[Lldb-commits] [PATCH] D70277: [Signal] Allow llvm clients to opt into one-shot SIGPIPE handling

[Pavel Labath via Phabricator via lldb-commits]

labath added a comment.

As I said in the other discussion, though this is a (small) step in what I consider to be the right direction for this, it still seems like a workaround to me. But if everyone else is fine with that, then I won't stand in your way.

One thing I am not sure of is why you've chosen clang as the only recipient of the old SIGPIPE behavior. It seems like pretty much every non-interactive llvm utility would benefit from the previously-default SIGPIPE behavior. This is why I was talking about building this functionality into InitLLVM somehow -- one could make it so that InitLLVM installs the sigpipe handler by default, and lldb could pass some extra argument or something to disable that behavior.



================
Comment at: llvm/lib/Support/Unix/Signals.inc:328
     registerHandler(S, SignalKind::IsKill);
+  if (OneShotPipeSignalFunction)
+    registerHandler(SIGPIPE, SignalKind::IsKill);
----------------
This bit is tricky. It means that `SetOneShotPipeSignalFunction` will take effect only if it is called before we install any signal handlers. Calling it later will not cause the handler to be installed. However, if one does set a handler function before signals are installed, then it is possible to change it with `SetOneShotPipeSignalFunction`. This makes its behavior a lot more complex than `SetInfoSignalFunction`, which it tries to emulate.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D70277/new/

https://reviews.llvm.org/D70277