labath added inline comments.
================
Comment at: lit/tools/lldb-mi/target/inputs/target-select-so-path.py:8-11
+def get_free_port():
+ s = socket.socket()
+ s.bind(('', 0))
+ return s.getsockname()[1]
----------------
apolyakov wrote:
> labath wrote:
> > This is still racy, because the port can be snatched from under you between
> > the time you get the free port and the time when lldb-server binds to it.
> > If this was the only test doing it then it might be fine, but since this is
> > going to be running concurrently with other tests, all of which are
> > fetching free ports, the chances of that happening add up.
> >
> > (Also, binding to the wildcard address will trigger a firewall popup on
> > some machines.)
> There is a problem with getting port from lldb-server. If we run `lldb-server
> gdbserver --pipe 0 ocalhost:0`, it'll print port number to its stdout, but we
> can't get it using pipes since to do this we need to wait until lldb-server
> finishes that isn't what we want.
Aha, I see. lldb-server does not really expect you to pass std file descriptor
as the --pipe argument. Normally you would create a separate fd and pass that
instead. Doing that from python is a bit icky, but doable:
```
(r, w) = os.pipe()
kwargs = {}
if sys.version_info >= (3,0):
kwargs["pass_fds"] = [w]
llgs = subprocess.Popen(["lldb-server", "gdbserver", "--pipe", str(w), ...],
**kwargs)
port_bytes = os.read(r, 10)
port = int(port_bytes.decode("utf-8").strip('\x00'))
```
Alternatively, we could modify lldb-server to print the port number to stdout
in addition to any --pipe arguments (sounds like a good addition anyway, as it
enables easy free port selection for a shell user), and then you can sniff that
text from here.
https://reviews.llvm.org/D49739
_______________________________________________
lldb-commits mailing list
lldb-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
