%%
%% %CopyrightBegin%
%%
%% Copyright Ericsson AB 2001-2013. All Rights Reserved.
%%
%% Licensed under the Apache License, Version 2.0 (the "License");
%% you may not use this file except in compliance with the License.
%% You may obtain a copy of the License at
%%
%% http://www.apache.org/licenses/LICENSE-2.0
%%
%% Unless required by applicable law or agreed to in writing, software
%% distributed under the License is distributed on an "AS IS" BASIS,
%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
%% See the License for the specific language governing permissions and
%% limitations under the License.
%%
%% %CopyrightEnd%
%%
-module(file_sorter).
%% Avoid warning for local function error/2 clashing with autoimported BIF.
-compile({no_auto_import,[error/2]}).
-export([sort/1, sort/2, sort/3,
keysort/2, keysort/3, keysort/4,
merge/2, merge/3,
keymerge/3, keymerge/4,
check/1, check/2,
keycheck/2, keycheck/3]).
-include_lib("kernel/include/file.hrl").
-define(CHUNKSIZE, 16384).
-define(RUNSIZE, 524288).
-define(NOMERGE, 16).
-define(MERGESIZE, ?CHUNKSIZE).
-define(MAXSIZE, (1 bsl 31)).
-record(w, {keypos, runs = [[]], seq = 1, in, out, fun_out, prefix, temp = [],
format, runsize, no_files, order, chunksize, wfd, ref, z, unique,
hdlen, inout_value}).
-record(opts, {format = binary_term_fun(), size = ?RUNSIZE,
no_files = ?NOMERGE, tmpdir = default, order = ascending,
compressed = false, unique = false, header = 4}).
-compile({inline, [{badarg, 2}, {make_key,2}, {make_stable_key,3}, {cfun,3}]}).
%%%
%%% Exported functions
%%%
-export_type([reason/0]).
-type(file_name() :: file:name()).
-type(file_names() :: [file:name()]).
-type(i_command() :: read | close).
-type(i_reply() :: end_of_input | {end_of_input, value()}
| {[object()], infun()} | input_reply()).
-type(infun() :: fun((i_command()) -> i_reply())).
-type(input() :: file_names() | infun()).
-type(input_reply() :: term()).
-type(o_command() :: {value, value()} | [object()] | close).
-type(o_reply() :: outfun() | output_reply()).
-type(object() :: term() | binary()).
-type(outfun() :: fun((o_command()) -> o_reply())).
-type(output() :: file_name() | outfun()).
-type(output_reply() :: term()).
-type(value() :: term()).
-type(options() :: [option()] | option()).
-type(option() :: {compressed, boolean()}
| {header, header_length()}
| {format, format()}
| {no_files, no_files()}
| {order, order()}
| {size, size()}
| {tmpdir, tmp_directory()}
| {unique, boolean()}).
-type(format() :: binary_term | term | binary | format_fun()).
-type(format_fun() :: fun((binary()) -> term())).
-type(header_length() :: pos_integer()).
-type(key_pos() :: pos_integer() | [pos_integer()]).
-type(no_files() :: pos_integer()). % > 1
-type(order() :: ascending | descending | order_fun()).
-type(order_fun() :: fun((term(), term()) -> boolean())).
-type(size() :: non_neg_integer()).
-type(tmp_directory() :: [] | file:name()).
-type(reason() :: bad_object
| {bad_object, file_name()}
| {bad_term, file_name()}
| {file_error, file_name(),
file:posix() | badarg | system_limit}
| {premature_eof, file_name()}).
-spec(sort(FileName) -> Reply when
FileName :: file_name(),
Reply :: ok | {error, reason()} | input_reply() | output_reply()).
sort(FileName) ->
sort([FileName], FileName).
-spec(sort(Input, Output) -> Reply when
Input :: input(),
Output :: output(),
Reply :: ok | {error, reason()} | input_reply() | output_reply()).
sort(Input, Output) ->
sort(Input, Output, []).
-spec(sort(Input, Output, Options) -> Reply when
Input :: input(),
Output :: output(),
Options :: options(),
Reply :: ok | {error, reason()} | input_reply() | output_reply()).
sort(Input0, Output0, Options) ->
case {is_input(Input0), maybe_output(Output0), options(Options)} of
{{true,Input}, {true,Output}, #opts{}=Opts} ->
do_sort(0, Input, Output, Opts, sort);
T ->
badarg(culprit(tuple_to_list(T)), [Input0, Output0, Options])
end.
-spec(keysort(KeyPos, FileName) -> Reply when
KeyPos :: key_pos(),
FileName :: file_name(),
Reply :: ok | {error, reason()} | input_reply() | output_reply()).
keysort(KeyPos, FileName) ->
keysort(KeyPos, [FileName], FileName).
-spec(keysort(KeyPos, Input, Output) -> Reply when
KeyPos :: key_pos(),
Input :: input(),
Output :: output(),
Reply :: ok | {error, reason()} | input_reply() | output_reply()).
keysort(KeyPos, Input, Output) ->
keysort(KeyPos, Input, Output, []).
-spec(keysort(KeyPos, Input, Output, Options) -> Reply when
KeyPos :: key_pos(),
Input :: input(),
Output :: output(),
Options :: options(),
Reply :: ok | {error, reason()} | input_reply() | output_reply()).
keysort(KeyPos, Input0, Output0, Options) ->
R = case {is_keypos(KeyPos), is_input(Input0),
maybe_output(Output0), options(Options)} of
{_, _, _, #opts{format = binary}} ->
{Input0,Output0,[{badarg,format}]};
{_, _, _, #opts{order = Order}} when is_function(Order) ->
{Input0,Output0,[{badarg,order}]};
{true, {true,In}, {true,Out}, #opts{}=Opts} ->
{In,Out,Opts};
T ->
{Input0,Output0,tuple_to_list(T)}
end,
case R of
{Input,Output,#opts{}=O} ->
do_sort(KeyPos, Input, Output, O, sort);
{_,_,O} ->
badarg(culprit(O), [KeyPos, Input0, Output0, Options])
end.
-spec(merge(FileNames, Output) -> Reply when
FileNames :: file_names(),
Output :: output(),
Reply :: ok | {error, reason()} | output_reply()).
merge(Files, Output) ->
merge(Files, Output, []).
-spec(merge(FileNames, Output, Options) -> Reply when
FileNames :: file_names(),
Output :: output(),
Options :: options(),
Reply :: ok | {error, reason()} | output_reply()).
merge(Files0, Output0, Options) ->
case {is_files(Files0), maybe_output(Output0), options(Options)} of
%% size not used
{{true,Files}, {true,Output}, #opts{}=Opts} ->
do_sort(0, Files, Output, Opts, merge);
T ->
badarg(culprit(tuple_to_list(T)), [Files0, Output0, Options])
end.
