[Swift-devel] remote file/directory stuff (bug 22)

[Mihael Hategan]

I thought we had this discussion before: swft at ... Subject "function vs.
mapper".

On Mon, 2007-02-26 at 21:14 +0000, Ben Clifford wrote:
> 
> On Mon, 26 Feb 2007, Mihael Hategan wrote:
> 
> > We may also want to do the same to the input, because some even more
> > twisted apps will not even accept that as a parameter.
> 
>  So:
> > (file k, file m, file n) myapp(file l) {
> >   app{
> >     l>"input.txt";
> >     myapp;
> >     k<"output.crd"
> >     m<"output.prd"
> >     n<"output.rtf"
> >   }
> > }
> 
> somewhat related to that is the way that the air align_warp program wants 
> its input files:
> 
>   $ ls Raw/ 
>   anatomy1.hdr anatomy2.hdr anatomy3.hdr anatomy4.hdr reference.hdr
>   anatomy1.img anatomy2.img anatomy3.img anatomy4.img reference.img
> 
> All paired up - but the app is told just:
> 
>   $ align_warp /path/to/anatomy1.img /otherpath/reference.img
> 
> with it being implicit that the corresponding hdr files exists at:
> 
>  "/path/to/anatomy1.img" - "img" + "hdr"
>  "/otherpath/reference.img" - "img" + "hdr"
> 
> In practice, I think this is OK in the short-medium term because as long 
> as the corresponding hdr and img files sit next to each other in the 
> submit side, they'll map through to execute side files that also sit next 
> to each other, because of the way that we map.
> 
> But in the longer term future, that is not necessarily going to be the 
> case - as source-side mapping becomes richer, to pull data from different 
> places that maybe don't have a single source filename (eg replicas, other 
> kinds of data store), the above invariant may become untrue. So there's at 
> least consideration needed there - either in terms of the invariants that 
> we assume are true but are never written down in a spec, or in terms of 
> how we make that stuff map properly on the execute side.
>