From: Andrew Burgess Date: Wed, 29 Dec 2021 11:53:49 +0000 (+0000) Subject: gdb: don't pass nullptr to sigwait X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=926ac872e924e6618441f76917d58088d92dd11d;p=binutils-gdb.git gdb: don't pass nullptr to sigwait I tried building GDB on GNU/Hurd, and ran into this warning: gdbsupport/scoped_ignore_signal.h:78:16: error: null argument where non-null required (argument 2) [-Werror=nonnull] This is because in this commit: commit 99624310dd82542c389c89c2e55d8cae36bb74e1 Date: Sun Jun 27 15:13:14 2021 -0400 gdb: fall back on sigpending + sigwait if sigtimedwait is not available A call to sigwait was introduced that passes nullptr as the second argument, this call is only reached if sigtimedwait is not supported. The original patch was written for macOS, I assume on that target passing nullptr as the second argument is fine. On my GNU/Linux box, the man-page for sigwait doesn't mention that nullptr is allowed for the second argument, so my assumption would be that nullptr is not OK, and, if I change the '#ifdef HAVE_SIGTIMEDWAIT' introduced by the above patch to '#if 0', and rebuild on GNU/Linux, I see the same warning that I see on GNU/Hurd. I propose that we stop passing nullptr as the second argument to sigwait, and instead pass a valid int pointer. The value returned in the int can then be used in an assert. For testing, I (locally) made the change to the #ifdef I mentioned above, compiled GDB, and ran the usual tests, this meant I was using sigwait instead on sigtimedwait on GNU/Linux, I saw no regressions. --- diff --git a/gdbsupport/scoped_ignore_signal.h b/gdbsupport/scoped_ignore_signal.h index eff240fcbad..cb46812224c 100644 --- a/gdbsupport/scoped_ignore_signal.h +++ b/gdbsupport/scoped_ignore_signal.h @@ -75,7 +75,12 @@ public: sigpending (&pending); if (sigismember (&pending, Sig)) - sigwait (&set, nullptr); + { + int sig_found; + + sigwait (&set, &sig_found); + gdb_assert (sig_found == Sig); + } #endif }