0]->transaction_depth; } diff --git a/lib/DBIx/Class/Storage/DBI/mysql.pm b/lib/DBIx/Class/Storage/DBI/mysql.pm index 616d7d7..2eea92a 100644 --- a/lib/DBIx/Class/Storage/DBI/mysql.pm +++ b/lib/DBIx/Class/Storage/DBI/mysql.pm @@ -17,6 +17,8 @@ __PACKAGE__->_use_multicolumn_in (1); # We turn FOREIGN_KEY_CHECKS off, do a transaction, then turn them back on right # before the COMMIT so that they can be checked during the COMMIT. +# nothing tests this hypothesis, why? + sub with_deferred_fk_checks { my ($self, $sub) = @_;