bug-gnulib
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH] flexmember: update comment


From: Bruno Haible
Subject: Re: [PATCH] flexmember: update comment
Date: Sat, 25 May 2019 01:33:19 +0200
User-agent: KMail/5.1.3 (Linux/4.4.0-145-generic; KDE/5.18.0; x86_64; ; )

Hi Paul,

> > 1) What is the alignment problem if the array element type is 'char'?
> 
> Suppose we have a platform where the alignment of each basic type is 
> equal to its size, where sizeof (int) == 4 and sizeof (char) == 1, and 
> where we have 'struct s { int n; char d[]; };' and suppose we want to 
> allocate a 'struct s' with a 3-element flexible array member 'd'. Then 
> 'offsetof (struct s, d) + 3 * sizeof (char)' is 7, and when we call 
> 'malloc (7)' malloc is entitled to assume that memory is being allocated 
> for an array of 7 one-byte objects, so it can return an address that is 
> not a multiple of 4, which means that the pointer that malloc returns is 
> invalid for 'struct s *'. To fix the problem, we can use 'malloc 
> (FLEXSIZEOF (struct s, d, 3))' instead, as FLEXSIZEOF yields 8 on this 
> platform, and malloc (8) must return an address that is a multiple of 4.
> 
> > don't we need to apply a 'ceil'-like alignment to the malloc result?
> No, because if malloc is given an argument like 8 that is a multiple of 
> sizeof (int), it must return a pointer suitable for 'int *'.

Oh, I see. Thanks. I was assuming that malloc (N) always returns a
multiple of max_align_t.

> Perhaps I should update the comment to make this all clearer, though it 
> is hard to be clear and accurate and terse in this area.

You found the right words now :) - I would expect to see such
explanations in the .h file, not in the .m4 file.

Bruno




reply via email to

[Prev in Thread] Current Thread [Next in Thread]