-spec(keymerge(KeyPos, FileNames, Output) -> Reply when
KeyPos :: key_pos(),
FileNames :: file_names(),
Output :: output(),
Reply :: ok | {error, reason()} | output_reply()).
keymerge(KeyPos, Files, Output) ->
keymerge(KeyPos, Files, Output, []).
-spec(keymerge(KeyPos, FileNames, Output, Options) -> Reply when
KeyPos :: key_pos(),
FileNames :: file_names(),
Output :: output(),
Options :: options(),
Reply :: ok | {error, reason()} | output_reply()).
keymerge(KeyPos, Files0, Output0, Options) ->
R = case {is_keypos(KeyPos), is_files(Files0),
maybe_output(Output0), options(Options)} of
{_, _, _, #opts{format = binary}} ->
{Files0,Output0,[{badarg,format}]};
{_, _, _, #opts{order = Order}} when is_function(Order) ->
{Files0,Output0,[{badarg,order}]};
{true, {true,Fs}, {true,Out}, #opts{}=Opts} ->
{Fs,Out,Opts};
T ->
{Files0,Output0,tuple_to_list(T)}
end,
case R of
{Files,Output,#opts{}=O} ->
do_sort(KeyPos, Files, Output, O, merge);
{_,_,O} ->
badarg(culprit(O), [KeyPos, Files0, Output0, Options])
end.
-spec(check(FileName) -> Reply when
FileName :: file_name(),
Reply :: {ok, [Result]} | {error, reason()},
Result :: {FileName, TermPosition, term()},
TermPosition :: pos_integer()).
check(FileName) ->
check([FileName], []).
-spec(check(FileNames, Options) -> Reply when
FileNames :: file_names(),
Options :: options(),
Reply :: {ok, [Result]} | {error, reason()},
Result :: {FileName, TermPosition, term()},
FileName :: file_name(),
TermPosition :: pos_integer()).
check(Files0, Options) ->
case {is_files(Files0), options(Options)} of
{{true,Files}, #opts{}=Opts} ->
do_sort(0, Files, undefined, Opts, check);
T ->
badarg(culprit(tuple_to_list(T)), [Files0, Options])
end.
-spec(keycheck(KeyPos, FileName) -> Reply when
KeyPos :: key_pos(),
FileName :: file_name(),
Reply :: {ok, [Result]} | {error, reason()},
Result :: {FileName, TermPosition, term()},
TermPosition :: pos_integer()).
keycheck(KeyPos, FileName) ->
keycheck(KeyPos, [FileName], []).
-spec(keycheck(KeyPos, FileNames, Options) -> Reply when
KeyPos :: key_pos(),
FileNames :: file_names(),
Options :: options(),
Reply :: {ok, [Result]} | {error, reason()},
Result :: {FileName, TermPosition, term()},
FileName :: file_name(),
TermPosition :: pos_integer()).
keycheck(KeyPos, Files0, Options) ->
R = case {is_keypos(KeyPos), is_files(Files0), options(Options)} of
{_, _, #opts{format = binary}} ->
{Files0,[{badarg,format}]};
{_, _, #opts{order = Order}} when is_function(Order) ->
{Files0,[{badarg,order}]};
{true, {true,Fs}, #opts{}=Opts} ->
{Fs,Opts};
T ->
{Files0,tuple_to_list(T)}
end,
case R of
{Files,#opts{}=O} ->
do_sort(KeyPos, Files, undefined, O, check);
{_,O} ->
badarg(culprit(O), [KeyPos, Files0, Options])
end.
%%%
%%% Local functions
%%%
%%-define(debug, true).
-ifdef(debug).
-define(DEBUG(S, A), io:format(S, A)).
-else.
-define(DEBUG(S, A), ok).
-endif.
culprit([{error, _} = E | _]) ->
E;
culprit([{badarg, _} = B | _]) ->
B;
culprit([_ | B]) ->
culprit(B).
%% Inlined.
badarg({error, _} = E, _Args) ->
E;
badarg({badarg, _} = B, Args) ->
erlang:error(B, Args).
options(Options) when is_list(Options) ->
options(Options, #opts{});
options(Option) ->
options([Option]).
options([{format, Format} | L], Opts) when Format =:= binary;
Format =:= term;
is_function(Format),
is_function(Format, 1) ->
options(L, Opts#opts{format = Format});
options([{format, binary_term} | L], Opts) ->
options(L, Opts#opts{format = binary_term_fun()});
options([{size, Size} | L], Opts) when is_integer(Size), Size >= 0 ->
options(L, Opts#opts{size = erlang:max(Size, 1)});
options([{no_files, NoFiles} | L], Opts) when is_integer(NoFiles),
NoFiles > 1 ->
options(L, Opts#opts{no_files = NoFiles});
options([{tmpdir, ""} | L], Opts) ->
options(L, Opts#opts{tmpdir = default});
options([{tmpdir, Dir} | L], Opts) ->
case catch filename:absname(Dir) of
{'EXIT', _} ->
{badarg, Dir};
FileName ->
options(L, Opts#opts{tmpdir = {dir, FileName}})
end;
options([{order, Fun} | L], Opts) when is_function(Fun), is_function(Fun, 2) ->
options(L, Opts#opts{order = Fun});
options([{order, Order} | L], Opts) when Order =:= ascending;
Order =:= descending ->
options(L, Opts#opts{order = Order});
options([{compressed, Bool} | L], Opts) when is_boolean(Bool) ->
options(L, Opts#opts{compressed = Bool});
options([{unique, Bool} | L], Opts) when is_boolean(Bool) ->
options(L, Opts#opts{unique = Bool});
options([{header, Len} | L], Opts)
when is_integer(Len), Len > 0, Len < ?MAXSIZE ->
options(L, Opts#opts{header = Len});
options([], Opts) ->
if
Opts#opts.format =:= term, Opts#opts.header =/= 4 ->
{badarg, header};
true ->
Opts
end;
options([Bad | _], _Opts) ->
{badarg, Bad};
options(Bad, _Opts) ->
{badarg, Bad}.
-define(OBJ(X, Y), {X, Y}).
