[erlang-questions] dialyzer and #record vs tuple

[Jonathan Leivent]

I have a similar problem with Dialyzer (which I run all the time using 
Emacs flymake mode), but on match specs of records with typed fields. 
In this case, Dialyzer sees atoms like '$1' and '_' as field values, 
which don't match the record field types.  It will then invariably 
complain with "Warning: Function ... has no local return" for functions 
that contain the resulting match specs.  This happens whether the match 
specs are composed by hand or by ets:fun2ms/ms_transform.hrl magic.

To prevent this, I add the following:

-ifdef(dialyzer).
dialyzer_fake_fun2ms(F) when is_function(F, 1) ->
     [{ok, [ok], [ok]}].
-define(msify(X), dialyzer_fake_fun2ms(X)).
-else.
-define(msify(X), ets:fun2ms(X)).
-endif.

Then put the functions being transformed within ?msify macro calls, like so:

-record(foo, {a :: integer(), b :: integer(), c :: integer()}).

get_foos() ->
   ets:select(foos, ?msify(fun (#foo{a=0, b=X, c=Y})
                                when X > 5 -> Y end)).

Then run Dialyzer (but not the compiler) with -Ddialyzer.

-- Jonathan Leivent