test erlang 21

[getong]

No description provided.

[getong]

@paulo-ferraz-oliveira @Vagabond
The stacktrace_compat seems to not work with otp 21. See the https://github.com/g-andrade/stacktrace_compat/blob/master/.travis.yml

#    - 21.0 FIXME wait for Travis to get OTP 21 -compatible rebar3

And in hex.pm, the poolboy has 856403 downloads, it needs a new tag and push to hex.pm.
And this pr uses OTP_RELEASE and test otp 21, and the ci build result shows this pr works.
I think this pr can be merged asap.
Feedback is welcome.

[fenollp]

Just wanted to say I like your stacktrace macros very much!
LGTM.

[getong]

@fenollp
It is copied from meck

[paulo-ferraz-oliveira]

@getong: it does work with OTP 21 :D I've already deployed it to live systems with no issues. It was created for OTP 21 compatibility.

On the other hand, even if you don't consider it a good alternative (some might) it was recently merged into elli: elli-lib/elli#61 (though I later figured out that it wasn't necessary, so removed it).

I also like your approach (though it involves macros and changing code+), I'm just saying that there's no better approach in this case.

+imagine you have a huge project with 20/30 try catch statements, while stacktrace_compat might use parse_transforms but it requires no change to existing code.

[getong]

@paulo-ferraz-oliveira
The stacktrace_compat is so complicated, and the code is hard to read, and I thnk it is hard to maintain.
And in the travis-ci test, it does not pass the otp 21 test.
It is designed to solve the otp 21 problem, but it does not fix up for the version. So confused.
Don't use it until it fix the version problem.

[g-andrade]

@getong

The stacktrace_compat seems to not work with otp 21.

It does work ;-). Test cases are passed and coverage is full. What's broken is the rebar3 version that's bundled with Travis, as mentioned in this issue. I've solved it by always downloading the latest rebar3 when running tests under Travis.

As for it being complex, well there's not much of a counterargument I can offer - parse transforms are complex almost by definition, as they manipulate code in the abstract format. But I wouldn't expect it to be particularly hard to maintain; there isn't anything else to be done on features at this point.

[getong]

@Vagabond
Would you spend some time merging this code?

[Vagabond]

Alright I merged this and fixed the dialyzer specs.

[Vagabond]

I tagged 1.5.2 but I can't push to hex. @devinus can you grant me access?