-define(SK(T, I), [T | I]). % stable key
do_sort(KeyPos0, Input0, Output0, Opts, Do) ->
#opts{format = Format0, size = Size, no_files = NoFiles,
tmpdir = TmpDir, order = Order, compressed = Compressed,
unique = Unique, header = HdLen} = Opts,
Prefix = tmp_prefix(Output0, TmpDir),
ChunkSize = ?CHUNKSIZE,
Ref = make_ref(),
KeyPos = case KeyPos0 of [Kp] -> Kp; _ -> KeyPos0 end,
{Format, Input} = wrap_input(Format0, Do, Input0),
Z = if Compressed -> [compressed]; true -> [] end,
{Output, FunOut} = wrap_output_terms(Format0, Output0, Z),
W = #w{keypos = KeyPos, out = Output, fun_out = FunOut,
prefix = Prefix, format = Format, runsize = Size,
no_files = NoFiles, order = Order, chunksize = ChunkSize,
ref = Ref, z = Z, unique = Unique, hdlen = HdLen,
inout_value = no_value},
try
doit(Do, Input, W)
catch {Ref,Error} ->
Error
end.
doit(sort, Input, W) ->
files(1, [], 0, W, Input);
doit(merge, Input, W) ->
last_merge(Input, W);
doit(check, Input, W) ->
check_files(Input, W, []).
wrap_input(term, check, Files) ->
Fun = fun(File) ->
Fn = merge_terms_fun(file_rterms(no_file, [File])),
{fn, Fn, File}
end,
{binary_term_fun(), [Fun(F) || F <- Files]};
wrap_input(Format, check, Files) ->
{Format, Files};
wrap_input(term, merge, Files) ->
Fun = fun(File) -> merge_terms_fun(file_rterms(no_file, [File])) end,
Input = lists:reverse([Fun(F) || F <- Files]),
{binary_term_fun(), Input};
wrap_input(Format, merge, Files) ->
Input = lists:reverse([merge_bins_fun(F) || F <- Files]),
{Format, Input};
wrap_input(term, sort, InFun) when is_function(InFun, 1) ->
{binary_term_fun(), fun_rterms(InFun)};
wrap_input(term, sort, Files) ->
{binary_term_fun(), file_rterms(no_file, Files)};
wrap_input(Format, sort, Input) ->
{Format, Input}.
merge_terms_fun(RFun) ->
fun(close) ->
RFun(close);
({I, [], _LSz, W}) ->
case RFun(read) of
end_of_input ->
eof;
{Objs, NRFun} when is_function(NRFun), is_function(NRFun, 1) ->
{_, [], Ts, _} = fun_objs(Objs, [], 0, ?MAXSIZE, I, W),
{{I, Ts, ?CHUNKSIZE}, merge_terms_fun(NRFun)};
Error ->
error(Error, W)
end
end.
merge_bins_fun(FileName) ->
fun(close) ->
ok;
({_I, _L, _LSz, W} = A) ->
Fun = read_fun(FileName, user, W),
Fun(A)
end.
wrap_output_terms(term, OutFun, _Z) when is_function(OutFun),
is_function(OutFun, 1) ->
{fun_wterms(OutFun), true};
wrap_output_terms(term, File, Z) when File =/= undefined ->
{file_wterms(name, File, Z++[write]), false};
wrap_output_terms(_Format, Output, _Z) ->
{Output, is_function(Output) and is_function(Output, 1)}.
binary_term_fun() ->
fun binary_to_term/1.
check_files([], _W, L) ->
{ok, lists:reverse(L)};
check_files([FN | FNs], W, L) ->
{IFun, FileName} =
case FN of
{fn, Fun, File} ->
{Fun, File};
File ->
{read_fun(File, user, W), File}
end,
NW = W#w{in = IFun},
check_run(IFun, FileName, FNs, NW, L, 2, nolast).
check_run(IFun, F, FNs, W, L, I, Last) ->
case IFun({{merge,I}, [], 0, W}) of
{{_I, Objs, _LSz}, IFun1} ->
NW = W#w{in = IFun1},
check_objs0(IFun1, F, FNs, NW, L, I, Last, lists:reverse(Objs));
eof ->
NW = W#w{in = undefined},
check_files(FNs, NW, L)
end.
check_objs0(IFun, F, FNs, W, L, I, nolast, [?OBJ(T,_BT) | Os]) ->
check_objs1(IFun, F, FNs, W, L, I, T, Os);
check_objs0(IFun, F, FNs, W, L, I, Last, []) ->
check_run(IFun, F, FNs, W, L, I, Last);
check_objs0(IFun, F, FNs, W, L, I, {last, Last}, Os) ->
check_objs1(IFun, F, FNs, W, L, I, Last, Os).
check_objs1(IFun, F, FNs, W, L, I, LastT, Os) ->
case W of
#w{order = ascending, unique = true} ->
ucheck_objs(IFun, F, FNs, W, L, I, LastT, Os);
#w{order = ascending, unique = false} ->
check_objs(IFun, F, FNs, W, L, I, LastT, Os);
#w{order = descending, unique = true} ->
rucheck_objs(IFun, F, FNs, W, L, I, LastT, Os);
#w{order = descending, unique = false} ->
rcheck_objs(IFun, F, FNs, W, L, I, LastT, Os);
#w{order = CF, unique = true} ->
uccheck_objs(IFun, F, FNs, W, L, I, LastT, Os, CF);
#w{order = CF, unique = false} ->
ccheck_objs(IFun, F, FNs, W, L, I, LastT, Os, CF)
end.
check_objs(IFun, F, FNs, W, L, I, Last, [?OBJ(T,_BT) | Os]) when T >= Last ->
check_objs(IFun, F, FNs, W, L, I+1, T, Os);
check_objs(IFun, F, FNs, W, L, I, _Last, [?OBJ(_T,BT) | _]) ->
culprit_found(IFun, F, FNs, W, L, I, BT);
check_objs(IFun, F, FNs, W, L, I, Last, []) ->
check_run(IFun, F, FNs, W, L, I, {last, Last}).
rcheck_objs(IFun, F, FNs, W, L, I, Last, [?OBJ(T,_BT) | Os]) when T =< Last ->
rcheck_objs(IFun, F, FNs, W, L, I+1, T, Os);
rcheck_objs(IFun, F, FNs, W, L, I, _Last, [?OBJ(_T,BT) | _]) ->
culprit_found(IFun, F, FNs, W, L, I, BT);
rcheck_objs(IFun, F, FNs, W, L, I, Last, []) ->
check_run(IFun, F, FNs, W, L, I, {last, Last}).
ucheck_objs(IFun, F, FNs, W, L, I, LT, [?OBJ(T,_BT) | Os]) when T > LT ->
ucheck_objs(IFun, F, FNs, W, L, I+1, T, Os);
ucheck_objs(IFun, F, FNs, W, L, I, _LT, [?OBJ(_T,BT) | _]) ->
culprit_found(IFun, F, FNs, W, L, I, BT);
ucheck_objs(IFun, F, FNs, W, L, I, LT, []) ->
check_run(IFun, F, FNs, W, L, I, {last, LT}).
