Date: Sat, 28 Apr 2018 11:39:02 +1000 (EST) From: Bruce Evans <brde@optusnet.com.au> To: John Baldwin <jhb@freebsd.org> Cc: arch@freebsd.org Subject: Re: LIBC_SCCS Message-ID: <20180428110152.Q4737@besplex.bde.org> In-Reply-To: <1711113.VelFtdTVS7@ralph.baldwin.cx> References: <1711113.VelFtdTVS7@ralph.baldwin.cx>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, 27 Apr 2018, John Baldwin wrote: > I suspect no one cares, but for whatever reason our current handling of the > LIBC_SCCS macro in some of our libraries annoys me. In theory it seems like > LIBC_SCCS's purpose is to control whether or not old SCCS IDs from Berkeley > are included in libc's sources when libc is built. (Similar to how macros > control the behavior of __FBSDID().) However, we use an odd construct in > the tree. First, we define LIBC_SCCS by default in the CFLAGS of various > libraries (libkvm, libutil, libthr, libc, etc.) which in theory would enable > the IDs, but then we explicitly wrap them in #if 0, e.g.: > > #if defined(LIBC_SCCS) && !defined(lint) > #if 0 > static char sccsid[] = "@(#)kvm_hp300.c 8.1 (Berkeley) 6/4/93"; > #endif > #endif /* LIBC_SCCS and not lint */ Most aren't actually wrapped with '#if 0'. E.g., in libc/*/*.c there are 839 files but only 47 of these have any '#if 0' at all. SO this can be fixed without much churn. I thought there is a problem with the above not actually compiling if LIBC_SCCS is defined, but WARNS is only 2 for libc and it takes WARNS >= 4 to give -Wwrite-strings. style(9) says to use '#if 0', but not as above. It says to put the #if 0 around everything, but only if there is not already a suitable ifdef like the LIBC_SCCS one. The above style is apparently from NetBSD. It is now used a lot on libedit, with further ugliness and bugs: - first there is an #if-#else clause with a NetBSD __RCSID() in the #else clause, so that the #if 0 selects between the sccsid and the NetBSD id - __RCSID() may be killed by defining NO__RCSID. It is a bug that the application identifier NO__RCSID affects and system header. sys/cdefs.h is an especially important system header and is one of the few that tries to avoid namespace pollution bugs like this. The default is to keep the NetBSD id. - after this idfef tangle, there is an unconditional __FBSDID() to give the FreeBSD id - __FBSDID() may also be killed by defining either lint or STRIP_FBSDID. sys/cdefs.h is actually careless about namespace pollution bugs like this. > I'd rather that we make LIBC_SCCS actually work by removing the #if 0 (and > perhaps the lint baggage) but then remove it from the default CFLAGS to > preserve the existing behavior by default. Does anyone else care if I do > this? I don't mind removing FreeBSD mistakes like the '#if 0'. Editing to add const poisoning is hopefully not often needed. Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20180428110152.Q4737>