| Message ID | 20141009193327.B7BD02C3ACF@topped-with-meat.com |
|---|---|
| State | New |
| Headers | show |
On 10/09/2014 09:33 PM, Roland McGrath wrote: >> Uhm, the current lv package in Fedora and Debian uses it: > > Indeed. Surprising. The use of SV_INTERRUPT means it would actually > change the behavior to simply start failing that configure check and use > plain signal instead of sigvec. (signal by default is equivalent to > sigaction using SA_RESTART; SV_INTERRUPT is the inverse of SA_RESTART.) > > It's of course a simple 5-minute job to convert any old use of sigvec to > use sigaction instead. I did it for lv and the patch is below. Would you > like to take care of sending that to the lv maintainer and/or using it in > the Fedora package? I've submitted your patch to Fedora: <https://bugzilla.redhat.com/show_bug.cgi?id=1151982> > Do you think we should do a staged deprecation for sigvec instead? Probably not, especially since we aren't actually removing sigvec as such (the subject got me worried for a moment).
> I've submitted your patch to Fedora: > > <https://bugzilla.redhat.com/show_bug.cgi?id=1151982> Thanks. > > Do you think we should do a staged deprecation for sigvec instead? > > Probably not, especially since we aren't actually removing sigvec as > such (the subject got me worried for a moment). We are indeed removing sigvec from the API. Of course we are not breaking ABI compatibility; that goes without saying. Thanks, Roland
On 10/13/2014 10:24 PM, Roland McGrath wrote: >> I've submitted your patch to Fedora: >> >> <https://bugzilla.redhat.com/show_bug.cgi?id=1151982> > > Thanks. Also reported to Debian <https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=769546> >>> Do you think we should do a staged deprecation for sigvec instead? >> >> Probably not, especially since we aren't actually removing sigvec as >> such (the subject got me worried for a moment). > > We are indeed removing sigvec from the API. > Of course we are not breaking ABI compatibility; that goes without saying. For the record, I think this change can go ahead.
Committed. check-abi passes and there are no changes to generated code except for the static sigvec.o going away. Thanks, Roland
Hi, make install fails due: LANGUAGE=C LC_ALL=C makeinfo -P build-dir/manual/ --output=build-dir/manual/libc.info libc.texinfo src-dir/manual//signal.texi:2142: Cross reference to nonexistent node `BSD Handler' (perhaps incorrect sectioning?). makeinfo: Removing output file `build-dir/manual/libc.info' due to errors; use --force to preserve. Bye Stefan On 11/14/2014 08:07 PM, Roland McGrath wrote: > Committed. check-abi passes and there are no changes to generated code > except for the static sigvec.o going away. > > Thanks, > Roland >
--- ./src/configure.in.~1~ 2004-01-04 22:35:44.000000000 -0800 +++ ./src/configure.in 2014-10-09 11:14:47.782210631 -0700 @@ -34,7 +34,7 @@ AC_CHECK_HEADERS(fcntl.h sys/ioctl.h sys dnl Checks for typedefs, structures, and compiler characteristics. dnl Checks for library functions. -AC_CHECK_FUNCS(sigvec tgetnum setlocale) +AC_CHECK_FUNCS(sigaction tgetnum setlocale) AC_FUNC_GETPGRP AC_PROG_GCC_TRADITIONAL AC_TYPE_SIGNAL --- ./src/console.c.~1~ 2004-01-04 23:27:46.000000000 -0800 +++ ./src/console.c 2014-10-09 11:16:59.627943378 -0700 @@ -158,9 +158,9 @@ private RETSIGTYPE InterruptHandler( int { kb_interrupted = TRUE; -#ifndef HAVE_SIGVEC +#ifndef HAVE_SIGACTION signal( SIGINT, InterruptHandler ); -#endif /* HAVE_SIGVEC */ +#endif /* HAVE_SIGACTION */ } public void ConsoleEnableInterrupt() @@ -235,9 +235,9 @@ private RETSIGTYPE WindowChangeHandler( ConsoleGetWindowSize(); -#ifndef HAVE_SIGVEC +#ifndef HAVE_SIGACTION signal( SIGWINCH, WindowChangeHandler ); -#endif /* HAVE_SIGVEC */ +#endif /* HAVE_SIGACTION */ } #endif /* UNIX */ @@ -388,24 +388,24 @@ public void ConsoleSetUp() signal( SIGINT, InterruptIgnoreHandler ); #endif /* MSDOS */ -#ifdef HAVE_SIGVEC - struct sigvec sigVec; +#ifdef HAVE_SIGACTION + struct sigaction sa; - sigVec.sv_handler = WindowChangeHandler; - sigVec.sv_mask = sigmask( SIGWINCH ); - sigVec.sv_flags = SV_INTERRUPT; - sigvec( SIGWINCH, &sigVec, NULL ); - - sigVec.sv_handler = InterruptHandler; - sigVec.sv_mask = sigmask( SIGINT ); - sigVec.sv_flags = SV_INTERRUPT; - sigvec( SIGINT, &sigVec, NULL ); + sa.sa_handler = WindowChangeHandler; + sigemptyset( &sa.sa_mask ); + sa.sa_flags = 0; /* No SA_RESTART means interrupt syscalls. */ + sigaction( SIGWINCH, &sa, NULL ); + + sa.sa_handler = InterruptHandler; + sigemptyset( &sa.sa_mask ); + sa.sa_flags = 0; /* No SA_RESTART means interrupt syscalls. */ + sigaction( SIGINT, &sa, NULL ); #else # ifdef SIGWINCH signal( SIGWINCH, WindowChangeHandler ); # endif signal( SIGINT, InterruptHandler ); -#endif /* HAVE_SIGVEC */ +#endif /* HAVE_SIGACTION */ #ifdef UNIX #ifdef HAVE_TERMIOS_H
> Uhm, the current lv package in Fedora and Debian uses it: Indeed. Surprising. The use of SV_INTERRUPT means it would actually change the behavior to simply start failing that configure check and use plain signal instead of sigvec. (signal by default is equivalent to sigaction using SA_RESTART; SV_INTERRUPT is the inverse of SA_RESTART.) It's of course a simple 5-minute job to convert any old use of sigvec to use sigaction instead. I did it for lv and the patch is below. Would you like to take care of sending that to the lv maintainer and/or using it in the Fedora package? This is a case where the plain removal of sigvec would not have broken the package's build and would instead have subtly changed the behavior of the program. So it certainly seems like a red flag. But I'm not sure that a staged deprecation plan would do any better on that score. There would be compile-time warnings about using a deprecated interface, but that would not have broken the package's build either. Perhaps distros will (should? do?) build with -Werror=deprecated-declarations specifically to ensure such things get caught. Do you think we should do a staged deprecation for sigvec instead? Thanks, Roland