rucheck_objs(IFun, F, FNs, W, L, I, LT, [?OBJ(T,_BT) | Os]) when T < LT ->
rucheck_objs(IFun, F, FNs, W, L, I+1, T, Os);
rucheck_objs(IFun, F, FNs, W, L, I, _LT, [?OBJ(_T,BT) | _]) ->
culprit_found(IFun, F, FNs, W, L, I, BT);
rucheck_objs(IFun, F, FNs, W, L, I, LT, []) ->
check_run(IFun, F, FNs, W, L, I, {last, LT}).
ccheck_objs(IFun, F, FNs, W, L, I, LT, [?OBJ(T,BT) | Os], CF) ->
case CF(LT, T) of
true -> % LT =< T
ccheck_objs(IFun, F, FNs, W, L, I+1, T, Os, CF);
false -> % LT > T
culprit_found(IFun, F, FNs, W, L, I, BT)
end;
ccheck_objs(IFun, F, FNs, W, L, I, LT, [], _CF) ->
check_run(IFun, F, FNs, W, L, I, {last, LT}).
uccheck_objs(IFun, F, FNs, W, L, I, LT, [?OBJ(T,BT) | Os], CF) ->
case CF(LT, T) of
true -> % LT =< T
case CF(T, LT) of
true -> % T equal to LT
culprit_found(IFun, F, FNs, W, L, I, BT);
false -> % LT < T
uccheck_objs(IFun, F, FNs, W, L, I+1, T, Os, CF)
end;
false -> % LT > T
culprit_found(IFun, F, FNs, W, L, I, BT)
end;
uccheck_objs(IFun, F, FNs, W, L, I, LT, [], _CF) ->
check_run(IFun, F, FNs, W, L, I, {last, LT}).
culprit_found(IFun, F, FNs, W, L, I, [_Size | BT]) ->
IFun(close),
check_files(FNs, W, [{F,I,binary_to_term(BT)} | L]).
files(_I, L, _LSz, #w{seq = 1, out = Out}=W, []) ->
%% No temporary files created, everything in L.
case Out of
Fun when is_function(Fun) ->
SL = internal_sort(L, W),
W1 = outfun(binterm_objects(SL, []), W),
NW = close_input(W1),
outfun(close, NW);
Out ->
_ = write_run(L, W, Out),
ok
end;
files(_I, L, _LSz, W, []) ->
W1 = write_run(L, W),
last_merge(lists:append(W1#w.runs), W1);
files(I, L, LSz, W, Fun) when is_function(Fun) ->
NW = W#w{in = Fun},
fun_run(I, L, LSz, NW, []);
files(I, L, LSz, W, [FileName | FileNames]) ->
InFun = read_fun(FileName, user, W),
NW = W#w{in = InFun},
file_run(InFun, FileNames, I, L, LSz, NW).
file_run(InFun, FileNames, I, L, LSz, W) when LSz < W#w.runsize ->
case InFun({I, L, LSz, W}) of
{{I1, L1, LSz1}, InFun1} ->
NW = W#w{in = InFun1},
file_run(InFun1, FileNames, I1, L1, LSz1, NW);
eof ->
NW = W#w{in = undefined},
files(I, L, LSz, NW, FileNames)
end;
file_run(InFun, FileNames, I, L, _LSz, W) ->
NW = write_run(L, W),
file_run(InFun, FileNames, I, [], 0, NW).
fun_run(I, L, LSz, W, []) ->
case infun(W) of
{end_of_input, NW} ->
files(I, L, LSz, NW, []);
{cont, NW, Objs} ->
fun_run(I, L, LSz, NW, Objs)
end;
fun_run(I, L, LSz, #w{runsize = Runsize}=W, Objs) when LSz < Runsize ->
{NI, NObjs, NL, NLSz} = fun_objs(Objs, L, LSz, Runsize, I, W),
fun_run(NI, NL, NLSz, W, NObjs);
fun_run(I, L, _LSz, W, Objs) ->
NW = write_run(L, W),
fun_run(I, [], 0, NW, Objs).
write_run([], W) ->
W;
write_run(L, W) ->
{W1, Temp} = next_temp(W),
NW = write_run(L, W1, Temp),
[R | Rs] = NW#w.runs,
merge_runs([[Temp | R] | Rs], [], NW).
write_run(L, W, FileName) ->
SL = internal_sort(L, W),
BTs = binterms(SL, []),
{Fd, W1} = open_file(FileName, W),
write(Fd, FileName, BTs, W1),
close_file(Fd, W1).
%% Returns a list in reversed order.
internal_sort([]=L, _W) ->
L;
internal_sort(L, #w{order = CFun, unique = Unique}) when is_function(CFun) ->
Fun = fun(?OBJ(T1, _), ?OBJ(T2, _)) -> CFun(T1, T2) end,
RL = lists:reverse(L),
lists:reverse(if
Unique ->
lists:usort(Fun, RL);
true ->
lists:sort(Fun, RL)
end);
internal_sort(L, #w{unique = true, keypos = 0}=W) ->
rev(lists:usort(L), W);
internal_sort(L, #w{unique = false, keypos = 0}=W) ->
rev(lists:sort(L), W);
internal_sort(L, #w{unique = true}=W) ->
rev(lists:ukeysort(1, lists:reverse(L)), W);
internal_sort(L, #w{unique = false}=W) ->
rev(lists:keysort(1, lists:reverse(L)), W).
rev(L, #w{order = ascending}) ->
lists:reverse(L);
rev(L, #w{order = descending}) ->
L.
last_merge(R, W) when length(R) =< W#w.no_files ->
case W#w.out of
Fun when is_function(Fun) ->
{Fs, W1} = init_merge(lists:reverse(R), 1, [], W),
?DEBUG("merging ~tp~n", [lists:reverse(R)]),
W2 = merge_files(Fs, [], 0, nolast, W1),
NW = close_input(W2),
outfun(close, NW);
Out ->
_ = merge_files(R, W, Out),
ok
end;
last_merge(R, W) ->
L = lists:sublist(R, W#w.no_files),
{M, NW} = merge_files(L, W),
last_merge([M | lists:nthtail(W#w.no_files, R)], NW).
merge_runs([R | Rs], NRs0, W) when length(R) < W#w.no_files ->
NRs = lists:reverse(NRs0) ++ [R | Rs],
W#w{runs = NRs};
merge_runs([R], NRs0, W) ->
{M, NW} = merge_files(R, W),
NRs = [[] | lists:reverse([[M] | NRs0])],
NW#w{runs = NRs};
merge_runs([R, R1 | Rs], NRs0, W) ->
{M, NW} = merge_files(R, W),
merge_runs([[M | R1] | Rs], [[] | NRs0], NW).
merge_files(R, W) ->
{W1, Temp} = next_temp(W),
?DEBUG("merging ~tp~nto ~tp~n", [lists:reverse(R), Temp]),
{Temp, merge_files(R, W1, Temp)}.
merge_files(R, W, FileName) ->
{Fs, W1} = init_merge(lists:reverse(R), 1, [], W),
{Fd, W2} = open_file(FileName, W1),
W3 = W2#w{wfd = {Fd, FileName}},
W4 = merge_files(Fs, [], 0, nolast, W3),
NW = W4#w{wfd = undefined},
close_file(Fd, NW).
