X-Git-Url: https://git.xandkar.net/?p=dups.git;a=blobdiff_plain;f=dups.ml;h=1483754f4103268754678c9f63c4b59d1b6ac1f9;hp=c627369c5de0c44d6b6d6cbb12921f9382d17276;hb=refs%2Fheads%2Fmulti-samples;hpb=8c54ccb832581554a2c272c0f08b4300081d22b5 diff --git a/dups.ml b/dups.ml index c627369..1483754 100644 --- a/dups.ml +++ b/dups.ml @@ -1,149 +1,10 @@ open Printf -module Array = ArrayLabels -module List = ListLabels +module List = ListLabels module StrSet = Set.Make(String) -module Unix = UnixLabels +module Unix = UnixLabels -module Stream : sig - type 'a t - - val create : (unit -> 'a option) -> 'a t - - val iter : 'a t -> f:('a -> unit) -> unit - - val map : 'a t -> f:('a -> 'b) -> 'b t - - val filter : 'a t -> f:('a -> bool) -> 'a t - - val concat : ('a t) list -> 'a t -end = struct - module S = Stream - - type 'a t = - {mutable streams : ('a S.t) list} - - let create f = - {streams = [S.from (fun _ -> f ())]} - - let rec next t = - match t.streams with - | [] -> - None - | s :: streams -> - (match S.next s with - | exception Stream.Failure -> - t.streams <- streams; - next t - | x -> - Some x - ) - - let map t ~f = - create (fun () -> - match next t with - | None -> None - | Some x -> Some (f x) - ) - - let filter t ~f = - let rec filter () = - match next t with - | None -> - None - | Some x when f x -> - Some x - | Some _ -> - filter () - in - create filter - - let iter t ~f = - List.iter t.streams ~f:(S.iter f) - - let concat ts = - {streams = List.concat (List.map ts ~f:(fun {streams} -> streams))} -end - -module In_channel : sig - val lines : in_channel -> string Stream.t -end = struct - let lines ic = - Stream.create (fun () -> - match input_line ic with - | exception End_of_file -> - None - | line -> - Some line - ) -end - -module File : sig - type t = - { path : string - ; size : int - } - - val find : string -> t Stream.t - (** Find all files in the directory tree, starting from the given root path *) - - val lookup : string Stream.t -> t Stream.t - (** Lookup file info for given paths *) - - module Set : sig include Set.S with type elt := t end -end = struct - type t = - { path : string - ; size : int - } - - let compare {path=p1; _} {path=p2; _} = - Stdlib.compare p1 p2 - - let lookup paths = - Stream.map paths ~f:(fun path -> - let {Unix.st_size = size; _} = Unix.lstat path in - {path; size} - ) - - let find root = - let dirs = Queue.create () in - let files = Queue.create () in - let explore parent = - Array.iter (Sys.readdir parent) ~f:(fun child -> - let path = Filename.concat parent child in - let {Unix.st_kind = file_kind; st_size; _} = Unix.lstat path in - match file_kind with - | Unix.S_REG -> - let file = {path; size = st_size} in - Queue.add file files - | Unix.S_DIR -> - Queue.add path dirs - | Unix.S_CHR - | Unix.S_BLK - | Unix.S_LNK - | Unix.S_FIFO - | Unix.S_SOCK -> - () - ) - in - explore root; - let rec next () = - match Queue.is_empty files, Queue.is_empty dirs with - | false, _ -> Some (Queue.take files) - | true , true -> None - | true , false -> - explore (Queue.take dirs); - next () - in - Stream.create next - - module Set = Set.Make(struct - type elt = t - type t = elt - let compare = compare - end) -end +module M = Metrics type input = | Stdin @@ -156,40 +17,27 @@ type output = type opt = { input : input ; output : output - ; ignore : Str.regexp option + ; ignore : string -> bool ; sample : int + ; njobs : int + ; delim_null : bool } -type count = - { considered : int ref - ; empty : int ref - ; ignored : int ref - ; unique_size : int ref - ; unique_sample : int ref - ; hashed : int ref - } - -let make_input_stream input ignore count = +let make_input_stream input ignore ~metrics ~delim_null = let input = match input with | Stdin -> - File.lookup (In_channel.lines stdin) + File.lookup (In_channel.lines stdin ~delim_null) | Directories paths -> let paths = StrSet.elements (StrSet.of_list paths) in Stream.concat (List.map paths ~f:File.find) in Stream.filter input ~f:(fun {File.path; size} -> - incr count.considered; + M.file_considered metrics ~size; let empty = size = 0 in - let ignored = - match ignore with - | Some regexp when (Str.string_match regexp path 0) -> - true - | Some _ | None -> - false - in - if empty then incr count.empty; - if ignored then incr count.ignored; + let ignored = ignore path in + if empty then M.file_empty metrics; + if ignored then M.file_ignored metrics ~size; (not empty) && (not ignored) ) @@ -197,7 +45,7 @@ let make_output_fun = function | Stdout -> fun digest n_files files -> printf "%s %d\n%!" (Digest.to_hex digest) n_files; - List.iter (File.Set.elements files) ~f:(fun {File.path; _} -> + List.iter files ~f:(fun {File.path; _} -> printf " %S\n%!" path ) | Directory dir -> @@ -206,56 +54,23 @@ let make_output_fun = function let dir = Filename.concat dir (String.sub digest 0 2) in Unix.mkdir dir ~perm:0o700; let oc = open_out (Filename.concat dir digest) in - List.iter (File.Set.elements files) ~f:(fun {File.path; _} -> + List.iter files ~f:(fun {File.path; _} -> output_string oc (sprintf "%S\n%!" path) ); close_out oc -let sample path ~len = - let buf = Bytes.make len ' ' in - let ic = open_in_bin path in - let rec read pos len = - assert (len >= 0); - if len = 0 then - () - else begin - let chunk_size = input ic buf pos len in - if chunk_size = 0 then (* EOF *) - () - else - read (pos + chunk_size) (len - chunk_size) - end - in - read 0 len; - close_in ic; - Bytes.to_string buf +let time_wall () = + Unix.gettimeofday () -let main {input; output; ignore; sample = sample_len} = - let t0 = Sys.time () in - let count = - { considered = ref 0 - ; empty = ref 0 - ; ignored = ref 0 - ; unique_size = ref 0 - ; hashed = ref 0 - ; unique_sample = ref 0 - } - in +let time_proc () = + Sys.time () + +let main {input; output; ignore; sample = sample_len; njobs; delim_null} = + let wt0_all = time_wall () in + let pt0_all = time_proc () in + let metrics = M.init () in let output = make_output_fun output in - let input = make_input_stream input ignore count in - let files_by_size = Hashtbl.create 1_000_000 in - let files_by_sample = Hashtbl.create 1_000_000 in - let files_by_digest = Hashtbl.create 1_000_000 in - let process tbl ~group ~file = - let count, files = - match Hashtbl.find_opt tbl group with - | None -> - (0, File.Set.empty) - | Some (n, files) -> - (n, files) - in - Hashtbl.replace tbl group (count + 1, File.Set.add file files) - in + let input = make_input_stream input ignore ~metrics ~delim_null in (* TODO: Make a nice(r) abstraction to re-assemble pieces in the pipeline: * * from input to files_by_size @@ -265,53 +80,84 @@ let main {input; output; ignore; sample = sample_len} = * * input |> files_by_size |> files_by_sample |> files_by_digest |> output *) - Stream.iter input ~f:(fun ({File.size; _} as file) -> - process files_by_size ~group:size ~file + + let files = input in + + let wt0_group_by_size = time_wall () in + let pt0_group_by_size = time_proc () in + eprintf "[debug] filtering-out files with unique size\n%!"; + let files = File.filter_out_unique_sizes files ~metrics in + let pt1_group_by_size = time_proc () in + let wt1_group_by_size = time_wall () in + + let wt0_group_by_sample = wt1_group_by_size in + let pt0_group_by_sample = pt1_group_by_size in + eprintf "[debug] filtering-out files with unique heads\n%!"; + let files = + if njobs > 1 then begin + let q = Queue.create () in + files + |> Stream.bag_map ~njobs ~f:(File.head ~len:sample_len ~metrics) + |> Stream.group_by ~f:snd + |> Stream.map ~f:(fun (d, n, pairs) -> (d, n, List.map pairs ~f:fst)) + |> Stream.filter ~f:(fun (_, n, _) -> n > 1) + |> Stream.iter ~f:(fun (_, _, fs) -> List.iter fs ~f:(fun f -> Queue.add f q)) + ; + Stream.of_queue q + end else + File.filter_out_unique_heads files ~len:sample_len ~metrics + in + let pt1_group_by_sample = time_proc () in + let wt1_group_by_sample = time_wall () in + + let wt0_group_by_digest = wt1_group_by_sample in + let pt0_group_by_digest = pt1_group_by_sample in + eprintf "[debug] hashing\n%!"; + let groups = + if njobs > 1 then + let with_digests = + Stream.bag_map files ~njobs ~f:(fun {File.path; _} -> Digest.file path) + in + Stream.map (Stream.group_by with_digests ~f:snd) ~f:( + fun (digest, n, file_digest_pairs) -> + let files = + List.map file_digest_pairs ~f:(fun (file, _) -> + M.file_hashed metrics ~size:file.File.size; + file + ) + in + (digest, n, files) + ) + else + Stream.group_by files ~f:(fun {File.path; size} -> + M.file_hashed metrics ~size; + Digest.file path + ) + in + let pt1_group_by_digest = time_proc () in + let wt1_group_by_digest = time_wall () in + + eprintf "[debug] reporting\n%!"; + Stream.iter groups ~f:(fun (d, n, files) -> + M.digest metrics; + if n > 1 then begin + M.redundant_data metrics ~size:(n * (List.hd files).File.size); + output d n files + end ); - Hashtbl.iter - (fun _ (n, files) -> - (* Skip files with unique sizes *) - if n > 1 then - File.Set.iter - (fun ({File.path; _} as file) -> - process - files_by_sample - ~group:(sample path ~len:sample_len) - ~file - ) - files - else - incr count.unique_size; - ) - files_by_size; - Hashtbl.iter - (fun _ (n, files) -> - (* Skip files with unique samples *) - if n > 1 then - File.Set.iter - (fun ({File.path; _} as file) -> - incr count.hashed; - process files_by_digest ~group:(Digest.file path) ~file - ) - files - else - incr count.unique_sample; - ) - files_by_sample; - Hashtbl.iter - (fun d (n, files) -> - if n > 1 then - output d n files - ) - files_by_digest; - let t1 = Sys.time () in - eprintf "Time : %f seconds\n%!" (t1 -. t0); - eprintf "Considered : %d\n%!" !(count.considered); - eprintf "Hashed : %d\n%!" !(count.hashed); - eprintf "Skipped due to 0 size : %d\n%!" !(count.empty); - eprintf "Skipped due to unique size : %d\n%!" !(count.unique_size); - eprintf "Skipped due to unique sample : %d\n%!" !(count.unique_sample); - eprintf "Ignored due to regex match : %d\n%!" !(count.ignored) + + let pt1_all = time_proc () in + let wt1_all = time_wall () in + + M.report metrics + ~wall_time_all: (wt1_all -. wt0_all) + ~wall_time_group_by_size: (wt1_group_by_size -. wt0_group_by_size) + ~wall_time_group_by_head: (wt1_group_by_sample -. wt0_group_by_sample) + ~wall_time_group_by_digest:(wt1_group_by_digest -. wt0_group_by_digest) + ~proc_time_all: (pt1_all -. pt0_all) + ~proc_time_group_by_size: (pt1_group_by_size -. pt0_group_by_size) + ~proc_time_group_by_head: (pt1_group_by_sample -. pt0_group_by_sample) + ~proc_time_group_by_digest:(pt1_group_by_digest -. pt0_group_by_digest) let get_opt () : opt = let assert_ test x msg = @@ -328,8 +174,10 @@ let get_opt () : opt = in let input = ref Stdin in let output = ref Stdout in - let ignore = ref None in - let sample = ref 256 in + let ignore = ref (fun _ -> false) in + let sample = ref 512 in + let njobs = ref 6 in + let input_delim_null = ref false in let spec = [ ( "-out" , Arg.String (fun path -> @@ -340,13 +188,29 @@ let get_opt () : opt = , " Output to this directory instead of stdout." ) ; ( "-ignore" - , Arg.String (fun regexp -> ignore := Some (Str.regexp regexp)) + , Arg.String (fun regexp -> + let regexp = Str.regexp regexp in + ignore := fun string -> Str.string_match regexp string 0) , " Ignore file paths which match this regexp pattern (see Str module)." ) ; ( "-sample" , Arg.Set_int sample , (sprintf " Byte size of file samples to use. Default: %d" !sample) ) + ; ( "-j" + , Arg.Set_int njobs + , (sprintf " Number of parallel jobs. Default: %d" !njobs) + ) + ; ( "-0" + , Arg.Set input_delim_null + , ( sprintf + ( " Delimit input paths by null character instead of a newline." + ^^" Meaningful only when reading candidate paths from stdin." + ^^" Default: %B" + ) + !input_delim_null + ) + ) ] in Arg.parse @@ -369,6 +233,8 @@ let get_opt () : opt = ; output = !output ; ignore = !ignore ; sample = !sample + ; njobs = !njobs + ; delim_null = !input_delim_null } let () =