On Fri, July 18, 2008 9:11 pm, Carl Lowenstein wrote:
> On Fri, Jul 18, 2008 at 8:44 PM, James G. Sack (jim) <[EMAIL PROTECTED]>
> wrote:
>> Carl Lowenstein wrote:
>>>
>>> If I may allude to an old discussion, isn't this an argument for
>>> embedding people-readable tags in object files?
>>
>> I only kinda-vaguely recall that discussion. What was the context and
>> conclusion -- do you remember?
>>
>> In any case, I can think of something else to observe: embedded labels
>> can not really be used to verify files -- you need a real hash
>> authentication mechanism to do that, eh? But that wasn't the question
>> here, I don't think ..so.. nevermind. :-)
>
> It started with my observation that gcc now optimized away strings
> that were never used.  Said strings were to be embedded in object
> files as a record of RCS ID or equivalent.
>
> On the kplug-lpsg list with subject: "gcc optimizes out program ID
> string" and much rancor.
>
> The accepted answer is that the previous behavior of gcc has been
> deprecated for some time, and there is now
> __attribute__((unused))
> to deal with cases like this.
>
>     carl

<big grin>


Me recalls an embedded C program years ago that looped over a variable in
dual channel memory to see if the HW had set a value. We marked it
volatile, but nonetheless M$ C 7.XX looked at it and said "I didn't change
that, so it couldn't be changed" and took the loop out with hilarious
results.

When my boss called M$ support (after finally finding someone who knew
what he was talking about), their answer was that "'volatile' is
implemented syntactically but not semantically," by which he meant the
compiler wouldn't error on the token, but didn't act on it either.

Micro$oft -- where quality is job 0.

-- 
Lan Barnes

SCM Analyst              Linux Guy
Tcl/Tk Enthusiast        Biodiesel Brewer


-- 
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list

Reply via email to