> > It occurs to me that might lend itself a bit more to what you're looking
> > for here by avoiding the need for a new instance of the iter loop (I
> > assume there is some outstanding work that is affected by this?). Given
> > that this series was kind of a side quest for a band-aid performance fix
> > in the meantime, and it's not likely 6.13 material anyways, I think I'm
> > going to put it in a holding pattern and keep it in the back pocket in
> > favor of trying to move that alternate approach along, at least to where
> > I can post an RFC for discussion.
> >
> > If that doesn't work out or there proves some critical need for it in
> > the meantime, then I'll post v4 for an easy fix. I'll post a v2 of patch
> > 4 separately since that is an independent fix..
>
> I thought the bug robots were complaining about the performance hit,
> so at least this part should go in sooner than later.
>
Technically, yeah.. I've just been waffling over it because I'd rather
try to make more progress on that over trying too hard to polish up this
one. Christoph was grumbling a bit on factoring related to unrolling
these loops in the future or some such thing, so I'm not really sure
where he is on that.
Ok, maybe I'll do this.. I've already fixed the outstanding nits for v4,
so I might as well just post it. If there's anything critical that needs
fixing on review then obviously I'll address it, but otherwise I'll
prioritize working toward an RFC for the batching thing over pure
factoring/aesthetic changes, because that likely means much of this code
just goes away anyways.
Brian
> --D
>
> > Brian
> >
> > > > > + while ((ret = iomap_iter(&iter, ops)) > 0) {
> > > > > + const struct iomap *s = iomap_iter_srcmap(&iter);
> > > > > +
> > > > > + if (s->type == IOMAP_HOLE || s->type == IOMAP_UNWRITTEN) {
> > > > > + loff_t p = iomap_length(&iter);
> > > >
> > > > Also please stick to variable names that are readable and preferably
> > > > the same as in the surrounding code, e.g. s -> srcmap p -> pos.
> > > >
> > >
> > > Sure. I think I did this to avoid long lines, but I can change it.
> > > Thanks.
> > >
> > > Brian
> > >
> > >
> >
> >
>
