casey@lll-crg.llnl.gov (Casey Leedom) (01/05/89)
I haven't seen this mentioned before, so I'll post it. The problem occurred on a VAX 785 running 4.3BSD + all BSD patches + 1988 April networking code. I've checked the stock 4.3BSD source code and you'd run into the problem there too. Essentially fd.h and <sys/types.h> both define the typedef fd_set. My patch uses a test of the defined status of NFDBITS to determine whether or not fd_set has been typedef'd. Unfortunately C doesn't provide a mechanism to test for the existence of typedef's. I don't like my fix, but I think it should work on most systems. Fd_set and NFDBITS came into existence at the same time. Casey ----- *** fd.h-dist Tue Sep 6 13:29:39 1988 --- fd.h Wed Jan 4 23:57:44 1989 *************** *** 42,49 **** --- 42,51 ---- #define FD_SETSIZE 256 #endif + #ifndef NFDBITS typedef long fd_mask; #define NFDBITS (sizeof(fd_mask) * NBBY) /* bits per mask */ + #endif #ifndef howmany #define howmany(x, y) (((x)+((y)-1))/(y)) #endif
swick@ATHENA.MIT.EDU (Ralph R. Swick) (01/05/89)
> Essentially fd.h and <sys/types.h> both define the typedef fd_set.
This was fixed in X11R3 by renaming the type (in lib/Xt/fd.h) used
internally by Xt.casey@GAUSS.LLNL.GOV (01/05/89)
| From: Ralph R. Swick <swick@ATHENA.MIT.EDU> | | > Essentially fd.h and <sys/types.h> both define the typedef fd_set. | | This was fixed in X11R3 by renaming the type (in lib/Xt/fd.h) used | internally by Xt. I assume that you're talking about the type "Fd_set" which would have conflicted with the type "fd_set" defined in <sys/types.h>. However, that still leaves the problem of the type "fd_mask". This isn't a problem on any 4.2BSD based machine because NFDBITS, fs_set, FD_SET, FD_CLR, etc. were all brought out with 4.3BSD. "fd_set" was brought out in a later release of 4.1BSD (I believe release c). If you like I can send you the specific versions of types.h for each of the above systems. Also, I should point out, my fix really isn't as good as it should be since the definitions of FD_SET, FD_CRL, FD)ISSET, and FD_ZERO should also be similarly ifdefed. I didn't see that till now when I sat down and looked closely at all the different versions of types.h ... Casey