Legend:
Page
Library
Module
Module type
Parameter
Class
Class type
Source
Page
Library
Module
Module type
Parameter
Class
Class type
Source
patience_diff.ml
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105
open! Core include Patience_diff_intf module Hunk = Hunk module Hunks = Hunks module Matching_block = Matching_block module Range = Range module Move_id = Move_id let ( <|> ) ar (i, j) = if j <= i then [||] else Array.slice ar i j (* Does the nitty gritty of turning indexes into line numbers and reversing the ranges, returning a nice new hunk *) let create_hunk prev_start prev_stop next_start next_stop ranges : _ Hunk.t = { prev_start = prev_start + 1 ; prev_size = prev_stop - prev_start ; next_start = next_start + 1 ; next_size = next_stop - next_start ; ranges = List.rev ranges } ;; module Ordered_sequence : sig type elt = int * int [@@deriving compare] (* A [t] has its second coordinates in increasing order *) type t = private elt array [@@deriving sexp_of] val create : (int * int) list -> t val is_empty : t -> bool end = struct type elt = int * int [@@deriving sexp_of] let compare_elt a b = Comparable.lexicographic [ (fun (_, y0) (_, y1) -> Int.compare y0 y1) ; (fun (x0, _) (x1, _) -> Int.compare x0 x1) ] a b ;; type t = elt array [@@deriving sexp_of] let create l = let t = Array.of_list l in Array.sort t ~compare:compare_elt; t ;; let is_empty = Array.is_empty end (* This is an implementation of the patience sorting algorithm as explained at http://en.wikipedia.org/wiki/Patience_sorting *) module Patience : sig val longest_increasing_subsequence : Ordered_sequence.t -> (int * int) list end = struct module Pile = struct type 'a t = 'a Stack.t let create x = let t = Stack.create () in Stack.push t x; t ;; let top t = Stack.top t |> Option.value_exn let put_on_top t x = Stack.push t x end module Piles = struct type 'a t = 'a Pile.t Deque.t let empty () : 'a t = Deque.create ~never_shrink:true () let get_ith_pile t i dir = let get index offset = Option.bind (index t) ~f:(fun index -> Deque.get_opt t (index + offset)) in match dir with | `From_left -> get Deque.front_index i | `From_right -> get Deque.back_index (-i) ;; let new_rightmost_pile t pile = Deque.enqueue_back t pile end module Backpointers = struct (* in the terminology of the Wikipedia article, this corresponds to a card together with its backpointers *) type 'a tag = 'a t and 'a t = { value : 'a ; tag : 'a tag option } let to_list t = let rec to_list acc t = match t.tag with | None -> t.value :: acc | Some t' -> to_list (t.value :: acc) t' in to_list [] t ;; end module Play_patience : sig val play_patience : Ordered_sequence.t -> get_tag: (pile_opt:int option -> piles:Ordered_sequence.elt Backpointers.t Piles.t -> Ordered_sequence.elt Backpointers.tag option) -> Ordered_sequence.elt Backpointers.t Piles.t end = struct let optimized_findi_from_left piles x = (* first see if any work *) let last_pile = Piles.get_ith_pile piles 0 `From_right in (* [x_pile] is a dummy pile just used for comparisons *) let x_pile = Pile.create { Backpointers.value = x, 0; tag = None } in let compare_top_values pile1 pile2 = let top pile = fst (Pile.top pile).Backpointers.value in Int.compare (top pile1) (top pile2) in let%bind.Option last_pile = last_pile in if compare_top_values last_pile x_pile < 0 then None else (* do binary search *) Deque.binary_search piles `First_strictly_greater_than x_pile ~compare:compare_top_values ;; (* [play_patience ar ~get_tag] plays patience with the greedy algorithm as described in the Wikipedia article, taking [ar] to be the deck of cards. It returns the resulting [Piles.t]. Before putting an element of [ar] in a pile, it tags it using [get_tag]. [get_tag] takes as its arguments the full [Piles.t] in its current state, and also the specific [Pile.t] that the element of [ar] is being added to. *) let play_patience ar ~get_tag = let ar = (ar : Ordered_sequence.t :> Ordered_sequence.elt array) in if Array.length ar = 0 then raise (Invalid_argument "Patience_diff.play_patience"); let piles = Piles.empty () in Array.iter ar ~f:(fun x -> let pile_opt = optimized_findi_from_left piles (fst x) in let tagged_x = { Backpointers.value = x; tag = get_tag ~pile_opt ~piles } in match pile_opt with | None -> Piles.new_rightmost_pile piles (Pile.create tagged_x) | Some i -> let pile = Deque.get piles i in Pile.put_on_top pile tagged_x); piles ;; end let longest_increasing_subsequence ar = if Ordered_sequence.is_empty ar then [] else let module P = Play_patience in let get_tag ~pile_opt ~piles = match pile_opt with | None -> Piles.get_ith_pile piles 0 `From_right |> Option.map ~f:Pile.top | Some i -> if i = 0 then None else Piles.get_ith_pile piles (i - 1) `From_left |> Option.value_exn |> Pile.top |> Option.some in let piles = P.play_patience ar ~get_tag in Piles.get_ith_pile piles 0 `From_right |> Option.value_exn |> Pile.top |> Backpointers.to_list ;; end let compare_int_pair = Tuple.T2.compare ~cmp1:Int.compare ~cmp2:Int.compare let _longest_increasing_subsequence ar = let ar = (ar : Ordered_sequence.t :> (int * int) array) in let len = Array.length ar in if len <= 1 then Array.to_list ar else ( let maxlen = ref 0 in let m = Array.create ~len:(len + 1) (-1) in let pred = Array.create ~len:(len + 1) (-1) in for i = 0 to len - 1 do let p = Array.binary_search ~compare:Ordered_sequence.compare_elt ar `First_greater_than_or_equal_to ar.(i) ~len:(max (!maxlen - 1) 0) ~pos:1 |> Option.value ~default:0 in pred.(i) <- m.(p); if p = !maxlen || compare_int_pair ar.(i) ar.(p + 1) < 0 then ( m.(p + 1) <- i; if p + 1 > !maxlen then maxlen := p + 1) done; let rec loop ac p = if p = -1 then ac else loop (ar.(p) :: ac) pred.(p) in loop [] m.(!maxlen)) ;; (* Configurable parameters for [semantic_cleanup] and [unique_lcs], all chosen based on empirical observation. *) (* This function is called on the edge case of semantic cleanup, when there's a change that's exactly the same length as the size of the match. If the insert on the next side is a LOT larger than the match, it should be semantically cleaned up, but most of the time it should be left alone. *) let should_discard_if_other_side_equal ~big_enough = 100 / big_enough (* These are the numerator and denominator of the cutoff for aborting the patience diff algorithm in [unique_lcs]. (This will result in us using [Plain_diff] instead.) Lowering [switch_to_plain_diff_numerator] / [switch_to_plain_diff_denominator] makes us switch to plain diff less often. The range of this cutoff is from 0 to 1, where 0 means we always switch and 1 means we never switch. *) let switch_to_plain_diff_numerator = 1 let switch_to_plain_diff_denominator = 10 module Make (Elt : Hashtbl.Key) = struct module Table = Hashtbl.Make (Elt) type elt = Elt.t (* This is an implementation of the patience diff algorithm by Bram Cohen as seen in Bazaar version 1.14.1 *) module Line_metadata = struct type t = | Unique_in_a of { index_in_a : int } | Unique_in_a_b of { index_in_a : int ; index_in_b : int } | Not_unique of { occurrences_in_a : int } end let unique_lcs (alpha, alo, ahi) (bravo, blo, bhi) = (* Create a hashtable which takes elements of a to their index in a iff they're unique. If an element is not unique, it takes it to its frequency in a. *) let unique : (elt, Line_metadata.t) Table.hashtbl = Table.create ~size:(Int.min (ahi - alo) (bhi - blo)) () in for x's_pos_in_a = alo to ahi - 1 do let x = alpha.(x's_pos_in_a) in match Hashtbl.find unique x with | None -> Hashtbl.set unique ~key:x ~data:(Unique_in_a { index_in_a = x's_pos_in_a }) | Some (Unique_in_a _) -> Hashtbl.set unique ~key:x ~data:(Not_unique { occurrences_in_a = 2 }) | Some (Not_unique { occurrences_in_a = n }) -> Hashtbl.set unique ~key:x ~data:(Not_unique { occurrences_in_a = n + 1 }) (* This case doesn't occur until the second pass through [unique] *) | Some (Unique_in_a_b _) -> assert false done; (* [num_pairs] is the size of the list we use for Longest Increasing Subsequence. [intersection_size] is the number of tokens in the intersection of the two sequences, with multiplicity, and is an upper bound on the size of the LCS. *) let num_pairs = ref 0 in let intersection_size = ref 0 in for x's_pos_in_b = blo to bhi - 1 do let x = bravo.