%% A file number, I, is used for making the merge phase stable.
init_merge([FN | FNs], I, Fs, W) ->
IFun = case FN of
_ when is_function(FN) ->
%% When and only when merge/2,3 or keymerge/3,4 was called.
FN;
_ ->
read_fun(FN, fsort, W)
end,
W1 = W#w{temp = [IFun | lists:delete(FN, W#w.temp)]},
case read_more(IFun, I, 0, W1) of
{Ts, _LSz, NIFun, NW} ->
InEtc = {I, NIFun},
init_merge(FNs, I+1, [[Ts | InEtc] | Fs], NW);
{eof, NW} -> % can only happen when merging files
init_merge(FNs, I+1, Fs, NW)
end;
init_merge([], _I, Fs0, #w{order = ascending}=W) ->
{lists:sort(Fs0), W};
init_merge([], _I, Fs0, #w{order = descending}=W) ->
{lists:reverse(lists:sort(Fs0)), W};
init_merge([], _I, Fs0, #w{order = Order}=W) when is_function(Order) ->
{lists:sort(cfun_files(W#w.order), lists:reverse(Fs0)), W}.
cfun_files(CFun) ->
fun(F1, F2) ->
[[?OBJ(T1,_) | _] | _] = F1,
[[?OBJ(T2,_) | _] | _] = F2,
CFun(T1, T2)
end.
%% The argument Last is used when unique = true. It is the last kept
%% element.
%% LSz is not the sum of the sizes of objects in L. Instead it is
%% the number of bytes read. After init_merge it is set to 0, which
%% means that the first chunk written may be quite large (it may take
%% a while before buffers are exhausted).
merge_files([F1, F2 | Fs], L0, LSz, Last0, W) when LSz < ?MERGESIZE ->
[Ts0 | InEtc] = F1,
Kind = merge_kind(W),
{Last, L, Ts} = case {Last0, Kind} of
{{last, Lst}, Kind} ->
{Lst, L0, Ts0};
{nolast, {ukmerge, _Kp}} ->
[?OBJ(?SK(T, _I), BT) | Ts1] = Ts0,
{T, [BT], Ts1};
{nolast, {rukmerge, _Kp}} ->
[?OBJ(?SK(T, _I), BT) | Ts1] = Ts0,
{{T, BT}, [], Ts1};
{nolast, _} ->
[?OBJ(T, BT) | Ts1] = Ts0,
{T, [BT], Ts1}
end,
[[?OBJ(T2, BT2) | Ts2T] = Ts2 | InEtc2] = F2,
{NInEtc, NFs, NL, NLast} =
case Kind of
umerge ->
umerge_files(L, F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Last);
{ukmerge, Kp} ->
ukmerge_files(L, F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Kp, Last);
merge ->
merge_files(L, F2, Fs, InEtc2, BT2, Ts2T, Ts, InEtc, T2);
rumerge ->
rumerge_files(L, F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Last);
{rukmerge, Kp} ->
{Lt, LtBT} = Last,
rukmerge_files(L, F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Kp,
Lt, LtBT);
rmerge ->
rmerge_files(L, F2, Fs, InEtc2, BT2, Ts2T, Ts, InEtc, T2);
{ucmerge, CF} ->
{I2, _} = InEtc2,
{I, _} = InEtc,
ucmerge_files(L, F2, Fs, InEtc2, Ts2, I2, Ts, I, InEtc, T2, CF,
Last);
{cmerge, CF} ->
{I2, _} = InEtc2,
{I, _} = InEtc,
cmerge_files(L, F2, Fs, InEtc2, BT2, Ts2T, I2, Ts, I, InEtc, T2,
CF)
end,
read_chunk(NInEtc, NFs, NL, LSz, NLast, W);
merge_files([F1], L, LSz, Last, W) when LSz < ?MERGESIZE ->
[Ts | InEtc] = F1,
NL = last_file(Ts, L, Last, merge_kind(W), W),
read_chunk(InEtc, [], NL, LSz, nolast, W);
merge_files([], [], 0, nolast, W) ->
%% When merging files, ensure that the output fun (if there is
%% one) is called at least once before closing.
merge_write(W, []);
merge_files([], L, _LSz, Last, W) ->
Last = nolast,
merge_write(W, L);
merge_files(Fs, L, _LSz, Last, W) ->
NW = merge_write(W, L),
merge_files(Fs, [], 0, Last, NW).
merge_kind(#w{order = ascending, unique = true, keypos = 0}) ->
umerge;
merge_kind(#w{order = ascending, unique = true, keypos = Kp}) ->
{ukmerge, Kp};
merge_kind(#w{order = ascending, unique = false}) ->
merge;
merge_kind(#w{order = descending, unique = true, keypos = 0}) ->
rumerge;
merge_kind(#w{order = descending, unique = true, keypos = Kp}) ->
{rukmerge, Kp};
merge_kind(#w{order = descending, unique = false}) ->
rmerge;
merge_kind(#w{order = CF, unique = true}) ->
{ucmerge, CF};
merge_kind(#w{order = CF, unique = false}) ->
{cmerge, CF}.
merge_write(W, L) ->
case {W#w.wfd, W#w.out} of
{undefined, Fun} when is_function(Fun) ->
outfun(objects(L, []), W);
{{Fd, FileName}, _} ->
write(Fd, FileName, lists:reverse(L), W),
W
end.
umerge_files(L, F2, Fs, InEtc2, Ts2, [?OBJ(T, _BT) | Ts], InEtc, T2, Last)
when T == Last ->
umerge_files(L, F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Last);
umerge_files(L, F2, Fs, InEtc2, Ts2, [?OBJ(T, BT) | Ts], InEtc, T2, _Last)
when T =< T2 ->
umerge_files([BT | L], F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, T);
umerge_files(L, F2, Fs, _InEtc2, _Ts2, [], InEtc, _T2, Last) ->
{InEtc, [F2 | Fs], L, {last, Last}};
umerge_files(L, _F2, Fs, InEtc2, Ts2, Ts, InEtc, _T2, Last) ->
[F3 | NFs] = insert([Ts | InEtc], Fs),
[[?OBJ(T3,_BT3) | _] = Ts3 | InEtc3] = F3,
umerge_files(L, F3, NFs, InEtc3, Ts3, Ts2, InEtc2, T3, Last).
