Re: add dd support for O_NOATIME

[Jim Meyering]

Paul Eggert <address@hidden> wrote:

> Jim Meyering <address@hidden> writes:
>
>> I'm tempted to say O_NOATIME should be the default for rm -r, but
>> maybe not for cp, find, du, ls, etc.
>
> Yes, that makes sense.
>
> How about if we suggest instead that laptop owners mount their file
> systems with the noatime option?  That solves the problem in general,
> rather than forcing app writers to modify cat, ls, sh, etc.

It'd be ok if there were a noatime mount option that applied only to
directories.  Personally, I rarely find directory atimes useful,
but file atimes *are* sometimes useful.  I want to have my cake and
eat it, too.

>> CVS glibc does provide fdopendir now, but it's not yet commonly available.
>
> Can we modify gnulib fdopendir to muck with glibc's guts, if it knows
> it's using glibc?  That is, gnulib could contain a copy of glibc's
> internal function __alloc_dir, as well as of its internal data
> structure DIR.  This should be safe, since we know all the older glibc
> versions that we'd be using this code in, so we know their internal
> structure.
>
> Normally I wouldn't suggest this sort of thing, but if it's a major
> performance issue....  (On second thought maybe I shouldn't have
> suggested it anyway.  :-)

I noticed __alloc_dir and was tempted, too :-)
but figured it wasn't worth the effort/risk.
Of course, if you want to pursue it, I won't try
to dissuade you.

>> So, in some cases we'd have to call open twice for
>> each directory.
>
> If we have a --noatime option, it could fail for files you don't own.
> Then we wouldn't have to open twice.  That's how dd iflags=noatime
> works, as well as tar --atime-preserve=system.
>
> That noatime mount option is looking better and better....

I have to admit I agree.