> File Interoperability / User Defined Data Representations ---
> would need to be changed.
Specifically, references to etypes would become filetypes.
Presumably we would have to allow the implementation to
compress away holes when the data was copied, since
filetypes are even more likely than etypes to have big
holes in them. Furthermore, I think we would have to
allow an implementation to read a partial filetype for
the purpose of conversion. A filetype could be huge,
and if the implementation had to read the whole thing at
once into a filebuf, it could easily run out of memory
(since the user has already allocated a buffer large
enough to hold the converted data). If we allowed
this change, then the count and position would be in
units of basic types, rather than etypes or filetypes.
Of course, the implementation would then have to cut
up the filetype to describe each segment that it
passed to the filter, unless we add a second "position"
counter that specified where in the filetype the
converter should start accessing the filebuf.
John