[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: boolean issue with serf

From: Clark S. Cox III <clark.cox_at_apple.com>
Date: Thu, 16 Apr 2009 16:14:33 -0700

On Apr 16, 2009, at 3:07 PM, Michae Sinz wrote:

> On Thu, Apr 16, 2009 at 1:36 PM, Greg Stein <gstein_at_gmail.com> wrote:
> On Thu, Apr 16, 2009 at 21:46, Stefan Küng <tortoisesvn_at_gmail.com>
> wrote:
> > C. Michael Pilato wrote:
> [...]
> > Hmm - according to wikipedia:
> > http://en.wikipedia.org/wiki/Boolean_datatype
> >
> > "However, problems arise from the fact that any non-zero value
> > represents true in C, while the value TRUE is represented by a
> specific
> > value. So while in other languages, if (foo == TRUE) ... is merely
> > redundant, in C, it is actually incorrect code."
>
> svn_boolean_t is defined as one of {TRUE, FALSE}.
>
> Pass one of those values. Wikipedia does not define our svn_boolean_t.
>
> I agree that they don't define the svn_boolean_t but show me where
> the compiler will respect that or enforce that?
>
> Both at the user code path and in the actually evaluation of
> operations into a boolean.
>
> There are cases in code, depending on the compiler and optimizer,
> where in C code the compiler itself will not force the value to be 0
> or 1. In fact, a common optimization trick has been used to let
> whatever non-zero pass through or even use the non-zero to make -1
> (all bits on in whatever word size you expect). (Albeit these were
> highly optimizing embedded C compilers - it has been years since I
> have looked at C code generation and optimization paths and in other
> languages there is an actual boolean type that can only have two
> values and thus the behavior of the code is more strictly defined.)
>
> PS - on CPU architectures there are instructions that expand the CPU
> condition codes into a boolean value in a register such that no
> branching is needed at all to store the result of a boolean
> operation/test. In at least two of the cases these expand to -1 or
> 0 (all bits on or all bits off). I would expect that those
> platforms would end up having a natural boolean behavior of -1.
> This would make it impossible to correctly do:
>
> svn_boolean_t flag = (x > y);
>
> and expect flag to contain either TRUE or FALSE (it may contain
> another value).

This is not true. The C standard guarantees that the result of (x >
y), converted to any integer type (signed bitfields notwithstanding)
is always either exactly equal to 1 or exactly equal to 0. Any
compiler that fails to behave as such is broken.

>
> Thus the code would have to be written:
>
> svn_boolean_t flag = (x > y) ? TRUE : FALSE;
>
> which looks a bit redundant (and will be more code as the compiler
> will have to do a test-branch-store or test-failbranch-store-branch
> to store the correct value into flag or a register representing
> flag. With enough optimization smarts it may be able to figure out
> that you could do test-negate-store to get that to work out...
>
> --
> Michael Sinz Technology and Engineering Director/
> Consultant
> "Starting Startups" mailto:Michael.Sinz_at_sinz.org
> My place on the web http://www.sinz.org/Michael.Sinz

------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=1756735

  • application/pkcs7-signature attachment: smime.p7s
Received on 2009-04-17 01:15:12 CEST

This is an archived mail posted to the Subversion Dev mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.