rumerge_files(L, F2, Fs, InEtc2, Ts2, [?OBJ(T, _BT) | Ts], InEtc, T2, Last)
when T == Last ->
rumerge_files(L, F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Last);
rumerge_files(L, F2, Fs, InEtc2, Ts2, [?OBJ(T, BT) | Ts], InEtc, T2, _Last)
when T >= T2 ->
rumerge_files([BT | L], F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, T);
rumerge_files(L, F2, Fs, _InEtc2, _Ts2, [], InEtc, _T2, Last) ->
{InEtc, [F2 | Fs], L, {last, Last}};
rumerge_files(L, _F2, Fs, InEtc2, Ts2, Ts, InEtc, _T2, Last) ->
[F3 | NFs] = rinsert([Ts | InEtc], Fs),
[[?OBJ(T3,_BT3) | _] = Ts3 | InEtc3] = F3,
rumerge_files(L, F3, NFs, InEtc3, Ts3, Ts2, InEtc2, T3, Last).
merge_files(L, F2, Fs, InEtc2, BT2, Ts2, [?OBJ(T, BT) | Ts], InEtc, T2)
when T =< T2 ->
merge_files([BT | L], F2, Fs, InEtc2, BT2, Ts2, Ts, InEtc, T2);
merge_files(L, F2, Fs, _InEtc2, _BT2, _Ts2, [], InEtc, _T2) ->
{InEtc, [F2 | Fs], L, {last, foo}};
merge_files(L, _F2, Fs, InEtc2, BT2, Ts2, Ts, InEtc, _T2) ->
L1 = [BT2 | L],
[F3 | NFs] = insert([Ts | InEtc], Fs),
[[?OBJ(T3,BT3) | Ts3] | InEtc3] = F3,
merge_files(L1, F3, NFs, InEtc3, BT3, Ts3, Ts2, InEtc2, T3).
rmerge_files(L, F2, Fs, InEtc2, BT2, Ts2, [?OBJ(T, BT) | Ts], InEtc, T2)
when T >= T2 ->
rmerge_files([BT | L], F2, Fs, InEtc2, BT2, Ts2, Ts, InEtc, T2);
rmerge_files(L, F2, Fs, _InEtc2, _BT2, _Ts2, [], InEtc, _T2) ->
{InEtc, [F2 | Fs], L, {last, foo}};
rmerge_files(L, _F2, Fs, InEtc2, BT2, Ts2, Ts, InEtc, _T2) ->
L1 = [BT2 | L],
[F3 | NFs] = rinsert([Ts | InEtc], Fs),
[[?OBJ(T3,BT3) | Ts3] | InEtc3] = F3,
rmerge_files(L1, F3, NFs, InEtc3, BT3, Ts3, Ts2, InEtc2, T3).
ukmerge_files(L, F2, Fs, InEtc2, Ts2, [?OBJ(?SK(T, _I),_BT) | Ts], InEtc,
T2, Kp, Last) when T == Last ->
ukmerge_files(L, F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Kp, Last);
ukmerge_files(L, F2, Fs, InEtc2, Ts2, [?OBJ(?SK(T0,_I)=T,BT) | Ts], InEtc,
T2, Kp, _Last) when T =< T2 ->
ukmerge_files([BT | L], F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Kp, T0);
ukmerge_files(L, F2, Fs, _InEtc2, _Ts2, [], InEtc, _T2, _Kp, Last) ->
{InEtc, [F2 | Fs], L, {last, Last}};
ukmerge_files(L, _F2, Fs, InEtc2, Ts2, Ts, InEtc, _T2, Kp, Last) ->
[F3 | NFs] = insert([Ts | InEtc], Fs),
[[?OBJ(T3,_BT3) | _] = Ts3 | InEtc3] = F3,
ukmerge_files(L, F3, NFs, InEtc3, Ts3, Ts2, InEtc2, T3, Kp, Last).
rukmerge_files(L, F2, Fs, InEtc2, Ts2, [?OBJ(?SK(T, _I), BT) | Ts], InEtc,
T2, Kp, Last, _LastBT) when T == Last ->
rukmerge_files(L, F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Kp, T, BT);
rukmerge_files(L, F2, Fs, InEtc2, Ts2, [?OBJ(?SK(T0, _I)=T, BT) | Ts], InEtc,
T2, Kp, _Last, LastBT) when T >= T2 ->
rukmerge_files([LastBT|L], F2, Fs, InEtc2, Ts2, Ts, InEtc, T2, Kp, T0,BT);
rukmerge_files(L, F2, Fs, _InEtc2, _Ts2, [], InEtc, _T2, _Kp, Last, LastBT) ->
{InEtc, [F2 | Fs], L, {last, {Last, LastBT}}};
rukmerge_files(L, _F2, Fs, InEtc2, Ts2, Ts, InEtc, _T2, Kp, Last, LastBT) ->
[F3 | NFs] = rinsert([Ts | InEtc], Fs),
[[?OBJ(T3,_BT3) | _] = Ts3 | InEtc3] = F3,
rukmerge_files(L, F3, NFs, InEtc3, Ts3, Ts2, InEtc2, T3, Kp, Last,LastBT).
ucmerge_files(L, F2, Fs, InEtc2, Ts2, I2, [?OBJ(T, BT) | Ts] = Ts0, I,
InEtc, T2, CF, Last) when I < I2 ->
case CF(T, T2) of
true -> % T =< T2
case CF(T, Last) of
true ->
ucmerge_files(L, F2, Fs, InEtc2, Ts2, I2, Ts, I, InEtc, T2,
CF, Last);
false ->
ucmerge_files([BT | L], F2, Fs, InEtc2, Ts2, I2, Ts, I,
InEtc, T2, CF, T)
end;
false -> % T > T2
[F3 | NFs] = cinsert([Ts0 | InEtc], Fs, CF),
[[?OBJ(T3,_BT3) | _] = Ts3 | {I3,_} = InEtc3] = F3,
ucmerge_files(L, F3, NFs, InEtc3, Ts3, I3, Ts2, I2, InEtc2, T3, CF, Last)
end;
ucmerge_files(L, F2, Fs, InEtc2, Ts2, I2, [?OBJ(T, BT) | Ts] = Ts0, I,
InEtc, T2, CF, Last) -> % when I2 < I
case CF(T2, T) of
true -> % T2 =< T
[F3 | NFs] = cinsert([Ts0 | InEtc], Fs, CF),
[[?OBJ(T3,_BT3) | _] = Ts3 | {I3,_} = InEtc3] = F3,
ucmerge_files(L, F3, NFs, InEtc3, Ts3, I3, Ts2, I2, InEtc2, T3,
CF, Last);
false -> % T < T2
case CF(T, Last) of
true ->
ucmerge_files(L, F2, Fs, InEtc2, Ts2, I2, Ts, I, InEtc, T2,
CF, Last);
false ->
ucmerge_files([BT | L], F2, Fs, InEtc2, Ts2, I2, Ts, I,
InEtc, T2, CF, T)
end
end;
ucmerge_files(L, F2, Fs, _InEtc2, _Ts2, _I2, [], _I, InEtc, _T2, _CF, Last) ->
{InEtc, [F2 | Fs], L, {last, Last}}.