(x's_pos_in_b) in Hashtbl.find unique x |> Option.iter ~f:(fun pos -> match pos with | Not_unique { occurrences_in_a = n } -> if n > 0 then ( Hashtbl.set unique ~key:x ~data:(Not_unique { occurrences_in_a = n - 1 }); incr intersection_size) | Unique_in_a { index_in_a = x's_pos_in_a } -> incr num_pairs; incr intersection_size; Hashtbl.set unique ~key:x ~data: (Unique_in_a_b { index_in_a = x's_pos_in_a; index_in_b = x's_pos_in_b }) | Unique_in_a_b _ -> decr num_pairs; Hashtbl.set unique ~key:x ~data:(Not_unique { occurrences_in_a = 0 })) done; (* If we're ignoring almost all of the text when we perform the patience diff algorithm, it will often give bad results. *) if !num_pairs * switch_to_plain_diff_denominator < !intersection_size * switch_to_plain_diff_numerator then `Not_enough_unique_tokens else ( let a_b = let unique = Hashtbl.filter_map unique ~f:(function | Not_unique _ | Unique_in_a _ -> None | Unique_in_a_b { index_in_a = i_a; index_in_b = i_b } -> Some (i_a, i_b)) in Ordered_sequence.create (Hashtbl.data unique) in `Computed_lcs (Patience.longest_increasing_subsequence a_b)) ;; (* [matches a b] returns a list of pairs (i,j) such that a.(i) = b.(j) and such that the list is strictly increasing in both its first and second coordinates. This is done by first applying unique_lcs to find matches from a to b among those elements which are unique in both a and b, and then recursively applying [matches] to each subinterval determined by those matches. The uniqueness requirement is waived for blocks of matching lines at the beginning or end. I couldn't figure out how to do this efficiently in a functional way, so this is pretty much a straight translation of the original Python code. *) let matches alpha bravo = let matches_ref_length = ref 0 in let matches_ref = ref [] in let add_match m = incr matches_ref_length; matches_ref := m :: !matches_ref in let rec recurse_matches alo blo ahi bhi = (* printf "alo %d blo %d ahi %d bhi %d\n%!" alo blo ahi bhi; *) let old_length = !matches_ref_length in if not (alo >= ahi || blo >= bhi) then if Elt.compare alpha.(alo) bravo.(blo) = 0 then ( let alo = ref alo in let blo = ref blo in while !alo < ahi && !blo < bhi && Elt.compare alpha.(!alo) bravo.(!blo) = 0 do add_match (!alo, !blo); incr alo; incr blo done; recurse_matches !alo !blo ahi bhi) else if Elt.compare alpha.(ahi - 1) bravo.(bhi - 1) = 0 then ( let nahi = ref (ahi - 1) in let nbhi = ref (bhi - 1) in while !nahi > alo && !nbhi > blo && Elt.compare alpha.(!nahi - 1) bravo.(!nbhi - 1) = 0 do decr nahi; decr nbhi done; recurse_matches alo blo !nahi !nbhi; for i = 0 to ahi - !nahi - 1 do add_match (!nahi + i, !nbhi + i) done) else ( let last_a_pos = ref (alo - 1) in let last_b_pos = ref (blo - 1) in let plain_diff () = Plain_diff.iter_matches ~hashable:(module Elt) (Array.sub alpha ~pos:alo ~len:(ahi - alo)) (Array.sub bravo ~pos:blo ~len:(bhi - blo)) ~f:(fun (i1, i2) -> add_match (alo + i1, blo + i2)) in match unique_lcs (alpha, alo, ahi) (bravo, blo, bhi) with | `Not_enough_unique_tokens -> plain_diff () | `Computed_lcs lcs -> lcs |> List.iter ~f:(fun (apos, bpos) -> if !last_a_pos + 1 <> apos || !last_b_pos + 1 <> bpos then recurse_matches (!last_a_pos + 1) (!last_b_pos + 1) apos bpos; last_a_pos := apos; last_b_pos := bpos; add_match (apos, bpos)); if !matches_ref_length > old_length (* Did unique_lcs find anything at all? *) then recurse_matches (!last_a_pos + 1) (!last_b_pos + 1) ahi bhi else plain_diff ()) in recurse_matches 0 0 (Array.length alpha) (Array.length bravo); List.rev !matches_ref ;; let collapse_sequences matches = let collapsed = ref [] in let start_a = ref None in let start_b = ref None in let length = ref 0 in List.iter matches ~f:(fun (i_a, i_b) -> match !start_a, !start_b with | Some start_a_val, Some start_b_val when i_a = start_a_val + !length && i_b = start_b_val + !length -> incr length | _ -> (match !start_a, !start_b with | Some start_a_val, Some start_b_val -> let matching_block = { Matching_block.prev_start = start_a_val ; next_start = start_b_val ; length = !length } in collapsed := matching_block :: !collapsed | _ -> ()); start_a := Some i_a; start_b := Some i_b; length := 1); (match !start_a, !start_b with | Some start_a_val, Some start_b_val when !length <> 0 -> let matching_block = { Matching_block.prev_start = start_a_val ; next_start = start_b_val ; length = !length } in collapsed := matching_block :: !collapsed | _ -> ()); List.rev !collapsed ;; (* Given that there's an insert/delete of size [left_change] to the left, and an insert/delete of size [right_change] to the right, should we keep this block of length [block_len] in our list of matches, or discard it? *) let should_discard_match ~big_enough ~left_change ~right_change ~block_len = (* Throw away if its effective length is too small, relative to its surrounding inserts / deletes. *) block_len < big_enough && ((left_change > block_len && right_change > block_len) || (left_change >= block_len + should_discard_if_other_side_equal ~big_enough && right_change = block_len) || (right_change >= block_len + should_discard_if_other_side_equal ~big_enough && left_change = block_len)) ;; let change_between (left_matching_block : Matching_block.t) (right_matching_block : Matching_block.t) = max (right_matching_block.prev_start - left_matching_block.prev_start) (right_matching_block.next_start - left_matching_block.next_start) - left_matching_block.length ;; (* See the "Semantic Chaff" section of https://neil.fraser.name/writing/diff/ *) let basic_semantic_cleanup ~big_enough matching_blocks = if big_enough <= 1 then matching_blocks else ( match matching_blocks with | [] -> [] | first_block :: other_blocks -> let final_ans, final_pending = List.fold other_blocks ~init:([], first_block) ~f:(fun (ans, pending) current_block -> let rec loop ans pending = match ans with | [] -> ans, pending | hd :: tl -> if should_discard_match ~big_enough ~left_change:(change_between hd pending) ~right_change:(change_between pending current_block) ~block_len:pending.length then loop tl hd else ans, pending in let updated_ans, updated_pending = loop ans pending in updated_pending :: updated_ans, current_block) in List.rev (final_pending :: final_ans)) ;; (* Attempts to eliminate the "tunnel vision" problem described in the "Semantic Chaff" section of https://neil.fraser.name/writing/diff/. To do this, we go through each pair of consecutive matches and pretend to combine them into one match. If that match would be deleted by [basic_semantic_cleanup], we delete both. *) let advanced_semantic_cleanup ~big_enough matching_blocks = if big_enough <= 1 then matching_blocks else ( match matching_blocks with | [] -> [] | [ first_block ] -> [ first_block ] | first_block :: second_block :: other_blocks -> let final_ans, final_pendingA, final_pendingB = List.fold other_blocks ~init:([], first_block, second_block) ~f:(fun (ans, pendingA, pendingB) current_block -> let rec loop ans pendingA pendingB = match ans with | [] -> ans, pendingA, pendingB | hd :: tl -> if should_discard_match ~big_enough ~left_change:(change_between hd pendingA) ~right_change:(change_between pendingB current_block) ~block_len: (pendingB.length + min (pendingB.prev_start - pendingA.prev_start) (pendingB.next_start - pendingA.next_start)) then loop tl hd pendingA else ans, pendingA, pendingB in let updated_ans, updated_pendingA, updated_pendingB = loop ans pendingA pendingB in updated_pendingA :: updated_ans, updated_pendingB, current_block) in List.rev (final_pendingB :: final_pendingA :: final_ans) (* The loop above only deleted the second element of each pair we're supposed to delete. This call to [basic_semantic_cleanup] is guaranteed to finish the job by deleting the remaining element of those pairs. *) |> basic_semantic_cleanup ~big_enough) ;; (* Goal: eliminate small, semantically meaningless matches. *) let semantic_cleanup ~big_enough matching_blocks = basic_semantic_cleanup ~big_enough matching_blocks |> advanced_semantic_cleanup ~big_enough ;; (* When we have a choice, we'd prefer one block of equality to two. For example, instead of A <insert>B A</insert> C D E F, we prefer <insert>A B</insert> A C D E F. There are two reasons: (1) A is usually something like "let", and so the second version is more semantically accurate (2) Semantic cleanup may delete the lone A match, but it will not delete the A C D E F match). So by moving the A match, we've also saved it. *) let combine_equalities ~prev ~next ~matches = match matches with | [] -> [] | first_block :: tl -> List.fold tl ~init:([], first_block) ~f:(fun (ans, pending) block -> let rec loop ans ~(pending : Matching_block.t) ~(new_block : Matching_block.t) = if pending.length = 0 then ans, pending, new_block else ( let advance_in_prev = Elt.compare prev.