format hex packages in a nice/human way

[fenollp]

Changes

===> Verifying dependencies...
===> Upgrading lager ({pkg,<<"lager">>,<<"3.6.8">>})
===> Downloaded package, caching at /home/pete/.cache/rebar3/hex/hexpm/packages/lager-3.6.8.tar
===> No upgrade needed for amqp_client
===> No upgrade needed for jsx
===> Upgrading lager ({pkg,<<"lager">>,<<"3.6.9">>})
===> Version cached at /home/pete/.cache/rebar3/hex/hexpm/packages/lager-3.6.9.tar is up to date, reusing it

into

===> Verifying dependencies...
===> Upgrading lager v3.6.8
===> Version cached at /home/pete/.cache/rebar3/hex/hexpm/packages/lager-3.6.8.tar is up to date, reusing it
===> No upgrade needed for amqp_client
===> No upgrade needed for jsx
===> Upgrading lager v3.6.9
===> Version cached at /home/pete/.cache/rebar3/hex/hexpm/packages/lager-3.6.9.tar is up to date, reusing it

In short:

===> Fetching bbmustache ({pkg,<<"bbmustache">>,<<"1.6.0">>})
===> Fetching bbmustache v1.6.0

[fenollp]

(misclick)
Any help on making CI pass? The tests are a bit terse.

[ferd]

I believe that the tests for all the failing suites are parsing the warnings to make sure they are properly output. See for example:

• rebar3/test/rebar_install_deps_SUITE.erl

  Lines 460 to 490 in cc788f1

  	warning_calls() ->
  	History = meck:history(rebar_log),
  	[{Str, Args} || {_, {rebar_log, log, [warn, Str, Args]}, _} <- History].
  	error_calls() ->
  	History = meck:history(rebar_log),
  	[{Str, Args} || {_, {rebar_log, log, [error, Str, Args]}, _} <- History].
  	check_warnings(_, [], _) ->
  	ok;
  	check_warnings(Warns, [{Type, Name, Vsn} | Rest], mixed) ->
  	ct:pal("Checking for warning ~p in ~p", [{Name,Vsn},Warns]),
  	?assert(in_warnings(Type, Warns, Name, Vsn)),
  	check_warnings(Warns, Rest, mixed);
  	check_warnings(Warns, [{Name, Vsn} | Rest], Type) ->
  	ct:pal("Checking for warning ~p in ~p", [{Name,Vsn},Warns]),
  	?assert(in_warnings(Type, Warns, Name, Vsn)),
  	check_warnings(Warns, Rest, Type);
  	check_warnings(Warns, none, _Type) ->
  	ct:pal("Checking that there were no warnings", []),
  	?assert(Warns == []).
  	in_warnings(git, Warns, NameRaw, VsnRaw) ->
  	Name = iolist_to_binary(NameRaw),
  	1 =< length([1 || {_, [AppName, {git, _, {_, Vsn}}]} <- Warns,
  	AppName =:= Name, Vsn =:= VsnRaw]);
  	in_warnings(pkg, Warns, NameRaw, VsnRaw) ->
  	Name = iolist_to_binary(NameRaw),
  	Vsn = iolist_to_binary(VsnRaw),
  	1 =< length([1 || {_, [AppName, {pkg, _, AppVsn}]} <- Warns,
  	AppName =:= Name, AppVsn =:= Vsn]).

• rebar3/test/rebar_deps_SUITE.erl

  Lines 476 to 502 in cc788f1

  	run(Config) ->
  	{ok, RebarConfig} = file:consult(?config(rebarconfig, Config)),
  	rebar_test_utils:run_and_check(
  	Config, RebarConfig, ["install_deps"], ?config(expect, Config)
  	),
  	check_warnings(warning_calls(), ?config(warnings, Config), ?config(deps_type, Config)).
  	warning_calls() ->
  	History = meck:history(rebar_log),
  	[{Str, Args} || {_, {rebar_log, log, [warn, Str, Args]}, _} <- History].
  	check_warnings(_, [], _) ->
  	ok;
  	check_warnings(Warns, [{Name, Vsn} | Rest], Type) ->
  	ct:pal("Checking for warning ~p in ~p", [{Name,Vsn},Warns]),
  	?assert(in_warnings(Type, Warns, Name, Vsn)),
  	check_warnings(Warns, Rest, Type).
  	in_warnings(git, Warns, NameRaw, VsnRaw) ->
  	Name = iolist_to_binary(NameRaw),
  	1 =< length([1 || {_, [AppName, {git, _, {_, Vsn}}]} <- Warns,
  	AppName =:= Name, Vsn =:= VsnRaw]);
  	in_warnings(pkg, Warns, NameRaw, VsnRaw) ->
  	Name = iolist_to_binary(NameRaw),
  	Vsn = iolist_to_binary(VsnRaw),
  	1 =< length([1 || {_, [AppName, {pkg, _, AppVsn}]} <- Warns,
  	AppName =:= Name, AppVsn =:= Vsn]).

These are two suites that look for warnings (there may be more, but I think that's the largest bit).
By changing the output of the builds, you inadvertently broke the tests that relied on that format (as some external tools might have)

It's a question of updating the tests to care for the new warnings, and then possibly having a decent time where master contains the fix but without releasing to make sure anyone running tools find the problem early enough to not break their stuff. At least git resources will remain the same so that limits some potential breakage too.

[fenollp]

I can squash commits. Also maybe you'd prefer me find another way of making these tests pass?

[ferd]

The windows test appears to always be broken on CI, we might just want to remove it (CC @tsloughter ) -- I'll run this branch by hand instead on my windows computer

[tsloughter]

Ugh, yea, I guess we should.