Re: [Valgrind-users] Is this a gcc (or glibc) bug or a false positive?

[Nicholas N. <nj...@cs...>]

On Tue, 17 Jun 2008, Alan Jenkins wrote:

> The invalid read is caused by strlen() is reading past the end of a
> malloc()ed string.  For efficiency it is reading the string 4 bytes at a
> time.  valgrind reports this as an error.
>
> I believe this is acceptable *if the reads are guaranteed to be
> aligned*.  In this case they are, because the string is a variable
> length array at the end of a C structure, and the previous field is a
> pointer.  The string will start on an aligned address, so the read which
> includes the terminator will also be aligned.  There's no chance of it
> crossing a page boundary and provoking a segfault.

Valgrind tries to replace strlen with its own, less-optimised version, for 
exactly this reason.  Sometimes it doesn't work, eg. if your libc.so is 
stripped (IIRC).

As for "acceptable"... there's "acceptable according to the C standard" and 
"acceptable because it works on pretty much all modern machines", and those 
two aren't quite the same.  Valgrind/Memcheck tries to find a happy medium 
between those two, and --partial-loads-ok=yes is one example where you get 
to choose what behaviour you want.  It's surprising that 
--partial-loads-ok=yes isn't working, though.

Nick