Classic Mac OS: fix socklen_t, curl_off_t, long long

[ryandesign]

Change CURL_TYPEOF_CURL_SOCKLEN_T from int to unsigned int when __MWERKS__ is defined (i.e. Metrowerks compilers i.e. CodeWarrior).

This fixes a build failure when compiling for classic Mac OS with GUSI with CodeWarrior compilers.

---

The error was:

#----------------------------------------------------------
### MWC68K Compiler Error:
#	if(getsockname(sockfd, (struct sockaddr *) &add, &size) < 0) {
#	                                                      ^
# cannot convert
# 'int *' to
# 'unsigned int *'
#----------------------------------------------------------
File "connect.c"; Line 462
#----------------------------------------------------------
### MWC68K Compiler Error:
#	if(0 != getsockopt(sockfd, SOL_SOCKET, SO_ERROR, (void *)&err, &errSize))
#	                                                                       ^
# cannot convert
# 'int *' to
# 'unsigned int *'
#----------------------------------------------------------
File "connect.c"; Line 535
#----------------------------------------------------------

---

It's unclear to me what system was envisioned when the __MWERKS__ block was originally added in 2008 in 3ac6929. Metrowerks CodeWarrior was originally a classic Mac OS compiler in the 90s but, later, versions for BeOS, Windows, and Mac OS X existed, and it could cross-compile code for other systems as well. The Metrowerks company no longer exists today but CodeWarrior still does and is now used for embedded systems. I'm not certain that all of the systems CodeWarrior supports or supported will have the same values for the constants being defined here.

For example, POSIX says socklen_t should be unsigned.

• GUSI for classic Mac OS started using socklen_t in version 2. It is unsigned in the earliest version still posted on the site (2.1.3 from 2000) and in the last version (2.2.3 from 2002).
• Mac OS X first used socklen_t in version 10.3 (2003) and it was signed. By Mac OS X 10.4 (2005) it was unsigned.
• BeOS (1995-2001) apparently used a signed socklen_t at least for a time according to a quick Internet search. Don't know what Haiku (the reimplementation of BeOS) does.
• No idea what Windows or embedded systems do.

I'm not sure how these potential differences should be handled here.

[jay]

I'm not certain that all of the systems CodeWarrior supports or supported will have the same values for the constants being defined here.

Maybe we should stop using __MWERKS__ or add more qualifiers? I think we can assume that section was probably for old macs.

[dfandrich]

FWIW, Haiku defines socklen_t as uint32_t in sys/socket.h

[ryandesign]

Maybe we should stop using __MWERKS__ or add more qualifiers?

Perhaps so. Several of the other top-level conditions in that file are already about operating systems (e.g. _WIN32_WCE, __VMS, __OS400__), not compilers.

I think we can assume that section was probably for old macs.

Probably. I've rewritten the PR for this. I've divided it into a few hopefully sensibly distinct changes.

[jay]

Is SIZEOF_LONG defined for you? I don't see it in config-mac and there are a few places in the library where it's evaluated

lib/mprintf.c:#if (SIZEOF_CURL_OFF_T > SIZEOF_LONG)
lib/mprintf.c:#if (SIZEOF_SIZE_T > SIZEOF_LONG)
lib/mprintf.c:#if (SIZEOF_CURL_OFF_T > SIZEOF_LONG)

[ryandesign]

Good point! I've added a commit to define SIZEOF_LONG.

[bagder]

Thanks!