[erlang-questions] code:clash/0 borking when running eunit tests with rebar

[Jeremy Raymond]

I've tested the patch in my environment and it fixes the problem,
code:clash/0 no longer crashes on the invalid file path. Also note that a
patch to rebar_eunit is to be pushed out soon as well to use full paths
rather than relative ones as erl_prim_loader doesn't seem to like them.

On Fri, Apr 16, 2010 at 1:33 PM, Tuncer Ayaz <tuncer.ayaz@REDACTED> wrote:

> On Fri, Apr 16, 2010 at 4:46 PM, Tuncer Ayaz <tuncer.ayaz@REDACTED>
> wrote:
> > On Fri, Apr 16, 2010 at 3:42 PM, Vlad Dumitrescu <vladdu55@REDACTED>
> wrote:
> >> On Fri, Apr 16, 2010 at 15:31, Jeremy Raymond <jeraymond@REDACTED>
> wrote:
> >>> I can reproduce this without couchbeam as well by just calling
> >>> code:clash() directly in my eunit test:
> >>>
> >>> *** context setup failed ***
> >>> ::error:function_clause
> >>>  in function code:filter/3
> >>>    called as filter(".beam",".eunit",error)
> >>>  in call from code:build/1
> >>>  in call from code:clash/0
> >>>  in call from cdb_data_tests:setup/0
> >>
> >> Hi,
> >>
> >> in code.erl:
> >>
> >> build([]) -> [];
> >> build([Dir|Tail]) ->
> >>    Files = filter(objfile_extension(), Dir,
> >>                   erl_prim_loader:list_dir(Dir)),
> >>    [decorate(Files, Dir) | build(Tail)].
> >>
> >> filter(_Ext, Dir, {error,_}) ->
> >>    io:format("** Bad path can't read ~s~n", [Dir]), [];
> >> filter(Ext, _, {ok,Files}) ->
> >>    filter2(Ext, length(Ext), Files).
> >>
> >> and in erl_prim_loader.erl
> >>
> >> -spec list_dir(string()) -> {'ok', [string()]} | 'error'.
> >> list_dir(Dir) ->
> >>    check_file_result(list_dir, Dir, request({list_dir,Dir})).
> >>
> >> Where you can see that erl_prim_loader returns 'error' and filter
> >> expects {ok, _} or {error, _}. It feels like Dialyzer should have
> >> caught that?
> >
> > Thanks, nice catch!
> >
> > I've introduced that in 49da83 while fixing another
> > code:clash/0 issue.
> > Working on a patch with extended tests atop 79194d.
>
> Fix posted to erlang-patches@REDACTED
>
> I think this hasn't been caught earlier as usually invalid
> paths are caught in add_path(). A possible failure case is when
> a file/directory which is part of the code path disappears.
>
> I've extended the tests to catch that.
>



-- 
Jeremy Raymond