Re: [PATCH] index-pack usage of mmap() is unacceptably slower on many OSes other than Linux

From: Junio C Hamano <junkio@cox.net>
Date: Dec 20, 2006, 21:52 UTC
Message-ID: <7v1wmu8cbw.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <Pine.LNX.4.64.0612201243490.3576@woody.osdl.org>

Linus Torvalds <torvalds@osdl.org> writes:

Show 9 quoted lines

> On Wed, 20 Dec 2006, Junio C Hamano wrote:
>> 
>> I needed to fix the changes to git-compat-util.h a bit from the
>> version I sent earlier to make OpenBSD happy (sys/types.h there
>> did not expose u_int unless _BSD_SOURCE was set, and
>> netinet/in.h was duplicated by mistake).
>
> Please don't use "u_int" in the first place. It's an abomination of a 
> type. It should never be used.

I did not want to touch imported sources in compat/, but

Show 7 quoted lines

> The only user in git seems to have been copied from a source that is 
> insane. It does
>
> 	u_int words[NS_IN6ADDRSZ / NS_INT16SZ];
>
> which is just insane. It actually seems to want to use "uint16_t", which 
> at least would make sense, and be a type that has some _point_ to it.

the above argument makes 100% sense.  Will fix.

However, on sane platforms we do not even compile that file. The problem I observed was that <include/netinet/tcp.h> on OpenBSD uses u_int wants the source code that uses that header file to first include <sys/types.h> to get u_int, which in turn requires __BSD_VISIBLE to be in effect.

Unfortunately I think _BSD_SOURCE needs to stay for this reason.