cmerge_files(L, F2, Fs, InEtc2, BT2, Ts2, I2, [?OBJ(T, BT) | Ts] = Ts0, I,
InEtc, T2, CF) when I < I2 ->
case CF(T, T2) of
true -> % T =< T2
cmerge_files([BT|L], F2, Fs, InEtc2, BT2, Ts2, I2, Ts, I, InEtc, T2, CF);
false -> % T > T2
L1 = [BT2 | L],
[F3 | NFs] = cinsert([Ts0 | InEtc], Fs, CF),
[[?OBJ(T3,BT3) | Ts3] | {I3,_} = InEtc3] = F3,
cmerge_files(L1, F3, NFs, InEtc3, BT3, Ts3, I3, Ts2, I2, InEtc2, T3, CF)
end;
cmerge_files(L, F2, Fs, InEtc2, BT2, Ts2, I2, [?OBJ(T, BT) | Ts] = Ts0, I,
InEtc, T2, CF) -> % when I2 < I
case CF(T2, T) of
true -> % T2 =< T
L1 = [BT2 | L],
[F3 | NFs] = cinsert([Ts0 | InEtc], Fs, CF),
[[?OBJ(T3,BT3) | Ts3] | {I3,_} = InEtc3] = F3,
cmerge_files(L1, F3, NFs, InEtc3, BT3, Ts3, I3, Ts2, I2, InEtc2, T3, CF);
false -> % T < T2
cmerge_files([BT|L], F2, Fs, InEtc2, BT2, Ts2, I2, Ts, I, InEtc, T2, CF)
end;
cmerge_files(L, F2, Fs, _InEtc2, _BT2, _Ts2, _I2, [], _I, InEtc, _T2, _CF) ->
{InEtc, [F2 | Fs], L, {last, foo}}.
last_file(Ts, L, {last, T}, {ukmerge,_}, _W) ->
kulast_file(Ts, T, L);
last_file(Ts, L, {last, {T,BT}}, {rukmerge,_}, _W) ->
ruklast_file(Ts, T, BT, L);
last_file(Ts, L, {last, T}, {ucmerge,CF}, _W) ->
uclast_file(Ts, T, CF, L);
last_file(Ts, L, {last, T}, _Kind, #w{unique = true}) ->
ulast_file(Ts, T, L);
last_file(Ts, L, _Last, _Kind, _W) ->
last_file(Ts, L).
ulast_file([?OBJ(T, _BT) | Ts], Last, L) when Last == T ->
last_file(Ts, L);
ulast_file(Ts, _Last, L) ->
last_file(Ts, L).
kulast_file([?OBJ(?SK(T, _I), _BT) | Ts], Last, L) when Last == T ->
last_file(Ts, L);
kulast_file(Ts, _Last, L) ->
last_file(Ts, L).
ruklast_file([?OBJ(?SK(T, _I), BT) | Ts], Last, _LastBT, L) when Last == T ->
last_file(Ts, [BT | L]);
ruklast_file(Ts, _Last, LastBT, L) ->
last_file(Ts, [LastBT | L]).
uclast_file([?OBJ(T, BT) | Ts], Last, CF, L) ->
case CF(T, Last) of
true ->
last_file(Ts, L);
false ->
last_file(Ts, [BT | L])
end.
last_file([?OBJ(_Ta, BTa), ?OBJ(_Tb, BTb) | Ts], L) ->
last_file(Ts, [BTb, BTa | L]);
last_file([?OBJ(_T, BT) | Ts], L) ->
last_file(Ts, [BT | L]);
last_file([], L) ->
L.
%% OK for 16 files.
insert(A, [X1, X2, X3, X4 | Xs]) when A > X4 ->
[X1, X2, X3, X4 | insert(A, Xs)];
insert(A, [X1, X2, X3 | T]) when A > X3 ->
[X1, X2, X3, A | T];
insert(A, [X1, X2 | Xs]) when A > X2 ->
[X1, X2, A | Xs];
insert(A, [X1 | T]) when A > X1 ->
[X1, A | T];
insert(A, Xs) ->
[A | Xs].
rinsert(A, [X1, X2, X3, X4 | Xs]) when A < X4 ->
[X1, X2, X3, X4 | rinsert(A, Xs)];
rinsert(A, [X1, X2, X3 | T]) when A < X3 ->
[X1, X2, X3, A | T];
rinsert(A, [X1, X2 | Xs]) when A < X2 ->
[X1, X2, A | Xs];
rinsert(A, [X1 | T]) when A < X1 ->
[X1, A | T];
rinsert(A, Xs) ->
[A | Xs].
-define(CINSERT(F, A, T1, T2),
case cfun(CF, F, A) of
true -> [F, A | T2];
false -> [A | T1]
end).
cinsert(A, [F1 | [F2 | [F3 | [F4 | Fs]=T4]=T3]=T2]=T1, CF) ->
case cfun(CF, F4, A) of
true -> [F1, F2, F3, F4 | cinsert(A, Fs, CF)];
false ->
case cfun(CF, F2, A) of
true -> [F1, F2 | ?CINSERT(F3, A, T3, T4)];
false -> ?CINSERT(F1, A, T1, T2)
end
end;
cinsert(A, [F1 | [F2 | Fs]=T2]=T1, CF) ->
case cfun(CF, F2, A) of
true -> [F1, F2 | cinsert(A, Fs, CF)];
false -> ?CINSERT(F1, A, T1, T2)
end;
cinsert(A, [F | Fs]=T, CF) ->
?CINSERT(F, A, T, Fs);
cinsert(A, _, _CF) ->
[A].
