Handling of "shortcut" values

[bradh]

I'm working on updating the uncompressed codec implementation to reflect the FDIS and to correct some bugs / limitations.

One aspect to that is working with shortcut values. The concept here is that instead of having a cmpd box with the various components, and an uncC box that references it, there are a few cases where you can just put the profile value in, and skip the rest. That is, the uncC box just says "version == 1, profile == rgb3", and there is nothing else. The meaning of rgb3 is what you would expect - 8 bit RGB, with pixel interleave and no subsampling. There are a couple of other cases: rgba and argb.

I can see at least four ways to deal with that:

• when hitting one of those cases, the parser generates the extra boxes
• just keeping the profile and version value, and special casing all the getters.
• filling in the extra box values based on the profile template, and changing the codec to handle the special cases (e.g. when doing the cmpd box lookup).
• subclassing uncompressed codec box

I liked the first and fourth options, but that is hard to do with the current parsing logic, which is basically to instantiate the box and tell it to fill in its values.
I think I like the third option better than the second option.

This is broader than just the uncompressed codec though. There is an MPEG work item to define a "smaller header", which basically packs the various item properties into a mini box, which is a combination of a tighter packing of small values, and some assumptions about images - see "Exploration on low-overhead HEIF-compatible image file format.

Would appreciate thoughts on this.

[farindk]

I prefer the first option. But I would describe it in a bit more abstract way: The class HeifFile presents the file boxes always as extended boxes, even if the file uses the minimized version. This also applies when we write a file. Boxes are always set like in the full file format, but if the content allows it, HeifFile may optionally save this in the minimized format.

[bradh]

I prefer the first option. But I would describe it in a bit more abstract way: The class HeifFile presents the file boxes always as extended boxes, even if the file uses the minimized version. This also applies when we write a file. Boxes are always set like in the full file format, but if the content allows it, HeifFile may optionally save this in the minimized format.

So if you ask for HeifFile::get_properties for example, where one of the properties is an uncC box using version==1 (as written in the underlying file), you'd get back properties including a synthesized cmpd box and a uncC box in full (version == 0) form?

Similarly, if you ask for HeifFile::get_[infe,iref,ipco,ipma]_box(), where the file uses the proposed mini box, you would get an appropriately synthesized version of those boxes?

Do you see that being implemented as conditional logic in HeifFile::parse_heif_file()? I can basically see how it would work for the named boxes, but it could get a bit messy for the general properties.