Re: [Tinycc-devel] RE :Re: VLA implementation

[Edmund Grimley Evans]

> On i386, that recipe should detect the problem on 79_vla_continue.
>
> Yes, I reproduced this result:
> +==19621==  Address 0xbe9e4f84 is just below the stack ptr.  To
> suppress, use: --workaround-gcc296-bugs=yes

I've seen that, too, but I don't fully understand what causes it.
Perhaps worth investigating.

I'd expect there also to be warnings about uninitialised values coming from
the code in 79_vla_continue.c. I think I do understand where those come from:
the contents of the VLA are invalidated by the temporary move of the
stack pointer.

> PS: It looks that ld.so from glibc-2.8 have a problems
> +==19621== Conditional jump or move depends on uninitialised value(s)
> +==19621==    at 0x400A550: ??? (in /lib/ld-2.8.so)

I've not seen that, but I've not tried Valgrind with such an ancient
version of glibc.
That error might be caused by some kind of bad input to ld.so. In my experience
the dynamic linker does not validate its input: if you give it a bad
binary it may just
crash. Probably not worth investigating unless it happens also with a
newer glibc.

Edmund