Re: ftw/nftw is limited by stack size

[Petr Vandrovec]

On 17 Jan 03 at 17:56, Jim Meyering wrote:
> Ulrich Drepper <address@hidden> wrote:
> > Jim Meyering wrote:
> >
> >> What do you think about rewriting it to use space on the heap
> >> rather than on the stack?
> >
> > It makes the general case slower for the few lunatics with such
> > filesystem hierachies.  That's absolutely not desirable.
> 
> Hi Uli,
> 
> Are you sure it'd make a significant difference in performance?
> I want a portable version of ftw that is immune to such abuse,
> and am willing to incur a reasonable performance penalty, so I'll
> probably implement it and make du use the more robust version.
> 
> > Instead you could work around the problem (for some platforms) by using
> > not the normal stack, but instead some allocated memory.
> >
> > use mmap() to allocate many many megabytes of memory (maybe even you the
> > automatic-grow feature Linux provides, although this might also be
> > limited by the stack size).  This won't actually require any physical
> > memory.  Then use makecontext and setcontext to use this memory.
> 
> Yes, but should every application that uses ftw -- and that wants
> to avoid that limitation -- have to go to such lengths?

Most of apps expect that max. path length is PATH_MAX, and fail misserably
when faced longer path - for example getcwd() & co. returns truncated path
silently on some kernels... And as you cannot pass paths longer than 
PATH_MAX to open/stat anyway...
                                            Petr Vandrovec
                                            address@hidden