Reply to request for error correction ideas for new format

[Stuart Gathman]

The web page describing the tar replacement format asks for ideas to improve error recovery. Here is an important one that I learned from designing a file system that has no critical blocks (can survive corruption or loss anywhere with limited data loss). Each file (including meta files such as index data) should be assigned an "id" within the backup. The block header for each block should include the file id, and perhaps an offset within the file. This way, even with loss of meta data, blocks belonging to the same file can be identified, and files still recovered in a "lost+found" directory. The pointers in the index are not sufficient, because you often don't know how many bytes have been "skipped" when recovering from errors. (And you may not have the index.)