(pending.prev_start + pending.length - 1) prev.(new_block.prev_start - 1) = 0 in let advance_in_next = Elt.compare next.(pending.next_start + pending.length - 1) next.(new_block.next_start - 1) = 0 in if advance_in_prev && advance_in_next then loop ans ~pending: { prev_start = pending.prev_start ; next_start = pending.next_start ; length = pending.length - 1 } ~new_block: { prev_start = new_block.prev_start - 1 ; next_start = new_block.next_start - 1 ; length = new_block.length + 1 } else ans, pending, new_block) in let updated_ans, updated_pending, updated_new_block = loop ans ~pending ~new_block:block in (* In the original Google heuristic, we would either move all or none of pending. But because it might start with an unmatched `Newline(0, None), we are fine with moving all but one token of it. *) if updated_pending.length = 0 || updated_pending.length = 1 then ( let new_ans = if updated_pending.length = 0 then updated_ans else updated_pending :: updated_ans in new_ans, updated_new_block) else (* Do nothing *) pending :: ans, block) |> fun (ans, pending) -> List.rev (pending :: ans) ;; (* The aim here is to be able to transform e.g. foo f<insert>rob f</insert>lip into foo <insert>frob </insert>flip by scoring each boundary. *) (* {prev,next}_{elts,scorable} are the two arrays being diffed as (a) [Elt.t array]s which comes from mapping with the [transform] function and (b) ['a array]s where the [score] function operates on values of type ['a]. [blocks] is a list of [Matching_block.t]s which specify the contiguous chunks of prev/next which are the same (when index ranges are missing, that missing chunk is unique to the array they are missing from). *) let align_diffs ~prev_elts ~prev_scorable ~next_elts ~next_scorable ~max_slide ~score blocks = if max_slide = 0 then blocks else ( match blocks with | [] -> [] | (first_block : Matching_block.t) :: tl -> let score_elt_and_prev arr kind i = let i0 = i - 1 in let i1 = i in if i0 < 0 || i1 >= Array.length arr then 100 else score kind arr.(i0) arr.(i1) in (* [left] and [right] are both exclusive indexes, on either side of the inserted or deleted region. *) let score' arr ~left ~right = score_elt_and_prev arr `right right + score_elt_and_prev arr `left (left + 1) in let score ~prev_left ~prev_right ~next_left ~next_right = match prev_left + 1 = prev_right, next_left + 1 = next_right with | true, true -> (* This shouldn't happen; it means the two blocks of equal elements touch and should have been merged. *) 0 | true, false -> (* inserting *) score' next_scorable ~left:next_left ~right:next_right | false, true -> (* deleting *) score' prev_scorable ~left:prev_left ~right:prev_right | false, false -> (* editing. I think this also shouldn’t happen as if e.g. sliding right is viable you could just have had a bigger matching block on the right instead *) min (score' prev_scorable ~left:prev_left ~right:prev_right) (score' next_scorable ~left:next_left ~right:next_right) in let rec align_consecutive_blocks acc (left_block : Matching_block.t) right_blocks = let best_score = ref 0 in let offset_of_best_score = ref 0 in let score_initial (left_block : Matching_block.t) (right_block : Matching_block.t) = best_score := score ~prev_left:(left_block.prev_start + left_block.length - 1) ~prev_right:right_block.prev_start ~next_left:(left_block.next_start + left_block.length - 1) ~next_right:right_block.next_start in let rec try_to_slide_left ~i (left_block : Matching_block.t) (right_block : Matching_block.t) = let offset_into_left = left_block.length - i in if offset_into_left < 0 then () else if i > max_slide then () (* {v next: A B C D X P Q R S T X E F G H I prev: A B C D X E F G H I ^ ^ ^ ^ \..left../ \..right../ v} We want to see if shortening [left] and lengthening and moving-left [right] is viable. Requires last element of [left] to be equal to the element before [right]. *) else ( let prev_left = left_block.prev_start + offset_into_left - 1 in let prev_right = right_block.prev_start - i in let next_left = left_block.next_start + offset_into_left - 1 in let next_right = right_block.next_start - i in if Elt.compare next_elts.