[parted-devel] [RFC][PATCH] Remove alloca from libparted/arch/linux.c

[Frodo Baggins]

On 9/6/07, Jim Meyering <jim at meyering.net> wrote:
> "Frodo Baggins" <frodo.drogo at gmail.com> wrote:
> > Is this safe/desirable?
>
> Thanks for working on the code.
>
> However, please don't mix two types of changes like this.
> A global-substitution change like s/((m|re)alloc|free)/ped_$1/, belongs
> in a change set all by itself.  BTW, I'm luke-warm on that change,
> but won't object outright.  I don't see much to gain by using the
> ped_ wrappers.  IMHO, if it's worth using the ped_-prefixed allocation
> functions everywhere, then it's also worth writing a "make distcheck"
> rule (add it to Makefile.maint) to enforce the policy that all uses of
> malloc/calloc/realloc be through the ped_-prefixed functions.
>
> Uh oh... I just looked at the definitions of ped_realloc and ped_malloc
> and see that they call ped_exception_throw upon error.  Ick.  Worse
> still, ped_realloc returns 0/1, so your s/realloc/ped_realloc/ change
> would actually cause some serious damage.  IMHO, ped_realloc should be
> eliminated altogether (there are only a handful of uses) or changed to
> have its signature match that of realloc.

I am looking into this.

>
> I think it would be worthwhile (long term) to convert the other way,
> e.g., ped_free -> free, ped_malloc -> malloc, etc.
>
> --------------------------
> For the other part, why do you want to remove uses of alloca?
> Is the struct too large for someone's stack?
> Is there some portability problem?  Removing alloca is often
> an improvement, but it's best to be able to justify introducing
> a new way to fail.
>

How about this? Added this in a local branch.