On 1/28/20 4:26 PM, Dave Chinner wrote: > On Tue, Jan 28, 2020 at 08:48:40AM -0600, Eric Sandeen wrote: >> On 1/27/20 9:29 PM, Darrick J. Wong wrote: >>> On Mon, Jan 27, 2020 at 04:56:02PM -0600, Eric Sandeen wrote: >>>> From: Eric Sandeen <sandeen@xxxxxxxxxx> >>>> >>>> In each of these cases, db, logprint, and mdrestore are redeclaring >>>> as a global variable something which was already provided by a >>>> library they link with. >>> >>> Er... which library? >> >> libxfs and libxlog ... >> >> >> File Line >> 0 libxlog/util.c 10 int print_exit; >> 1 logprint/logprint.c 27 int print_exit = 1; >> >> File Line >> 0 db/init.c 30 libxfs_init_t x; >> 1 libxlog/util.c 13 libxfs_init_t x; >> >> File Line >> 0 fsr/xfs_fsr.c 31 char *progname; >> 1 io/init.c 14 char *progname; >> 2 libxfs/init.c 28 char *progname = "libxfs"; >> 3 mdrestore/xfs_mdrestore.c 10 char *progname; >> >> (fsr & io don't link w/ libxfs; mdrestore does) >> >> >>> >>> Also, uh...maybe we shouldn't be exporting globals across libraries? >>> >>> (He says having not looked for how many there are lurki... ye gods) >> >> Well, it's ugly for sure. >> >> We could either try to re-architect this to >> >> 1) pass stuff like progname all over the place, or >> 2) consistently make the library provide it as a global, or >> 3) consistently make utils provide it to the library as a global (?) > > IIRC, I chose #2 way back when I was consolidating all the libxfs > library code. There was code that declared libxfs_init_t x; on > stack, as local globals, in the libraries, etc. So I simply made the > library global the One True Global, and then had everyone use it > everywhere. > > That was just the simplest solution at the time to minimise the > amount of work to get userspace up to date with the kernel to allow > integration of the CRC work (userspace was years out of date at that > point). It was not pretty (like a lot of my code), but it worked. > > Feel free to do what you think is best :) I think the patch I sent is best - not quite sure how the stray globals snuck in as problems (or maybe got missed the first time) but here we are. gcc 10 complains about them btw, I should have mentioned that. All I need now is a review. :D -Eric
