diff options
Diffstat (limited to 'lib/common_test/src/ct_testspec.erl')
-rw-r--r-- | lib/common_test/src/ct_testspec.erl | 133 |
1 files changed, 86 insertions, 47 deletions
diff --git a/lib/common_test/src/ct_testspec.erl b/lib/common_test/src/ct_testspec.erl index 2b6abefb72..2c7cd3577c 100644 --- a/lib/common_test/src/ct_testspec.erl +++ b/lib/common_test/src/ct_testspec.erl @@ -68,7 +68,8 @@ prepare_tests(TestSpec) when is_record(TestSpec,testspec) -> %% Create initial list of {Node,{Run,Skip}} tuples NodeList = lists:map(fun(N) -> {N,{[],[]}} end, list_nodes(TestSpec)), %% Get all Run tests sorted per node basis. - NodeList1 = run_per_node(Run,NodeList), + NodeList1 = run_per_node(Run,NodeList, + TestSpec#testspec.merge_tests), %% Get all Skip entries sorted per node basis. NodeList2 = skip_per_node(Skip,NodeList1), %% Change representation. @@ -89,11 +90,17 @@ prepare_tests(TestSpec) when is_record(TestSpec,testspec) -> %% run_per_node/2 takes the Run list as input and returns a list %% of {Node,RunPerNode,[]} tuples where the tests have been sorted %% on a per node basis. -run_per_node([{{Node,Dir},Test}|Ts],Result) -> +run_per_node([{{Node,Dir},Test}|Ts],Result, MergeTests) -> {value,{Node,{Run,Skip}}} = lists:keysearch(Node,1,Result), - Run1 = merge_tests(Dir,Test,Run), - run_per_node(Ts,insert_in_order({Node,{Run1,Skip}},Result)); -run_per_node([],Result) -> + Run1 = case MergeTests of + false -> + append({Dir, Test}, Run); + true -> + merge_tests(Dir,Test,Run) + end, + run_per_node(Ts,insert_in_order({Node,{Run1,Skip}},Result), + MergeTests); +run_per_node([],Result,_) -> Result. merge_tests(Dir,Test={all,_},TestDirs) -> @@ -281,6 +288,8 @@ collect_tests(Terms,TestSpec,Relaxed) -> {Terms2, TestSpec3} = filter_init_terms(Terms, [], TestSpec2), add_tests(Terms2,TestSpec3). +get_global([{merge_tests, Bool} | Ts], Spec) -> + get_global(Ts,Spec#testspec{ merge_tests = Bool }); get_global([{alias,Ref,Dir}|Ts],Spec=#testspec{alias=Refs}) -> get_global(Ts,Spec#testspec{alias=[{Ref,get_absdir(Dir,Spec)}|Refs]}); get_global([{node,Ref,Node}|Ts],Spec=#testspec{nodes=Refs}) -> @@ -668,7 +677,7 @@ add_tests([{suites,Node,Dir,Ss}|Ts],Spec) -> Tests = Spec#testspec.tests, Tests1 = insert_suites(ref2node(Node,Spec#testspec.nodes), ref2dir(Dir,Spec#testspec.alias), - Ss,Tests), + Ss,Tests, Spec#testspec.merge_tests), add_tests(Ts,Spec#testspec{tests=Tests1}); %% --- groups --- @@ -694,13 +703,15 @@ add_tests([{groups,Node,Dir,Suite,Gs}|Ts],Spec) -> Tests = Spec#testspec.tests, Tests1 = insert_groups(ref2node(Node,Spec#testspec.nodes), ref2dir(Dir,Spec#testspec.alias), - Suite,Gs,all,Tests), + Suite,Gs,all,Tests, + Spec#testspec.merge_tests), add_tests(Ts,Spec#testspec{tests=Tests1}); add_tests([{groups,Node,Dir,Suite,Gs,{cases,TCs}}|Ts],Spec) -> Tests = Spec#testspec.tests, Tests1 = insert_groups(ref2node(Node,Spec#testspec.nodes), ref2dir(Dir,Spec#testspec.alias), - Suite,Gs,TCs,Tests), + Suite,Gs,TCs,Tests, + Spec#testspec.merge_tests), add_tests(Ts,Spec#testspec{tests=Tests1}); %% --- cases --- @@ -715,7 +726,7 @@ add_tests([{cases,Node,Dir,Suite,Cs}|Ts],Spec) -> Tests = Spec#testspec.tests, Tests1 = insert_cases(ref2node(Node,Spec#testspec.nodes), ref2dir(Dir,Spec#testspec.alias), - Suite,Cs,Tests), + Suite,Cs,Tests, Spec#testspec.merge_tests), add_tests(Ts,Spec#testspec{tests=Tests1}); %% --- skip_suites --- @@ -730,7 +741,8 @@ add_tests([{skip_suites,Node,Dir,Ss,Cmt}|Ts],Spec) -> Tests = Spec#testspec.tests, Tests1 = skip_suites(ref2node(Node,Spec#testspec.nodes), ref2dir(Dir,Spec#testspec.alias), - Ss,Cmt,Tests), + Ss,Cmt,Tests, + Spec#testspec.merge_tests), add_tests(Ts,Spec#testspec{tests=Tests1}); %% --- skip_groups --- @@ -752,13 +764,15 @@ add_tests([{skip_groups,Node,Dir,Suite,Gs,Cmt}|Ts],Spec) -> Tests = Spec#testspec.tests, Tests1 = skip_groups(ref2node(Node,Spec#testspec.nodes), ref2dir(Dir,Spec#testspec.alias), - Suite,Gs,all,Cmt,Tests), + Suite,Gs,all,Cmt,Tests, + Spec#testspec.merge_tests), add_tests(Ts,Spec#testspec{tests=Tests1}); add_tests([{skip_groups,Node,Dir,Suite,Gs,{cases,TCs},Cmt}|Ts],Spec) -> Tests = Spec#testspec.tests, Tests1 = skip_groups(ref2node(Node,Spec#testspec.nodes), ref2dir(Dir,Spec#testspec.alias), - Suite,Gs,TCs,Cmt,Tests), + Suite,Gs,TCs,Cmt,Tests, + Spec#testspec.merge_tests), add_tests(Ts,Spec#testspec{tests=Tests1}); %% --- skip_cases --- @@ -773,7 +787,7 @@ add_tests([{skip_cases,Node,Dir,Suite,Cs,Cmt}|Ts],Spec) -> Tests = Spec#testspec.tests, Tests1 = skip_cases(ref2node(Node,Spec#testspec.nodes), ref2dir(Dir,Spec#testspec.alias), - Suite,Cs,Cmt,Tests), + Suite,Cs,Cmt,Tests,Spec#testspec.merge_tests), add_tests(Ts,Spec#testspec{tests=Tests1}); %% --- handled/errors --- @@ -783,6 +797,9 @@ add_tests([{alias,_,_}|Ts],Spec) -> % handled add_tests([{node,_,_}|Ts],Spec) -> % handled add_tests(Ts,Spec); +add_tests([{merge_tests, _} | Ts], Spec) -> % handled + add_tests(Ts,Spec); + %% check if it's a CT term that has bad format or if the user seems to %% have added something of his/her own, which we'll let pass if relaxed %% mode is enabled. @@ -835,17 +852,22 @@ separate([],_,_,_) -> %% {Suite2,[{GrOrCase21,{skip,Cmt}},GrOrCase22,...]},...]} %% GrOrCase = {GroupName,[Case1,Case2,...]} | Case -insert_suites(Node,Dir,[S|Ss],Tests) -> - Tests1 = insert_cases(Node,Dir,S,all,Tests), - insert_suites(Node,Dir,Ss,Tests1); -insert_suites(_Node,_Dir,[],Tests) -> +insert_suites(Node,Dir,[S|Ss],Tests, MergeTests) -> + Tests1 = insert_cases(Node,Dir,S,all,Tests,MergeTests), + insert_suites(Node,Dir,Ss,Tests1,MergeTests); +insert_suites(_Node,_Dir,[],Tests,_MergeTests) -> Tests; -insert_suites(Node,Dir,S,Tests) -> - insert_suites(Node,Dir,[S],Tests). +insert_suites(Node,Dir,S,Tests,MergeTests) -> + insert_suites(Node,Dir,[S],Tests,MergeTests). -insert_groups(Node,Dir,Suite,Group,Cases,Tests) when is_atom(Group) -> - insert_groups(Node,Dir,Suite,[Group],Cases,Tests); -insert_groups(Node,Dir,Suite,Groups,Cases,Tests) when +insert_groups(Node,Dir,Suite,Group,Cases,Tests,MergeTests) + when is_atom(Group) -> + insert_groups(Node,Dir,Suite,[Group],Cases,Tests,MergeTests); +insert_groups(Node,Dir,Suite,Groups,Cases,Tests,false) when + ((Cases == all) or is_list(Cases)) and is_list(Groups) -> + Groups1 = [{Gr,Cases} || Gr <- Groups], + append({{Node,Dir},[{Suite,Groups1}]},Tests); +insert_groups(Node,Dir,Suite,Groups,Cases,Tests,true) when ((Cases == all) or is_list(Cases)) and is_list(Groups) -> case lists:keysearch({Node,Dir},1,Tests) of {value,{{Node,Dir},[{all,_}]}} -> @@ -859,9 +881,10 @@ insert_groups(Node,Dir,Suite,Groups,Cases,Tests) when Groups1 = [{Gr,Cases} || Gr <- Groups], insert_in_order({{Node,Dir},[{Suite,Groups1}]},Tests) end; -insert_groups(Node,Dir,Suite,Groups,Case,Tests) when is_atom(Case) -> +insert_groups(Node,Dir,Suite,Groups,Case,Tests, MergeTests) + when is_atom(Case) -> Cases = if Case == all -> all; true -> [Case] end, - insert_groups(Node,Dir,Suite,Groups,Cases,Tests). + insert_groups(Node,Dir,Suite,Groups,Cases,Tests, MergeTests). insert_groups1(_Suite,_Groups,all) -> all; @@ -891,7 +914,9 @@ insert_groups2([Group={GrName,Cases}|Groups],GrAndCases) -> insert_groups2([],GrAndCases) -> GrAndCases. -insert_cases(Node,Dir,Suite,Cases,Tests) when is_list(Cases) -> +insert_cases(Node,Dir,Suite,Cases,Tests,false) when is_list(Cases) -> + append({{Node,Dir},[{Suite,Cases}]},Tests); +insert_cases(Node,Dir,Suite,Cases,Tests,true) when is_list(Cases) -> case lists:keysearch({Node,Dir},1,Tests) of {value,{{Node,Dir},[{all,_}]}} -> Tests; @@ -901,8 +926,8 @@ insert_cases(Node,Dir,Suite,Cases,Tests) when is_list(Cases) -> false -> insert_in_order({{Node,Dir},[{Suite,Cases}]},Tests) end; -insert_cases(Node,Dir,Suite,Case,Tests) when is_atom(Case) -> - insert_cases(Node,Dir,Suite,[Case],Tests). +insert_cases(Node,Dir,Suite,Case,Tests,MergeTests) when is_atom(Case) -> + insert_cases(Node,Dir,Suite,[Case],Tests,MergeTests). insert_cases1(_Suite,_Cases,all) -> all; @@ -917,22 +942,28 @@ insert_cases1(Suite,Cases,Suites0) -> insert_in_order({Suite,Cases},Suites0) end. -skip_suites(Node,Dir,[S|Ss],Cmt,Tests) -> - Tests1 = skip_cases(Node,Dir,S,all,Cmt,Tests), - skip_suites(Node,Dir,Ss,Cmt,Tests1); -skip_suites(_Node,_Dir,[],_Cmt,Tests) -> +skip_suites(Node,Dir,[S|Ss],Cmt,Tests,MergeTests) -> + Tests1 = skip_cases(Node,Dir,S,all,Cmt,Tests,MergeTests), + skip_suites(Node,Dir,Ss,Cmt,Tests1,MergeTests); +skip_suites(_Node,_Dir,[],_Cmt,Tests,_MergeTests) -> Tests; -skip_suites(Node,Dir,S,Cmt,Tests) -> - skip_suites(Node,Dir,[S],Cmt,Tests). - -skip_groups(Node,Dir,Suite,Group,all,Cmt,Tests) when is_atom(Group) -> - skip_groups(Node,Dir,Suite,[Group],all,Cmt,Tests); -skip_groups(Node,Dir,Suite,Group,Cases,Cmt,Tests) when is_atom(Group) -> - skip_groups(Node,Dir,Suite,[Group],Cases,Cmt,Tests); -skip_groups(Node,Dir,Suite,Groups,Case,Cmt,Tests) when is_atom(Case), - Case =/= all -> - skip_groups(Node,Dir,Suite,Groups,[Case],Cmt,Tests); -skip_groups(Node,Dir,Suite,Groups,Cases,Cmt,Tests) when +skip_suites(Node,Dir,S,Cmt,Tests,MergeTests) -> + skip_suites(Node,Dir,[S],Cmt,Tests,MergeTests). + +skip_groups(Node,Dir,Suite,Group,all,Cmt,Tests,MergeTests) + when is_atom(Group) -> + skip_groups(Node,Dir,Suite,[Group],all,Cmt,Tests,MergeTests); +skip_groups(Node,Dir,Suite,Group,Cases,Cmt,Tests,MergeTests) + when is_atom(Group) -> + skip_groups(Node,Dir,Suite,[Group],Cases,Cmt,Tests,MergeTests); +skip_groups(Node,Dir,Suite,Groups,Case,Cmt,Tests,MergeTests) + when is_atom(Case),Case =/= all -> + skip_groups(Node,Dir,Suite,Groups,[Case],Cmt,Tests,MergeTests); +skip_groups(Node,Dir,Suite,Groups,Cases,Cmt,Tests,false) when + ((Cases == all) or is_list(Cases)) and is_list(Groups) -> + Suites1 = skip_groups1(Suite,[{Gr,Cases} || Gr <- Groups],Cmt,[]), + append({{Node,Dir},Suites1},Tests); +skip_groups(Node,Dir,Suite,Groups,Cases,Cmt,Tests,true) when ((Cases == all) or is_list(Cases)) and is_list(Groups) -> Suites = case lists:keysearch({Node,Dir},1,Tests) of @@ -943,9 +974,10 @@ skip_groups(Node,Dir,Suite,Groups,Cases,Cmt,Tests) when end, Suites1 = skip_groups1(Suite,[{Gr,Cases} || Gr <- Groups],Cmt,Suites), insert_in_order({{Node,Dir},Suites1},Tests); -skip_groups(Node,Dir,Suite,Groups,Case,Cmt,Tests) when is_atom(Case) -> +skip_groups(Node,Dir,Suite,Groups,Case,Cmt,Tests,MergeTests) + when is_atom(Case) -> Cases = if Case == all -> all; true -> [Case] end, - skip_groups(Node,Dir,Suite,Groups,Cases,Cmt,Tests). + skip_groups(Node,Dir,Suite,Groups,Cases,Cmt,Tests,MergeTests). skip_groups1(Suite,Groups,Cmt,Suites0) -> SkipGroups = lists:map(fun(Group) -> @@ -959,7 +991,10 @@ skip_groups1(Suite,Groups,Cmt,Suites0) -> insert_in_order({Suite,SkipGroups},Suites0) end. -skip_cases(Node,Dir,Suite,Cases,Cmt,Tests) when is_list(Cases) -> +skip_cases(Node,Dir,Suite,Cases,Cmt,Tests,false) when is_list(Cases) -> + Suites1 = skip_cases1(Suite,Cases,Cmt,[]), + append({{Node,Dir},Suites1},Tests); +skip_cases(Node,Dir,Suite,Cases,Cmt,Tests,true) when is_list(Cases) -> Suites = case lists:keysearch({Node,Dir},1,Tests) of {value,{{Node,Dir},Suites0}} -> @@ -969,8 +1004,8 @@ skip_cases(Node,Dir,Suite,Cases,Cmt,Tests) when is_list(Cases) -> end, Suites1 = skip_cases1(Suite,Cases,Cmt,Suites), insert_in_order({{Node,Dir},Suites1},Tests); -skip_cases(Node,Dir,Suite,Case,Cmt,Tests) when is_atom(Case) -> - skip_cases(Node,Dir,Suite,[Case],Cmt,Tests). +skip_cases(Node,Dir,Suite,Case,Cmt,Tests,MergeTests) when is_atom(Case) -> + skip_cases(Node,Dir,Suite,[Case],Cmt,Tests,MergeTests). skip_cases1(Suite,Cases,Cmt,Suites0) -> SkipCases = lists:map(fun(C) -> @@ -984,6 +1019,9 @@ skip_cases1(Suite,Cases,Cmt,Suites0) -> insert_in_order({Suite,SkipCases},Suites0) end. +append(Elem, List) -> + List ++ [Elem]. + insert_in_order([E|Es],List) -> List1 = insert_elem(E,List,[]), insert_in_order(Es,List1); @@ -1056,6 +1094,7 @@ valid_terms() -> {userconfig,2}, {userconfig,3}, {alias,3}, + {merge_tests,1}, {logdir,2}, {logdir,3}, {label,2}, |