(next_left + 1) next_elts.(next_right) <> 0 then () else if Elt.compare prev_elts.(prev_left + 1) prev_elts.(prev_right) <> 0 then () else ( let score = score ~prev_left ~prev_right ~next_left ~next_right in if score > !best_score then ( best_score := score; offset_of_best_score := -i); try_to_slide_left ~i:(i + 1) left_block right_block)) in let rec try_to_slide_right ~i (left_block : Matching_block.t) (right_block : Matching_block.t) = let offset_into_left = left_block.length + i - 1 in if i > right_block.length then () else if i > max_slide then () (* {v next: A B C D X P Q R S T X E F G H I prev: A B C D X E F G H I ^ ^ ^ ^ \.left./ \...right.../ v} We want to see if lengthening [left] and shortening and moving-right [right] is viable. Requires element after [left] to be equal to the first element of [right]. *) else ( let prev_left = left_block.prev_start + offset_into_left in let prev_right = right_block.prev_start + i in let next_left = left_block.next_start + offset_into_left in let next_right = right_block.next_start + i in if Elt.compare next_elts.(next_left) next_elts.(next_right - 1) <> 0 then () else if Elt.compare prev_elts.(prev_left) prev_elts.(prev_right - 1) <> 0 then () else ( let score = score ~prev_left ~prev_right ~next_left ~next_right in if score > !best_score then ( best_score := score; offset_of_best_score := i); try_to_slide_right ~i:(i + 1) left_block right_block)) in match right_blocks with | [] -> List.rev (left_block :: acc) | right_block :: rest -> (match left_block.length with | 0 -> (* It is possible to end up with a length=0 block due to sliding the previous block to include it. If this happens we should drop it (so long as it isn’t the last block; we want a length=0 block at the end). *) align_consecutive_blocks acc right_block rest | _ -> score_initial left_block right_block; try_to_slide_left ~i:1 left_block right_block; try_to_slide_right ~i:1 left_block right_block; (match !offset_of_best_score with | 0 -> align_consecutive_blocks (left_block :: acc) right_block rest | slide -> let new_left = { left_block with length = left_block.length + slide } in let new_right : Matching_block.t = { prev_start = right_block.prev_start + slide ; next_start = right_block.next_start + slide ; length = right_block.length - slide } in let acc = if new_left.length > 0 then new_left :: acc else acc in align_consecutive_blocks acc new_right rest)) in align_consecutive_blocks [] first_block tl) ;; let get_matching_blocks ~transform ?(big_enough = 1) ?(max_slide = 0) ?(score = fun _ _ _ -> 100) ~prev:prev_scorable ~next:next_scorable () = let prev = Array.map prev_scorable ~f:transform in let next = Array.map next_scorable ~f:transform in let matches = matches prev next |> collapse_sequences in let matches = combine_equalities ~prev ~next ~matches in let last_match = { Matching_block.prev_start = Array.length prev ; next_start = Array.length next ; length = 0 } in List.append matches [ last_match ] |> semantic_cleanup ~big_enough (* We want to score untransformed elements (e.g. still with whitespace) *) |> align_diffs ~prev_elts:prev ~prev_scorable ~next_elts:next ~next_scorable ~max_slide ~score ;; let get_ranges_rev ~transform ~big_enough ?max_slide ?score ~prev ~next () = let rec aux (matching_blocks : Matching_block.t list) i j l : _ Range.t list = match matching_blocks with | current_block :: remaining_blocks -> let prev_index, next_index, size = current_block.prev_start, current_block.next_start, current_block.length in (* Throw away crossover matches *) if prev_index < i || next_index < j then aux remaining_blocks i j l else ( let range_opt : _ Range.t option = if i < prev_index && j < next_index then ( let prev_range = prev <|> (i, prev_index) in let next_range = next <|> (j, next_index) in Some (Replace (prev_range, next_range, None))) else if i < prev_index then ( let prev_range = prev <|> (i, prev_index) in Some (Prev (prev_range, None))) else if j < next_index then ( let next_range = next <|> (j, next_index) in Some (Next (next_range, None))) else None in let l = match range_opt with | Some range -> range :: l | None -> l in let prev_stop, next_stop = prev_index + size, next_index + size in let l = if size = 0 then l else ( let prev_range = prev <|> (prev_index, prev_stop) in let next_range = next <|> (next_index, next_stop) in let range = Array.map2_exn prev_range next_range ~f:(fun x y -> x, y) in Same range :: l) in aux remaining_blocks prev_stop next_stop l) | [] -> List.rev l in let matching_blocks = get_matching_blocks ~transform ~big_enough ?max_slide ?score ~prev ~next () in aux matching_blocks 0 0 [] ;; let get_hunks ~transform ~context ?