Re: [PATCH] Add support for uintmax_t type on FreeBSD 4.9

From: Junio C Hamano <gitster@pobox.com>
Date: Oct 27, 2008, 06:17 UTC
Message-ID: <7v1vy2imt2.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <9a0027270810262246i56cf5515l5fa0875f91d90a7a@mail.gmail.com>

"David Syzdek" <syzdek@gmail.com> writes:

Show 24 quoted lines

> On Sun, Oct 26, 2008 at 9:30 PM, Junio C Hamano <gitster@pobox.com> wrote:
> ...
>> I have a stupid question.
>>
>> Would it be a more appropriate improvement to do it like this:
>>
>>        ifdef USE_THIS_AS_UINTMAX_T
>>            BASIC_CFLAGS += -Duintmax_t="$(USE_THIS_AS_UINTMAX_T)"
>>        endif
>>
>> and then add a section for FreeBSD 4.9-SECURITY like this:
>>
>>        ifeq ($(uname_R),4.9-SECURITY)
>>                USE_THIS_AS_UINTMAX_T = uint32_t
>>        endif
>>
>> That way, an oddball 64-bit machine can use uint64_t here if it wants to,
>> possibly including FreeBSD 4.9-SECURITY backported to 64-bit ;-).
>>
>
> Your suggestion provides more flexibility for other environments. I
> was making the assumption that 64-bit systems would define uintmax_t,
> however in retrospect that would be unwise.
> Would you like me to resubmit the patches with your modifications?

Actually there was a reason why I said this was a "stupid" question. I think your assumption on 64-bit platforms would hold in practice, and my suggestion could be an unnecessary overengineering. If nobody knows of a system that would benefit from such a generalization, your original patch would be better, partly because I think:

 (1) USE_THIS_AS_UINTMAX_T is just for demonstration of concept and is a
     terrible name we cannot possibly use in our Makefile.  We have to
     spend brain cycles to come up with a better name; and

 (2) It may be tricky to come up with autoconf macros to determine what to
     set USE_THIS_AS_UINTMAX_T to.

As a slightly unrelated aside, I find it somewhat unfortunate that the conditional says "4.9-SECURITY", which is a bit too explicit and specific. to my taste. I do not know how FreeBSD versioning scheme works, but wouldn't your change work equally well for 4.9-RELEASE or 4.11-RELEASE?

I suspect that you would want to say "$(uname_R) that begins with '4.' or smaller needs this workaround", as strtoul(3) manual page seems to appear first in FreeBSD 5.0-RELEASE (but not found in FreeBSD 4.11-RELEASE).