Re: additional HP-UX 11.11 HP ANSI C triggered cleanup.

[Guy Harris <guy () alum mit edu>]

On Jul 21, 2004, at 2:16 PM, Rick Jones wrote:

> First was print-esp.c - it was warning in three places about an 
> integer being converted to a pointer with the return value of strsep.  
> There is no strsep in HP-UX, and it seems that interface.h deals with 
> that, but print-esp.c was not including interface.h.

Actually, the problem there is that "netdissect.h" wasn't doing the 
same conditional declaration of "strsep()" that "interface.h" was; I 
checked in a change to do that (and fix some other not-up-to-date stuff 
in "netdissect.h").

> This leaves one warning behind, that I've not fixed:
>
> 	cc -O -DHAVE_CONFIG_H  -I./../libpcap  -I/usr/local/include 
> -I/usr//include -I./missing  -D_U_="" -I. -I./../libpcap  
> -I/usr/local/include -I/usr//include -I./missing -c ./print-esp.c
> cc: "print-esp.c", line 482: warning 604: Pointers are not 
> assignment-compatible.
> cc: "print-esp.c", line 482: warning 563: Argument #3 is not the 
> correct type.
>
> which seems to be this call to EVP_CipherInit:
>
>         if (sa->evp) {
>                 memset(&ctx, 0, sizeof(ctx));
>                 if (EVP_CipherInit(&ctx, sa->evp, secret, NULL, 0) < 0)
>                         (*ndo->ndo_warning)(ndo, "espkey init failed");
>
>                 blocksz = EVP_CIPHER_CTX_block_size(&ctx);

That *might* be a change in the OpenSSL (or whatever library/library 
set that comes from) API between versions.

> The next warning was a type mismatch in print-radius.c:

I've checked that fix in.

> the next is print-rsvp.c, which is another char * versus u_char * 
> thing:
>
> $ diff print-rsvp.c ../tcpdump-2004.07.20/print-rsvp.c
> 1383c1383
> <             if (rsvp_obj_print(subtptr,(u_char *)"\n\t    ", 
> subtlen) == -1)
> ---
> >             if (rsvp_obj_print(subtptr,"\n\t    ", subtlen) == -1)
> 1403c1403
> <         if (rsvp_obj_print(tptr,(u_char *)"\n\t  ", tlen) == -1)
> ---
> >         if (rsvp_obj_print(tptr,"\n\t  ", tlen) == -1)

Actually, the problem there is that the second argument to 
"rsvp_obj_print()" should be a "const char *" - it's just a string used 
as a tag in the output, not a pointer to packet data.  I've checked in 
a change to do that.

> next is another char * versus u_char * in print-telnet.c:

That's another case where a string argument should be "const char *"...

> and then more of the same in tcpdump.c.

...and the same applies there.  I've checked in changes to change the 
first argument to "ascii_print_with_offset()", "ascii_print()", 
"hex_print_with_offset()", and "hex_print()" from "const u_char *" to 
"const char *".

>  That also had an issue with redeclarations of ndo_error and 
> ndo_warning with a different storage class specifier.  I wasn't quite 
> sure what to do about that:

Dissectors using the new "netdissect" stuff (currently only the ESP 
dissector; the intent is to make a library out of the dissection code) 
will call those routines through a pointer in a "struct 
netdissect_options"; those functions should be static.  I've checked in 
a change to do that.

> nor the empty declaration warning.

That's probably just a stray semicolon; the change I checked in also 
deleted that.

> util.c had another char vs u_char:

...which should also be fixed by my "hex_print()" change.

> so, after a make clean, a make -k gives this:

Cool.  Sun C and your Alpha C compiler would probably pick up other 
issues; I don't have access to them any more, though.

-
This is the tcpdump-workers list.
Visit https://lists.sandelman.ca/ to unsubscribe.