(big_enough = 1) ?max_slide ?score ~prev ~next () = let ranges = get_ranges_rev ~transform ~big_enough ?max_slide ?score ~prev ~next () in let a = prev in let b = next in if context < 0 then ( let singleton_hunk = create_hunk 0 (Array.length a) 0 (Array.length b) (List.rev ranges) in [ singleton_hunk ]) else ( let rec aux ranges_remaining curr_ranges alo ahi blo bhi acc_hunks = match (ranges_remaining : _ Range.t list) with | [] -> (* Finish the last hunk *) let new_hunk = create_hunk alo ahi blo bhi curr_ranges in (* Add it to the accumulator *) let acc_hunks = new_hunk :: acc_hunks in (* Finished! Return the accumulator *) List.rev acc_hunks | [ Same range ] -> (* If the last range is a Same, we might need to crop to context. *) let stop = min (Array.length range) context in let new_range = Range.Same (range <|> (0, stop)) in let curr_ranges = new_range :: curr_ranges in (* Finish the current hunk *) let ahi = ahi + stop in let bhi = bhi + stop in let new_hunk = create_hunk alo ahi blo bhi curr_ranges in (* Add it to the accumulator *) let acc_hunks = new_hunk :: acc_hunks in (* Finished! Return the accumulator *) List.rev acc_hunks | Same range :: rest -> let size = Array.length range in if size > context * 2 then ( (* If this Same range is sufficiently large, split off a new hunk *) let new_range = Range.Same (range <|> (0, context)) in let curr_ranges = new_range :: curr_ranges in (* Advance both hi's by context *) let ahi = ahi + context in let bhi = bhi + context in (* Finish the current hunk *) let new_hunk = create_hunk alo ahi blo bhi curr_ranges in (* Add it to the accumulator *) let acc_hunks = new_hunk :: acc_hunks in (* Calculate ranges for the next hunk *) let alo = ahi + size - (2 * context) in let ahi = alo in let blo = bhi + size - (2 * context) in let bhi = blo in (* Push the remainder of the Equal range back onto the remaining_ranges *) let rest = Range.Same (range <|> (size - context, size)) :: rest in aux rest [] alo ahi blo bhi acc_hunks) else ( (* Otherwise, this range is small enough that it qualifies as context for the both the previous and forthcoming range, so simply add it to curr_ranges untouched *) let curr_ranges = Range.Same range :: curr_ranges in let ahi = ahi + size in let bhi = bhi + size in aux rest curr_ranges alo ahi blo bhi acc_hunks) | range :: rest -> (* Any range that isn't an Equal is important and not just context, so keep it in curr_ranges *) let curr_ranges = range :: curr_ranges in (* rest could be anything, so extract hunk_info from range *) let ahi, bhi = match range with | Same _ -> (* We eliminate the possibility of a Same above *) assert false | Unified _ -> (* get_ranges_rev never returns a Unified range *) assert false | Prev (_, Some _) | Next (_, Some _) | Replace (_, _, Some _) -> (* get_ranges_rev never finds moves *) assert false | Next (range, None) -> let stop = bhi + Array.length range in ahi, stop | Prev (range, None) -> let stop = ahi + Array.length range in stop, bhi | Replace (a_range, b_range, None) -> let prev_stop = ahi + Array.length a_range in let next_stop = bhi + Array.length b_range in prev_stop, next_stop in aux rest curr_ranges alo ahi blo bhi acc_hunks in let ranges, alo, ahi, blo, bhi = match ranges with (* If the first range is an Equal, shave off the front of the range, according to context. Keep it on the ranges list so hunk construction can see where the range begins *) | Same range :: rest -> let stop = Array.length range in let start = max 0 (stop - context) in let new_range = Range.Same (range <|> (start, stop)) in new_range :: rest, start, start, start, start | rest -> rest, 0, 0, 0, 0 in aux ranges [] alo ahi blo bhi []) ;; let match_ratio a b = (matches a b |> List.length |> ( * ) 2 |> float) /. (Array.length a + Array.length b |> float) ;; let collapse_multi_sequences matches = let collapsed = ref [] in let value_exn x = Option.value_exn x in if List.is_empty matches then [] else ( let start = Array.create ~len:(List.length (List.hd_exn matches)) None in let length = ref 0 in List.iter matches ~f:(fun il -> if Array.for_all start ~f:Option.is_some && List.mapi il ~f:(fun i x -> x = value_exn start.