There may be a role for negative offsets. But the combination of a negative offset with the base position must always be non-negative.
Certainly in TIFF, the various structures can be in any order in the file. So one will frequently combine an offset value that comes from the infoset with the base to get a new position that is less than the current position. But I believe there's no need to use negative offsets. But that's because in TIFF it is expressed in terms of offsets from a base that is before ALL the objects in the file. One can imagine a situation where the offsets are "self relative". I.e., an element sets the new dfdl:offsetBase to it's own start position. Then if a data element is before the current one, a negative offset would be needed. I don't know that this is required to express any actual format. It's theoretically possible that a format might be described in this way. ________________________________ From: Steve Lawrence <slawre...@apache.org> Sent: Friday, November 17, 2017 8:34:30 AM To: dev@daffodil.apache.org; Mike Beckerle Subject: Re: TIFF format - base+offset proposed features On 11/17/2017 08:05 AM, Mike Beckerle wrote: > This is lower priority than many things, but a while back we tried to create > a DFDL schema for TIFF and failed. > > > DFDL features for base+offset location of elements never were designed. They > were dropped in the haste to get DFDL v1.0 finished. > > > But I happened to ask Dave Sugar of the Tresys CADRE team whether the CADRE > engine (a binary data format description - for wire-speed applications) could > do TIFF? Turns out some work was done on this, and I was able to look at the > format specification for TIFF (a prototype thereof for CADRE). > > > So the features for DFDL to suppport base + offset locations don't seem too > daunting. > > > I created a wiki page to capture the design ideas here. There's more work to > do on this. We really should write the DFDL schema for TIFF using these > features to be sure we're happy with it. Thus far I've not taken the time to > do that. I've just recorded the basic ideas. > > > https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=74687382 > > You specify a requirement that dfdl:offset be non-negative--could that be too restrictive? For example, I could imagine a (really dumb) file format where it essentially chains blobs together, with the position of the next blob relative to this one, e.g.: 0: Blob-1, Next: +5 1: 2: Blob-4, Next: +2 3: Blob-3, Next: -1 4: Blob-5, End 5: Blob-2, Next: -2 I'm not sure supporting a format like this would be possible without negative offsets. If no format has something like this I'm not sure it matters and restricting to non-negative is completely reasonable, but I thought you could do something like this with TIFF? Maybe not? - Steve
