use lib qw(t/lib);
use DBICTest::RunMode;
-use DBICTest::Util::LeakTracer qw/populate_weakregistry assert_empty_weakregistry/;
+use DBICTest::Util::LeakTracer qw(populate_weakregistry assert_empty_weakregistry visit_refs);
+use Scalar::Util qw(weaken blessed reftype);
use DBIx::Class;
+use DBIx::Class::_Util qw(hrefaddr sigwarn_silencer);
BEGIN {
plan skip_all => "Your perl version $] appears to leak like a sieve - skipping test"
if DBIx::Class::_ENV_::PEEPEENESS;
# re-populate the registry while checking it, ewwww!)
return $obj if (ref $obj) =~ /^TB2::/;
- # weaken immediately to avoid weird side effects
+ # populate immediately to avoid weird side effects
return populate_weakregistry ($weak_registry, $obj );
};
my $getcol_rs = $cds_rs->get_column('me.cdid');
my $pref_getcol_rs = $cds_with_stuff->get_column('me.cdid');
- # fire the column getters
- my @throwaway = $pref_getcol_rs->all;
-
my $base_collection = {
resultset => $rs,
get_column_rs_pref => $pref_getcol_rs,
# twice so that we make sure only one H::M object spawned
- chained_resultset => $rs->search_rs ({}, { '+columns' => [ 'foo' ] } ),
- chained_resultset2 => $rs->search_rs ({}, { '+columns' => [ 'bar' ] } ),
+ chained_resultset => $rs->search_rs ({}, { '+columns' => { foo => 'artistid' } } ),
+ chained_resultset2 => $rs->search_rs ({}, { '+columns' => { bar => 'artistid' } } ),
row_object => $row_obj,
leaky_resultset_cond => $cond_rowobj,
};
- # this needs to fire, even if it can't find anything
- # see FIXME below
- # we run this only on smokers - trying to establish a pattern
- $rs_bind_circref->next
- if ( ($ENV{TRAVIS}||'') ne 'true' and DBICTest::RunMode->is_smoker);
+ # fire all resultsets multiple times, once here, more below
+ # some of these can't find anything (notably leaky_resultset)
+ my @rsets = grep {
+ blessed $_
+ and
+ (
+ $_->isa('DBIx::Class::ResultSet')
+ or
+ $_->isa('DBIx::Class::ResultSetColumn')
+ )
+ } values %$base_collection;
+
+
+ my $fire_resultsets = sub {
+ local $ENV{DBIC_COLUMNS_INCLUDE_FILTER_RELS} = 1;
+ local $SIG{__WARN__} = sigwarn_silencer(
+ qr/Unable to deflate 'filter'-type relationship 'artist'.+related object primary key not retrieved/
+ );
+
+ map
+ { $_, (blessed($_) ? { $_->get_columns } : ()) }
+ map
+ { $_->all }
+ @rsets
+ ;
+ };
+
+ push @{$base_collection->{random_results}}, $fire_resultsets->();
+
+ # FIXME - something throws a Storable for a spin if we keep
+ # the results in-collection. The same problem is seen above,
+ # swept under the rug back in 0a03206a, damned lazy ribantainer
+{
+ local $base_collection->{random_results};
require Storable;
%$base_collection = (
fresh_pager => $rs->page(5)->pager,
pager => $pager,
);
+}
+
+ # FIXME - ideally this kind of collector ought to be global, but attempts
+ # with an invasive debugger-based tracer did not quite work out... yet
+ # Manually scan the innards of everything we have in the base collection
+ # we assembled so far (skip the DT madness below) *recursively*
+ #
+ # Only do this when we do have the bits to look inside CVs properly,
+ # without it we are liable to pick up object defaults that are locked
+ # in method closures
+ if (DBICTest::Util::LeakTracer::CV_TRACING) {
+ visit_refs(
+ refs => [ $base_collection ],
+ action => sub {
+ populate_weakregistry ($weak_registry, $_[0]);
+ 1; # true means "keep descending"
+ },
+ );
+
+ # do a heavy-duty fire-and-compare loop on all resultsets
+ # this is expensive - not running on install
+ my $typecounts = {};
+ unless (DBICTest::RunMode->is_plain or $ENV{DBICTEST_IN_PERSISTENT_ENV}) {
+
+ # FIXME - ideally we should be able to just populate an alternative
+ # registry, subtract everything from the main one, and arrive at
+ # an "empty" resulting hash
+ # However due to gross inefficiencies in the ::ResultSet code we
+ # end up recalculating a new set of aliasmaps which could have very
+ # well been cached if it wasn't for... anyhow
+ # What we do here for the time being is similar to the lazy approach
+ # of Devel::LeakTrace - we just make sure we do not end up with more
+ # reftypes than when we started. At least we are not blanket-counting
+ # SVs like D::LT does, but going by reftype... sigh...
