[Duplicity-talk] thoughts about tar replacement

[Mark D. Anderson]

Just read your thoughts about replacing tar wrt duplicity.

First off, it reminds me a bit of the ELF format, which has
the requirement of fast random access to portions, and manages
extensible metadata.

But I also wondered why not keep the tar format, and implement
a sort of ranlib-on-steroids?
That way all traditional tools would continue to understand
the tar file at the format level (even if the tar members
may appear as garbage).

All the smarts would be in the special first member, just like
ranlib (which uses the special name '__.SYMDEF').
That member would have all the directory info:
- byte offset to each child
- encoding information about the child
- child's full name
- child's other attributes (ACL, etc.)
The ranlib member would also have global settings too (things
applying to all children).

This ranlib member could be in xml, for buzzword compliance.
Or it could use some sort of MIME.

One problem with ranlib is that it is variable length at the start
of the file, so that even small adjustments require significant temp space.
So one trick might be to have a fixed length special first member,
that has the offset to the special last member.
That approach would make in-place replacement of members more efficient.

-mda