(i) + !length) |> List.for_all ~f:(fun x -> x) then incr length else ( if Array.for_all start ~f:Option.is_some then collapsed := (Array.map start ~f:value_exn |> Array.to_list, !length) :: !collapsed; List.iteri il ~f:(fun i x -> start.(i) <- Some x); length := 1)); if Array.for_all start ~f:Option.is_some && !length <> 0 then collapsed := (Array.map start ~f:value_exn |> Array.to_list, !length) :: !collapsed; List.rev !collapsed) ;; type 'a segment = | Same of 'a array | Different of 'a array array type 'a merged_array = 'a segment list let array_mapi2 ar1 ar2 ~f = Array.zip_exn ar1 ar2 |> Array.mapi ~f:(fun i (x, y) -> f i x y) ;; let merge ar = if Array.length ar = 0 then [] else if Array.length ar = 1 then [ Same ar.(0) ] else ( let matches's = Array.map (ar <|> (1, Array.length ar)) ~f:(matches ar.(0)) in let len = Array.length ar in let hashtbl = Int.Table.create () ~size:0 in Array.iteri matches's ~f:(fun i matches -> List.iter matches ~f:(fun (a, b) -> match Hashtbl.find hashtbl a with | None -> Hashtbl.set hashtbl ~key:a ~data:[ i, b ] | Some l -> Hashtbl.set hashtbl ~key:a ~data:((i, b) :: l))); let list = Hashtbl.to_alist hashtbl |> List.filter_map ~f:(fun (a, l) -> if List.length l = len - 1 then Some (a :: (List.sort l ~compare:compare_int_pair |> List.map ~f:snd)) else None) |> List.sort ~compare:(List.compare Int.compare) in let matching_blocks = collapse_multi_sequences list in let last_pos = Array.create ~len:(Array.length ar) 0 in let merged_array = ref [] in List.iter matching_blocks ~f:(fun (l, len) -> let ar' = Array.of_list l in if Array.compare Int.compare last_pos ar' <> 0 then merged_array := Different (array_mapi2 last_pos ar' ~f:(fun i n m -> ar.(i) <|> (n, m))) :: !merged_array; merged_array := Same (ar.(0) <|> (ar'.(0), ar'.(0) + len)) :: !merged_array; Array.iteri last_pos ~f:(fun i _ -> last_pos.(i) <- ar'.(i) + len)); let trailing_lines = Array.existsi last_pos ~f:(fun i last_pos -> Array.length ar.(i) > last_pos) in if trailing_lines then merged_array := Different (Array.mapi last_pos ~f:(fun i n -> ar.(i) <|> (n, Array.length ar.(i)))) :: !merged_array; List.rev !merged_array) ;; end let%test_module _ = (module struct module P = Make (Int) let%test_unit _ = let check a b ~expect = [%test_result: (int * int) list] (P.matches a b) ~expect in check [||] [||] ~expect:[]; check [| 0 |] [| 0 |] ~expect:[ 0, 0 ]; check [| 0; 1; 1; 2 |] [| 3; 1; 4; 5 |] ~expect:[ 1, 1 ] ;; (* Needs the plain diff section *) let rec is_increasing a = function | [] -> true | hd :: tl -> Int.compare a hd <= 0 && is_increasing hd tl ;; let check_lis a = let b = Patience.longest_increasing_subsequence (Ordered_sequence.create a) in if is_increasing (-1) (List.map b ~f:fst) && is_increasing (-1) (List.map b ~f:snd) then () else failwiths ~here:[%here] "invariant failure" (a, b) [%sexp_of: (int * int) list * (int * int) list] ;; let%test_unit _ = check_lis [ 2, 0; 5, 1; 6, 2; 3, 3; 0, 4; 4, 5; 1, 6 ] let%test_unit _ = check_lis [ 0, 0; 2, 0; 5, 1; 6, 2; 3, 3; 0, 4; 4, 5; 1, 6 ] let%test_unit _ = check_lis [ 5, 1; 6, 2; 3, 3; 0, 4; 4, 5; 1, 6 ] let%test_unit _ = let check a b = let matches = P.matches a b in if is_increasing (-1) (List.map matches ~f:fst) && is_increasing (-1) (List.map matches ~f:snd) then () else failwiths ~here:[%here] "invariant failure" (a, b, matches) [%sexp_of: int array * int array * (int * int) list] in check [| 0; 1; 2; 3; 4; 5; 6 |] [| 2; 5; 6; 3; 0; 4; 1 |] ;; end) ;; module String = Make (String) module Stable = struct module Matching_block = Matching_block.Stable module Range = Range.Stable module Hunk = Hunk.Stable module Hunks = Hunks.Stable end