On Wed, May 17, 2006 at 12:25:18AM +0200, Michael Kerrisk wrote:
> Justin,
>
> > On Mon, May 15, 2006 at 10:06:29PM +0200, Michael Kerrisk wrote:
> > > > > > changes:
> > > > > > make mmap.2 reference mincore
> > > > >
> > > > > Done, for 2.33.
> > > > >
> > > > > > explicitly state that a read-only MAP_PRIVATE is equivalent to
> > > > > > MAP_SHARED
> > > > >
> > > > > *why?* You do not explain what benefit this serves.
> > > > Because, upon reflection, it isn't immediately clear from the wording
> > > > of the page as it exists now. In particular:
> > > >
> > > > Create a private copy-on-write mapping
> > >
> > > Umm -- we were talking about read-only mappings?
> > Yes; I wish to make it clear that, for a read-only map, the behaviour
> > is the same.
> >
> > > > This could be interpretted as meaning "create a new mapping, even if
> > > > the file is already mapped", which isn't what people usually want,
> > > > and isn't what it does anyway..
> > > >
> > > > > > An alternative would be to take the commented-out Linus quote
> > > > > > embedded
> > > > > > in mmap.2 which says precisely what I want and create a NOTES
> > > > > > section
> > > > > > out of it.
> > >
> > > Let me put things another way: why should a programmer care
> > > about whether read-only mappings are done the same way
> > > internally for both MAP_SHARED and MAP_PRIVATE?
> > A programmer should care whether using MAP_PRIVATE uses extra
> > resources than a MAP_SHARED (both maps presumably have some
> > overhead, for the initial mapping and for each additional mapping). I
> > question whether this is presently as clear as it should be.
>
> But:
>
> -- you have not demonstrated that there is any significant
> difference?
Do you mean differences between MAP_{PRIVATE,SHARED}? It is my
understanding that there is no (behavioural) difference for the case
of a read-only mapping. But I want this to be more obvious.
> -- whether there is a difference on other Unix systems
> (if they show different perfomance characteristics,
> then the man-pages should not make recommendations).
I haven't a clue; but I'm surprised that you think it shouldn't make a
recommendation, instead of documenting the differences.
> > I think it is typical to need to have *some* understanding of a lower
> > layer interface than one uses directly, even if the full
> > implementation details are deliberately opaque.
>
> I'm interested in specifics at this point.
Knowing that the effect of the two constants is the same for a
read-only case helps me to understand what the difference is for the
writable case.
Justin
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
