Syslinux - Sep 2015 - EFI: HP + syslinux = crash [ brown paper bag update ]

On Sun, Sep 13, 2015 at 9:22 AM, Patrick Masotta <masottaus at yahoo.com>
wrote:
>
>>>>
>  > It is also working here; I think this particular PNG issue is fixed.
>  >
>  > It's worth to mention that the replaced code looks valid to me;
that's why I think
>  > this is probably some sort of compiler bug. It was really hard to
find.
>  > We should keep an eye on this issue; it might appear somewhere else.
>
>  Eww.  The original code was merely an empty
>  define.  That's pretty nasty.
>
>  Alternatively, we could void them
>
>  -#  define Assert(cond,msg)
>  +#  define Assert(cond,msg) ((void)0)
>
>  --
>  -Gene
> <<<
>
>
> The empty define should work; actually it did work for many years...
>
> Your just proposed change was my first attempt for solving this but for
some
> reason it did not work. After testing a lot the only way I've found to
solve this
> bug was defining a "real" empty function. I know, 100% hacky but
it works.
If the voiding doesn't work, we'll need that function in another 44
lines of code, spread throughout the source.

-- 
-Gene

>>>
 >
 > The empty define should work; actually it did work for many years...
 >
 > Your just proposed change was my first attempt for solving this but for
some
 > reason it did not work. After testing a lot the only way I've found to
solve this
 > bug was defining a "real" empty function. I know, 100% hacky but
it works.
 
 If the voiding doesn't work, we'll need that function in another 44
 lines of code, spread throughout the source.
 
 -- 
 -Gene
<<<
 
You might be right; I have mentioned before that this bug could probably impact
other parts of the code.
So far I'm continually using vesamenu.c32 with PNG images w/o any problem.
But the Zlib code
is surely used in many more places!

I really do not know why the compiler is not correctly parsing the void function
or if
there is a particular condition that triggers this behavior in this particular
case.
This issue is very hard to trace/reproduce/debug/report, etc. a real PITA.

Best,
Patrick

On Sun, Sep 13, 2015 at 10:14 AM, Patrick Masotta <masottaus at yahoo.com>
wrote:
>
>>>>
>  >
>  > The empty define should work; actually it did work for many years...
>  >
>  > Your just proposed change was my first attempt for solving this but
for some
>  > reason it did not work. After testing a lot the only way I've
found to solve this
>  > bug was defining a "real" empty function. I know, 100%
hacky but it works.
>
>  If the voiding doesn't work, we'll need that function in another
44
>  lines of code, spread throughout the source.
>
>  --
>  -Gene
> <<<
>
> You might be right; I have mentioned before that this bug could probably
impact other parts of the code.
> So far I'm continually using vesamenu.c32 with PNG images w/o any
problem. But the Zlib code
> is surely used in many more places!
>
> I really do not know why the compiler is not correctly parsing the void
function or if
> there is a particular condition that triggers this behavior in this
particular case.
> This issue is very hard to trace/reproduce/debug/report, etc. a real PITA.
Smells of an optimizer/stripper bug :(

Patrick, Oscar, what specific build tools are you two using?

GNU ld (GNU Binutils for Debian) 2.22
gcc version 4.7.2 (Debian 4.7.2-5)
NASM version 2.10.01

Trying to recall if there's other utility packages.

-- 
-Gene