From: Rob Kinyon <rkinyon@cpan.org>
Date: Fri, 6 Mar 2009 19:55:57 +0000 (+0000)
Subject: Added solution for prefetch + hasmany bug
X-Git-Tag: v0.08240~45
X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=commitdiff_plain;h=87333f6cd45d93712940f7a8abd7989fe2385d7d;p=dbsrgits%2FDBIx-Class.git

Added solution for prefetch + hasmany bug
---

diff --git a/t/prefetch/multiple_hasmany.t b/t/prefetch/multiple_hasmany.t
index 6ef7a53..9f858fb 100644
--- a/t/prefetch/multiple_hasmany.t
+++ b/t/prefetch/multiple_hasmany.t
@@ -128,3 +128,43 @@ my $w;
             "warning on ->$_ attempt prefetching several same level has_manys (M -> 1 -> M + M)");
     }
 }
+
+__END__
+The solution is to rewrite ResultSet->_collapse_result() and
+ResultSource->resolve_prefetch() to focus on the final results from the collapse
+of the data. Right now, the code doesn't treat the columns from the various
+tables as grouped entities. While there is a concept of hierarchy (so that
+prefetching down relationships does work as expected), there is no idea of what
+the final product should look like and how the various columns in the row would
+play together. So, the actual prefetch datastructure from the search would be
+very useful in working through this problem. We already have access to the PKs
+and sundry for those. So, when collapsing the search result, we know we are
+looking for 1 cd object. We also know we're looking for tracks and tags records
+-independently- of each other. So, we can grab the data for tracks and data for
+tags separately, uniqueing on the PK as appropriate. Then, when we're done with
+the given cd object's datastream, we know we're good. This should work for all
+the various scenarios.
+
+My reccommendation is the row's data is preprocessed first, breaking it up into
+the data for each of the component tables. (This could be done in the single
+table case, too, but probably isn't necessary.) So, starting with something
+like:
+  my $row = {
+    t1.col1 => 1,
+    t1.col2 => 2,
+    t2.col1 => 3,
+    t2.col2 => 4,
+    t3.col1 => 5,
+    t3.col2 => 6,
+  };
+it is massaged to look something like:
+  my $row_massaged = {
+    t1 => { col1 => 1, col2 => 2 },
+    t2 => { col1 => 3, col2 => 4 },
+    t3 => { col1 => 5, col2 => 6 },
+  };
+At this point, find the stuff that's different is easy enough to do and slotting
+things into the right spot is, likewise, pretty straightforward.
+
+This implies that the collapse attribute can probably disappear or, at the
+least, be turned into a boolean (which is how it's used in every other place).