+
+ for (values %$weak_registry) {
+ if ( my $r = reftype($_->{weakref}) ) {
+ $typecounts->{$r}--;
+ }
+ }
+
+ # For now we can only reuse the same registry, see FIXME above/below
+ #for my $interim_wr ({}, {}) {
+ for my $interim_wr ( ($weak_registry) x 4 ) {
+
+ visit_refs(
+ refs => [ $fire_resultsets->(), @rsets ],
+ action => sub {
+ populate_weakregistry ($interim_wr, $_[0]);
+ 1; # true means "keep descending"
+ },
+ );
+
+ # FIXME - this is what *should* be here
+ #
+ ## anything we have seen so far is cool
+ #delete @{$interim_wr}{keys %$weak_registry};
+ #
+ ## moment of truth - the rest ought to be gone
+ #assert_empty_weakregistry($interim_wr);
+ }
+
+ for (values %$weak_registry) {
+ if ( my $r = reftype($_->{weakref}) ) {
+ $typecounts->{$r}++;
+ }
+ }
+ }
+
+ for (keys %$typecounts) {
+ fail ("Amount of $_ refs changed by $typecounts->{$_} during resultset mass-execution")
+ if ( abs ($typecounts->{$_}) > 1 ); # there is a pad caught somewhere, the +1/-1 can be ignored
+ }
+ }
if ($has_dt) {
my $rs = $base_collection->{icdt_rs} = $schema->resultset('Event');
$base_collection->{"DBI handle $_"} = $_;
}
- SKIP: {
- if ( DBIx::Class::Optional::Dependencies->req_ok_for ('test_leaks') ) {
- my @w;
- local $SIG{__WARN__} = sub { $_[0] =~ /\QUnhandled type: REGEXP/ ? push @w, @_ : warn @_ };
-
- Test::Memory::Cycle::memory_cycle_ok ($base_collection, 'No cycles in the object collection');
-
- if ( $] > 5.011 ) {
- local $TODO = 'Silence warning due to RT56681';
- is (@w, 0, 'No Devel::Cycle emitted warnings');
- }
- }
- else {
- skip 'Circular ref test needs ' . DBIx::Class::Optional::Dependencies->req_missing_for ('test_leaks'), 1;
- }
- }
-
populate_weakregistry ($weak_registry, $base_collection->{$_}, "basic $_")
for keys %$base_collection;
}
# T::B 2.0 has result objects and other fancyness
delete $weak_registry->{$addr};
}
- elsif ($names =~ /^Method::Generate::(?:Accessor|Constructor)/m) {
- # Moo keeps globals around, this is normal
- delete $weak_registry->{$addr};
- }
- elsif ($names =~ /^SQL::Translator::Generator::DDL::SQLite/m) {
- # SQLT::Producer::SQLite keeps global generators around for quoted
- # and non-quoted DDL, allow one for each quoting style
- delete $weak_registry->{$addr}
- unless $cleared->{sqlt_ddl_sqlite}->{@{$weak_registry->{$addr}{weakref}->quote_chars}}++;
- }
elsif ($names =~ /^Hash::Merge/m) {
# only clear one object of a specific behavior - more would indicate trouble
delete $weak_registry->{$addr}
unless $cleared->{hash_merge_singleton}{$weak_registry->{$addr}{weakref}{behavior}}++;
}
- elsif ($names =~ /^DateTime::TimeZone/m) {
- # DT is going through a refactor it seems - let it leak zones for now
- delete $weak_registry->{$addr};
+ elsif (
+ # if we can look at closed over pieces - we will register it as a global
+ !DBICTest::Util::LeakTracer::CV_TRACING
+ and
+ $names =~ /^SQL::Translator::Generator::DDL::SQLite/m
+ ) {
+ # SQLT::Producer::SQLite keeps global generators around for quoted
+ # and non-quoted DDL, allow one for each quoting style
+ delete $weak_registry->{$addr}
+ unless $cleared->{sqlt_ddl_sqlite}->{@{$weak_registry->{$addr}{weakref}->quote_chars}}++;
}
}
# FIXME !!!
# There is an actual strong circular reference taking place here, but because
-# half of it is in XS no leaktracer sees it, and Devel::FindRef is equally
-# stumped when trying to trace the origin. The problem is:
+# half of it is in XS, so it is a bit harder to track down (it stumps D::FR)
+# (our tracker does not yet do it, but it'd be nice)
+# The problem is:
#
# $cond_object --> result_source --> schema --> storage --> $dbh --> {CachedKids}
# ^ /
# \-------- bound value on prepared/cached STH <-----------/
#
{
- local $TODO = 'This fails intermittently - see RT#82942';
- if ( my $r = ($weak_registry->{'basic leaky_resultset_cond'}||{})->{weakref} ) {
+ my @circreffed;
+
+ for my $r (map
+ { $_->{weakref} }
+ grep
+ { $_->{slot_names}{'basic leaky_resultset_cond'} }
+ values %$weak_registry
+ ) {
+ local $TODO = 'Needs Data::Entangled or somesuch - see RT#82942';
ok(! defined $r, 'Self-referential RS conditions no longer leak!')
- or $r->result_source(undef);
+ or push @circreffed, $r;
+ }
+
+ if (@circreffed) {
+ is (scalar @circreffed, 1, 'One resultset expected to leak');
+
+ # this is useless on its own, it is to showcase the circref-diag
+ # and eventually test it when it is operational
+ local $TODO = 'Needs Data::Entangled or somesuch - see RT#82942';
+ while (@circreffed) {
+ weaken (my $r = shift @circreffed);
+
+ populate_weakregistry( (my $mini_registry = {}), $r );
+ assert_empty_weakregistry( $mini_registry );
+
+ $r->result_source(undef);
+ }
}
}
projects / dbsrgits/DBIx-Class.git / blobdiff