dialyzer: Improve compression of an ETS table

3 files changed, 46 insertions, 16 deletions

diff --git a/lib/dialyzer/src/dialyzer_codeserver.erl b/lib/dialyzer/src/dialyzer_codeserver.erl
index a1a7370eff..a8422d313e 100644
--- a/lib/dialyzer/src/dialyzer_codeserver.erl
+++ b/lib/dialyzer/src/dialyzer_codeserver.erl
@@ -304,9 +304,29 @@ lookup_temp_mod_records(Mod, #codeserver{temp_records = TempRecDict}) ->
 
 finalize_records(#codeserver{temp_records = TmpRecords,
                              records = Records} = CS) ->
-  true = ets:delete(Records),
-  ets:rename(TmpRecords, dialyzer_codeserver_records),
-  CS#codeserver{temp_records = clean, records = TmpRecords}.
+  %% The annotations of the abstract code are reset as they are no
+  %% longer needed, which makes the ETS table compression better.
+  A0 = erl_anno:new(0),
+  AFun = fun(_) -> A0 end,
+  FFun = fun({F, Abs, Type}) ->
+               NewAbs = erl_parse:map_anno(AFun, Abs),
+               {F, NewAbs, Type}
+         end,
+  ArFun = fun({Arity, Fields}) -> {Arity, lists:map(FFun, Fields)} end,
+  List = dialyzer_utils:ets_tab2list(TmpRecords),
+  true = ets:delete(TmpRecords),
+  Fun = fun({Mod, Map}) ->
+            MFun =
+              fun({record, _}, {FileLine, ArityFields}) ->
+                    {FileLine, lists:map(ArFun, ArityFields)};
+                 (_, {{M, FileLine, Abs, Args}, Type}) ->
+                    {{M, FileLine, erl_parse:map_anno(AFun, Abs), Args}, Type}
+              end,
+            {Mod, maps:map(MFun, Map)}
+        end,
+  NewList = lists:map(Fun, List),
+  true = ets:insert(Records, NewList),
+  CS#codeserver{temp_records = clean}.
 
 -spec lookup_mod_contracts(atom(), codeserver()) -> contracts().
 
diff --git a/lib/dialyzer/src/dialyzer_plt.erl b/lib/dialyzer/src/dialyzer_plt.erl
index fa242555c2..1c1814e49c 100644
--- a/lib/dialyzer/src/dialyzer_plt.erl
+++ b/lib/dialyzer/src/dialyzer_plt.erl
@@ -604,20 +604,8 @@ give_away(#mini_plt{info = ETSInfo,
   true = ets:give_away(ETSExpTypes, Pid, any),
   ok.
 
-%% Somewhat slower than ets:tab2list(), but uses less memory.
 tab2list(T) ->
-  tab2list(ets:first(T), T, []).
-
-tab2list('$end_of_table', T, A) ->
-  case ets:first(T) of % no safe_fixtable()...
-    '$end_of_table' -> A;
-    Key -> tab2list(Key, T, A)
-  end;
-tab2list(Key, T, A) ->
-  Vs = ets:lookup(T, Key),
-  Key1 = ets:next(T, Key),
-  ets:delete(T, Key),
-  tab2list(Key1, T, Vs ++ A).
+  dialyzer_utils:ets_tab2list(T).
 
 %%---------------------------------------------------------------------------
 %% Edoc
diff --git a/lib/dialyzer/src/dialyzer_utils.erl b/lib/dialyzer/src/dialyzer_utils.erl
index 7eaa03ec30..0dbb2bcf0a 100644
--- a/lib/dialyzer/src/dialyzer_utils.erl
+++ b/lib/dialyzer/src/dialyzer_utils.erl
@@ -39,6 +39,7 @@
          sets_filter/2,
 	 src_compiler_opts/0,
 	 refold_pattern/1,
+         ets_tab2list/1,
 	 parallelism/0,
          family/1
 	]).
@@ -989,6 +990,27 @@ label(Tree) ->
 
 %%------------------------------------------------------------------------------
 
+-spec ets_tab2list(ets:tid()) -> list().
+
+%% Deletes the contents of the table. Use:
+%%  ets_tab2list(T), ets:delete(T)
+%% instead of:
+%%  ets:tab2list(T), ets:delete(T)
+%% to save some memory at the expense of somewhat longer execution time.
+ets_tab2list(T) ->
+  tab2list(ets:first(T), T, []).
+
+tab2list('$end_of_table', T, A) ->
+  case ets:first(T) of % no safe_fixtable()...
+    '$end_of_table' -> A;
+    Key -> tab2list(Key, T, A)
+  end;
+tab2list(Key, T, A) ->
+  Vs = ets:lookup(T, Key),
+  Key1 = ets:next(T, Key),
+  ets:delete(T, Key),
+  tab2list(Key1, T, Vs ++ A).
+
 -spec parallelism() -> integer().
 
 parallelism() ->