Use lhttpc instead of httpc

[zp-sd]

• Aviod a crash in http_transport:close/2, happening when setting socket_opts in httpc in OTP 20.1.
  The fix for the bug is included in OTP 21 - https://bugs.erlang.org/browse/ERL-605.

09:55:22.062 [error] gen_server <0.1348.0> terminated with reason: no function clause matching http_transport:close(undefined, #Port<0.2267>) line 346
09:55:22.062 [error] CRASH REPORT Process <0.1348.0> with 0 neighbours crashed with reason: no function clause matching http_transport:close(undefined, #Port<0.2267>) line 346
09:55:22.062 [error] Supervisor httpc_handler_sup had child undefined started with {httpc_handler,start_link,undefined} at <0.1348.0> exit with reason no function clause matching http_transport:close(undefined, #Port<0.2267>) line 346 in context child_terminated

• Avoid crash report when httpc_handler_sup is killing its children after unsuccessful httpc:request/1 function call.
  It happens in some router softwares, with miniupnpd daemon installed, but UPnP/NAT-PMP functionality turned off.
  Then, during natupnp_v1/natupnp_v2 discovery, location is returned, but is it not responding (hence httpc:request/1 crash).

11:40:40.047 [error] gen_server <0.1176.0> terminated with reason: {tcp_error,#Port<0.246021>,etimedout}
11:40:40.069 [error] CRASH REPORT Process <0.1176.0> with 0 neighbours exited with reason: {tcp_error,#Port<0.246021>,etimedout} in gen_server:handle_common_reply/8 line 726
11:40:40.069 [error] Supervisor httpc_handler_sup had child undefined started with {httpc_handler,start_link,undefined} at <0.1176.0> exit with reason {tcp_error,#Port<0.246021>,etimedout} in context child_terminated

[benoitc]

@zp-sd as said on slack I would use hackney instead of lhttpc. I'm the maintainer of hackney so any updated needed would be easier. Can you make the change?

one question is if checking if setting the option to a profile using httpc:set_options/1 wouldn't do the trick so it wouldn't force the use to have a dependency to hackney and just rely on Erlang/OTP ?

[benoitc]

@zp-sd if you have a simple way to test we can eventually try the second method. Let me know :)

[lucafavatella]

Provided that the caller passes a profile not used by concurrent requests, would avoiding to place the httpc process in the supervisor solve the issue? Requires somehow determining profile to use in httpc call.

Refs:

• stand_alone.
• calling httpc with stand_alone pid.

Snippet:

httpc_request(Method, Request, HTTPOptions, Options, Profile) ->
    {ok, Pid} = inets:start(httpc, [{profile, Profile}], stand_alone),
    Response = httpc:request(Method, Request, HTTPOptions, Options, Pid),
    ok = gen_server:stop(Pid, normal, infinity),
    Response.

Past usages that may be referred to:

• aeternity/aeternity@81c07f46#diff-fd827a9d10583752b1e3ba0d9fa82fb2L377
• aeternity/aeternity@786e84b4

[jadeallenx]

Given that we're on OTP 22/23 these days, the PR seems moot?