Fix a qlc bug where filters were erroneously optimized away

Thanks to Sam Bobroff for reporting the bug.

1 files changed, 6 insertions, 5 deletions

diff --git a/lib/stdlib/src/qlc_pt.erl b/lib/stdlib/src/qlc_pt.erl
index c26764eb18..b6bb758dfb 100644
--- a/lib/stdlib/src/qlc_pt.erl
+++ b/lib/stdlib/src/qlc_pt.erl
@@ -1,7 +1,7 @@
 %%
 %% %CopyrightBegin%
 %%
-%% Copyright Ericsson AB 2004-2013. All Rights Reserved.
+%% Copyright Ericsson AB 2004-2014. All Rights Reserved.
 %%
 %% The contents of this file are subject to the Erlang Public License,
 %% Version 1.1, (the "License"); you may not use this file except in
@@ -1218,13 +1218,14 @@ lu_skip(ColConstants, FilterData, PatternFrame, PatternVars,
                  %% column, the filter will not be skipped.
                  %% (an example: {X=1} <- ..., X =:= 1).
                  length(D = Cols -- PatternColumns) =:= 1,
-                 Frame <- SFs,
-                 begin
+                 {{_,Col} = Column, Constants} <- D,
+                 %% Check that the following holds for all frames.
+                 lists:all(
+                   fun(Frame) ->
                      %% The column is compared/matched against a constant.
                      %% If there are no more comparisons/matches then
                      %% the filter can be replaced by the lookup of
                      %% the constant.
-                     [{{_,Col} = Column, Constants}] = D,
                      {VarI, FrameI} = unify_column(Frame, PV, Col, BindFun,
                                                    Imported),
                      VarValues = deref_skip(VarI, FrameI, LookupOp, Imported),
@@ -1253,7 +1254,7 @@ lu_skip(ColConstants, FilterData, PatternFrame, PatternVars,
                      length(VarValues) =< 1 andalso
                      (Constants -- LookedUpConstants =:= []) andalso
                      bindings_is_subset(Frame, F2, Imported)
-                 end],
+                   end, SFs)],
     ColFils = family_list(ColFil),
     %% The skip tag 'all' means that all filters are covered by the lookup.
     %% It does not imply that there is only one generator as is the case