Disable backtrace on mac

[conanoc]

fix #840

backtrace module does not support mac. So, disable crash log if BACKTRACE_SUPPORTED is 0.

The reason why this patch fix #840 is not fully identified. On mac, SIGSEGV and SIGBUS signal happens after jsonrpc only when signal handlers are registered with sigaction().

[isghe]

@lvaccaro @gabridome @Sjors
ACK 56ef036

[Sjors]

Awesome, this works for me on macOS. I'm now able to connect to peers and fund channels (master crashes on launch once I have some channels open).

[arowser]

Running ok on 10.12.4, but if run with lldb, show
`
2018-03-23T11:19:17.428Z lightningd(22701): Trying to guess public addresses...
2018-03-23T11:19:17.459Z lightningd(22701): Address 192.168.2.211:9735 is not routable
2018-03-23T11:19:17.467Z lightningd(22701): Failed to connect 30 socket: No route to host
2018-03-23T11:19:17.469Z lightningd(22701): testing /Users/danielaof/project/lightning/lightningd/lightning_channeld
2018-03-23T11:19:17.477Z lightningd(22701): testing /Users/danielaof/project/lightning/lightningd/lightning_closingd
2018-03-23T11:19:17.485Z lightningd(22701): testing /Users/danielaof/project/lightning/lightningd/lightning_gossipd
2018-03-23T11:19:17.492Z lightningd(22701): testing /Users/danielaof/project/lightning/lightningd/lightning_hsmd
2018-03-23T11:19:17.499Z lightningd(22701): testing /Users/danielaof/project/lightning/lightningd/lightning_onchaind
2018-03-23T11:19:17.508Z lightningd(22701): testing /Users/danielaof/project/lightning/lightningd/lightning_openingd
2018-03-23T11:19:17.569Z lightningd(22701): Client: Received message 11 from client
2018-03-23T11:19:17.594Z lightningd(22701): Client: Received message 9 from client
2018-03-23T11:19:17.595Z lightning_gossipd(22715): pid 22715, msgfd 13
2018-03-23T11:19:17.595Z lightningd(22701): Loaded 0 channels from DB
Waiting for /Users/danielaof/project/corebin/bitcoin-cli: Interrupted system call
Process 22701 stopped

• thread Just want to say 'Thank You'. #1, queue = 'com.apple.main-thread', stop reason = signal SIGABRT
  frame #0: 0x00007fffc80e7d42 libsystem_kernel.dylib__pthread_kill + 10 libsystem_kernel.dylib__pthread_kill:
  -> 0x7fffc80e7d42 <+10>: jae 0x7fffc80e7d4c ; <+20>
  0x7fffc80e7d44 <+12>: movq %rax, %rdi
  0x7fffc80e7d47 <+15>: jmp 0x7fffc80e0caf ; cerror_nocancel
  0x7fffc80e7d4c <+20>: retq
  (lldb) bt
• thread Just want to say 'Thank You'. #1, queue = 'com.apple.main-thread', stop reason = signal SIGABRT
  • frame #0: 0x00007fffc80e7d42 libsystem_kernel.dylib__pthread_kill + 10 frame #1: 0x00007fffc81d55bf libsystem_pthread.dylibpthread_kill + 90
    frame fixed typos in pdf #2: 0x00007fffc804d420 libsystem_c.dylibabort + 129 frame #3: 0x0000000100015b97 lightningdfatal(fmt="Waiting for %s: %s") at log.c:593
    frame Missing OP_CSV in Figure 4 of the pdf #4: 0x0000000100001f1f lightningdwait_for_bitcoind(bitcoind=0x0000000100302000) at bitcoind.c:760 frame #5: 0x0000000100004340 lightningdsetup_topology(topo=0x00000001003027f0, timers=0x0000000100302270, poll_time=timerel @ 0x00007fff5fbff680, first_blocknum=101) at chaintopology.c:730
    frame Don't use u32 locktime internally. #6: 0x0000000100012f91 lightningdmain(argc=5, argv=0x00007fff5fbff778) at lightningd.c:372 frame #7: 0x00007fffc7fb9235 libdyld.dylibstart + 1
    frame Python implementation of onion routing spitball #8: 0x00007fffc7fb9235 libdyld.dylibstart + 1 (lldb)

`

[conanoc]

yeah. I saw the same issue with lldb. But, it's not related with this patch. You could file a new issue for that.

[arowser]

Yes, it seems not related, and the issue have been addressed in #1093

[rustyrussell]

Ack a599399

ignoring dirty files in libwally-core is ugly, probably need a bug report there to improve their .gitignore?

[laanwj]

The reason why this patch fix #840 is not fully identified. On mac, SIGSEGV and SIGBUS signal happens after jsonrpc only when signal handlers are registered with sigaction().

What happens (see #840 (comment)) is that autodata (the library used for constructing RPC and other tables) uses SIGSEGV/SIGBUS on MacOSX and other BSDs to probe memory maps (in a subprocess). It registers its own handler for this, but apparently backtrace's handler also still triggers. This results in a logged error.

So this fix, removing backtrace, is overkill but it works. Another option would have been to disable backtrace handling just for autodata's subprocesses.