v4k-git-backup/tools/labs/fwk_diff.md

3.4 KiB

given a string A, we want it to be B.

A: hello world and thanks for the fish. B: hello cruel o_o world and thanks for the fish!

however, the instructions to reconstruct B must be as small as possible, to minimize transmission costs. this is why we dont transmit B entirely.

different algorithms as follow:

ALGORITHM 1

  • identify the first mismatching character (S).

       v-- S@6
    

A: hello world and thanks for the fish. B: hello cruel o_o world and thanks for the fish!

  • identify the last mismatch character (E).

                                              v-- E@-0
    

A: hello world and thanks for the fish. B: hello cruel o_o world and thanks for the fish!

  • we construct the patch now with 3 numbers:
    • number of bytes to copy from A[0] till A[S] (not included)
    • number of bytes to copy from B[S] till B[E] (not included) + plus the substring that is get copied.
    • number of bytes to copy from A[E] till end of A string.

6 40 "cruel o_o world and thanks for the fish!" 0

  • total patch size is 3 control bytes [6,40,0] + 40 (string) = 43 bytes

ALGORITHM 2

  • We delta every character in both strings, from head to tail, and from tail to head.

A: hello world and thanks for the fish.0000000000 B: hello cruel l_o world and thanks for the fish! C: 000000XXXXX0XXX0XXXXXXXXXXXXXXXXXXXXXXXXXXXXXX (-)

A: 0000000000hello world and thanks for the fish. B: hello cruel l_o world and thanks for the fish! C: XXXXXXXXXXXX0X0000000000000000000000000000000X (-)

  • Pick the choice with most zeros (2nd choice) (aka, select the choice with less Xs).
  • On C lane, promote gaps of one or two 0s into Xs.

A: 0000000000hello world and thanks for the fish.\0 B: hello cruel l_o world and thanks for the fish!\0 C: XXXXXXXXXXXXXX0000000000000000000000000000000X 0 (-) ^

  • Do the number of leading Xs(C) - number of leading 0s(A) (14-10). If positive, write it [4]. This number will be needed in the patch function.
  • Do { Encode every XXXX island in C as a positive number indicating how many bytes to copy from. Otherwise, run-length the number of zeros into a negative number. } repeat till lane is exhausted.

4 XXXXXXXXXXXXXXX: +14 "hello cruel l_" 0000000000000000000000000000000: -31 X: +1 "!"

  • Patch size is 4 control bytes [4,+14,-31,+1] + 14 (string) + 1 (string) = 19 bytes

ALGORITHM 3

  • start source (A) and target (B) strings. start with A:
  • select how many bytes to copy (positive), how many to skip (negative), or switch A<-->B lanes (zero).
  • repeat previous step over and over until both lanes do reach their respective string ends.

A: hello world and thanks for the fish. B: hello cruel o_o world and thanks for the fish!

A: +6 0 >> hello B: -6 +10 0 >> hello cruel o_o A: +24 -1 0 >> hello cruel o_o world and thanks for the fish B: -29 +1 >> hello cruel o_o world and thanks for the fish! (10 bytes)

what if we always specify a dual +/- operation per lane? lanes changed automatically after every tuple. in this case a 0 shall indicate a no op.

A: +6 0 >> hello B: -6 +10 >> hello cruel o_o A: +24 -1 >> hello cruel o_o world and thanks for the fish B: -29 +1 >> hello cruel o_o world and thanks for the fish! (8 bytes)

  • we construct the patch now. a substring excerpt must come after any positive operations on lane B.

+6 0 -6 +10 "cruel o_o " +24 -1 -29 +1 "!"

  • total patch size is 8 control bytes + 10 (string) + 1 (string) = 19 bytes