[erlang-bugs] Possible Dialyzer bug (reproducible test case)

[Jesper Louis Andersen]

I posted this as a gist as well,

https://gist.github.com/jlouis/8456179

There are many valuable comments in there.

It turns out that this is a proper dialyzer error where it crashes, on
different environments. It also crashes in R16B02, so the regression is
older than that.



On Thu, Jan 16, 2014 at 4:09 PM, Jesper Louis Andersen <
jesper.louis.andersen@REDACTED> wrote:

> I may be totally off here, but I have a strange bug which I have seen
> twice now on R16B03, OSX for the Dialyzer. To reproduce it, execute the
> following steps:
>
> # Clone repository and force a specific version
> ; git clone https://github.com/jlouis/msgpack-erlang.git
> ; cd msgpack-erlang
> ; git checkout origin/dialyzer-failure
> ; make
> ; make build_plt # Takes about 2 minutes
> ; make dialyzer
>
> Gives the following:
>
> =ERROR REPORT==== 16-Jan-2014::15:52:18 ===
> Error in process <0.67.0> with exit value:
> {function_clause,[{erl_types,unify_lists,[[{c,tuple,[{c,atom,[ext],unknown},{c,atom,[msgpack_term],unknown}],{2,{c,atom,[ext],unknown}}}],[],[],[{c,tuple,[{c,atom,[enable_str],unknown},{c,atom,[true],unknown}],{2,{c,atom...
>
>
> dialyzer: Analysis failed with error:
> {function_clause,[{erl_types,unify_lists,
>                              [[{c,tuple,[...],...}],[],[],[{...}|...],[]],
>                              [{file,"erl_types.erl"},{line,2695}]},
>                   {erl_types,t_unify,4,[{file,[...]},{line,...}]},
>                   {erl_types,t_unify,4,[{file,...},{...}]},
>                   {erl_types,t_unify,3,[{...}|...]},
>                   {dialyzer_typesig,solve_subtype,4,[...]},
>                   {dialyzer_typesig,mk_constraint,3,...},
>                   {dialyzer_typesig,state__store_conj,...},
>                   {dialyzer_typesig,...}]}
> Last messages in the log cache:
>   Reading files and computing callgraph... done in 0.27 secs
>   Removing edges... done in 0.04 secs
>
> =ERROR REPORT==== 16-Jan-2014::15:52:18 ===
> Error in process <0.69.0> with exit value:
> {badarg,[{ets,insert,[790562,[{{msgpack,pack,1},{{c,binary,[8,8],unknown},[{c,union,[none,{c,binary,[8,0],unknown},none,none,{c,list,[{c,union,[none,{c,binary,[8,0],unknown},none,none,{c,list,[any,{c,nil,[],unknown}],unknown...
>
> make: *** [dialyzer] Error 1
>
> If you run `git show origin/dialyzer-failure` you have the following patch
> chunk:
>
> diff --git a/include/msgpack.hrl b/include/msgpack.hrl
> index 6ef046d..4796492 100644
> --- a/include/msgpack.hrl
> +++ b/include/msgpack.hrl
> @@ -63,8 +63,8 @@
>            impl = erlang      :: erlang | nif,
>            allow_atom = none  :: none | pack, %% allows atom when packing
>            enable_str = false :: boolean(), %% true for new spec
> -          ext_packer = undefined :: msgpack_ext_packer(),
> -          ext_unpacker = undefined :: msgpack_ext_unpacker(),
> +          ext_packer = undefined :: undefined | msgpack_ext_packer(),
> +          ext_unpacker = undefined :: undefined | msgpack_ext_unpacker(),
>            original_list = []       :: msgpack_list_options()
>           }).
>
> Undoing this patch chunk makes the dialyzer report a slew of warnings, but
> not fail.
>
> Is it possible that someone can try doing the same on another setup to
> verify the presence/absence of the error? Also preferable on older releases
> to make a target for bisection.
>
>
> --
> J.
>



-- 
J.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-bugs/attachments/20140117/9b95a7dd/attachment.htm>