Re: [Nano-devel] valgrind complaining about uninitialized memory

[Benno Schulenberg]

Op 27-02-19 om 13:57 schreef Enrico Mioso:
> I was checking the other valgrind complain: winio.c, this line:
> if (on_a_vt && ioctl(0, TIOCLINUX, &modifiers) >= 0) {
> 
> was wondering what was wrong here.

Me too.  I have no idea.

> Am I wrong in thinking that &modifiers is an
> address on the function stack?

You're correct.  The &modifiers is an address pointing somewhere in
the function's stack.

> why does valgrind complain about us passing uninitialized memory to the ioctl 
> call?

Don't know.

> Was wondering if we could avoid these valgrind errors, to make it easier to 
> spot
> "real" issues in the future.

Maybe there is some option to valgrind that improves the check.

For example, when I run valgrind, I use --expensive-definedness-checks=yes
to avoid several occurrences of this:

==32641== Conditional jump or move depends on uninitialised value(s)
==32641==    at 0x538613D: re_compile_fastmap_iter.isra.26 (regcomp.c:328)
==32641== [...]

But that option doesn't help with the ioctl() issue you found.
(Apparently I've never run valgrind on a Linux console before.)


Hmmm...  The warning goes away when I make 'modifiers' a static variable
at the level of src/winio.c.  So apparently the warning is spurious and
there is nothing actually wrong with the call.

Benno

signature.asc
Description: OpenPGP digital signature