%% Inlined.
cfun(CF, F1, F2) ->
[[?OBJ(T1,_) | _] | {I1,_}] = F1,
[[?OBJ(T2,_) | _] | {I2,_}] = F2,
if
I1 < I2 ->
CF(T1, T2);
true -> % I2 < I1
not CF(T2, T1)
end.
binterm_objects([?OBJ(_T, [_Sz | BT]) | Ts], L) ->
binterm_objects(Ts, [BT | L]);
binterm_objects([], L) ->
L.
objects([[_Sz | BT] | Ts], L) ->
objects(Ts, [BT | L]);
objects([], L) ->
L.
binterms([?OBJ(_T1, BT1), ?OBJ(_T2, BT2) | Ts], L) ->
binterms(Ts, [BT2, BT1 | L]);
binterms([?OBJ(_T, BT) | Ts], L) ->
binterms(Ts, [BT | L]);
binterms([], L) ->
L.
read_chunk(InEtc, Fs, L, LSz, Last, W) ->
{I, IFun} = InEtc,
case read_more(IFun, I, LSz, W) of
{Ts, NLSz, NIFun, #w{order = ascending}=NW} ->
NInEtc = {I, NIFun},
NFs = insert([Ts | NInEtc], Fs),
merge_files(NFs, L, NLSz, Last, NW);
{Ts, NLSz, NIFun, #w{order = descending}=NW} ->
NInEtc = {I, NIFun},
NFs = rinsert([Ts | NInEtc], Fs),
merge_files(NFs, L, NLSz, Last, NW);
{Ts, NLSz, NIFun, NW} ->
NInEtc = {I, NIFun},
NFs = cinsert([Ts | NInEtc], Fs, NW#w.order),
merge_files(NFs, L, NLSz, Last, NW);
{eof, NW} ->
merge_files(Fs, L, LSz, Last, NW)
end.
%% -> {[{term() | binary()}], NewLSz, NewIFun, NewW} | eof | throw(Error)
read_more(IFun, I, LSz, W) ->
case IFun({{merge, I}, [], LSz, W}) of
{{_, [], NLSz}, NIFun} ->
read_more(NIFun, I, NLSz, W);
{{_, L, NLSz}, NInFun} ->
NW = case lists:member(IFun, W#w.temp) of
true ->
%% temporary file
W#w{temp = [NInFun | lists:delete(IFun, W#w.temp)]};
false ->
%% input file
W
end,
{lists:reverse(L), NLSz, NInFun, NW};
eof ->
%% already closed.
NW = W#w{temp = lists:delete(IFun, W#w.temp)},
{eof, NW}
end.
read_fun(FileName, Owner, W) ->
case file:open(FileName, [raw, binary, read, compressed]) of
{ok, Fd} ->
read_fun2(Fd, <<>>, 0, FileName, Owner);
Error ->
file_error(FileName, Error, W)
end.
read_fun2(Fd, Bin, Size, FileName, Owner) ->
fun(close) ->
close_read_fun(Fd, FileName, Owner);
({I, L, LSz, W}) ->
case read_objs(Fd, FileName, I, L, Bin, Size, LSz, W) of
{{I1, L1, Bin1, Size1}, LSz1} ->
NIFun = read_fun2(Fd, Bin1, Size1, FileName, Owner),
{{I1, L1, LSz1}, NIFun};
eof ->
close_read_fun(Fd, FileName, Owner),
eof
end
end.
close_read_fun(Fd, _FileName, user) ->
_ = file:close(Fd),
ok;
close_read_fun(Fd, FileName, fsort) ->
_ = file:close(Fd),
_ = file:delete(FileName),
ok.
read_objs(Fd, FileName, I, L, Bin0, Size0, LSz, W) ->
Max = erlang:max(Size0, ?CHUNKSIZE),
BSz0 = byte_size(Bin0),
Min = Size0 - BSz0 + W#w.hdlen, % Min > 0
NoBytes = erlang:max(Min, Max),
case read(Fd, FileName, NoBytes, W) of
{ok, Bin} ->
BSz = byte_size(Bin),
NLSz = LSz + BSz,
case catch file_loop(L, I, Bin0, Bin, Size0, BSz0, BSz, Min, W)
of
{'EXIT', _R} ->
error({error, {bad_object, FileName}}, W);
Reply ->
{Reply, NLSz}
end;
eof when byte_size(Bin0) =:= 0 ->
eof;
eof ->
error({error, {premature_eof, FileName}}, W)
end.
file_loop(L, I, _B1, B2, Sz, 0, _B2Sz, _Min, W) ->
file_loop(L, I, B2, Sz, W);
file_loop(L, I, B1, B2, Sz, _B1Sz, B2Sz, Min, W) when B2Sz > Min ->
{B3, B4} = split_binary(B2, Min),
{I1, L1, <<>>, Sz1} = file_loop(L, I, list_to_binary([B1, B3]), Sz, W),
file_loop(L1, I1, B4, Sz1, W);
file_loop(L, I, B1, B2, Sz, _B1Sz, _B2Sz, _Min, W) ->
file_loop(L, I, list_to_binary([B1, B2]), Sz, W).
file_loop(L, I, B, Sz, W) ->
#w{keypos = Kp, format = Format, hdlen = HdLen} = W,
file_loop1(L, I, B, Sz, Kp, Format, HdLen).
file_loop1(L, I, HB, 0, Kp, F, HdLen) ->
<<Size:HdLen/unit:8, B/binary>> = HB,
file_loop2(L, I, B, Size, <<Size:HdLen/unit:8>>, Kp, F, HdLen);
file_loop1(L, I, B, Sz, Kp, F, HdLen) ->
file_loop2(L, I, B, Sz, <<Sz:HdLen/unit:8>>, Kp, F, HdLen).
file_loop2(L, _I, B, Sz, SzB, 0, binary, HdLen) ->
{NL, NB, NSz, NSzB} = file_binloop(L, Sz, SzB, B, HdLen),
if
byte_size(NB) =:= NSz ->
<<Bin:NSz/binary>> = NB,
{0, [?OBJ(Bin, [NSzB | Bin]) | NL], <<>>, 0};
true ->
{0, NL, NB, NSz}
end;
file_loop2(L, _I, B, Sz, SzB, 0, Fun, HdLen) ->
file_binterm_loop(L, Sz, SzB, B, Fun, HdLen);
file_loop2(L, {merge, I}, B, Sz, SzB, Kp, Fun, HdLen) -> % when Kp =/= 0
merge_loop(Kp, I, L, Sz, SzB, B, Fun, HdLen);
file_loop2(L, I, B, Sz, SzB, Kp, Fun, HdLen) when is_integer(I) ->
key_loop(Kp, I, L, Sz, SzB, B, Fun, HdLen).
file_binloop(L, Size, SizeB, B, HL) ->
case B of
<<Bin:Size/binary, NSizeB:HL/binary, R/binary>> ->
<<NSize:HL/unit:8>> = NSizeB,
file_binloop([?OBJ(Bin, [SizeB | Bin]) | L], NSize, NSizeB, R, HL);
_ ->
{L, B, Size, SizeB}
end.
file_binterm_loop(L, Size, SizeB, B, Fun, HL) ->
case B of
<<BinTerm:Size/binary, NSizeB:HL/binary, R/binary>> ->
<<NSize:HL/unit:8>> = NSizeB,
BT = [SizeB | BinTerm],
Term = Fun(BinTerm),
file_binterm_loop([?OBJ(Term, BT) | L], NSize, NSizeB, R, Fun, HL);
<<BinTerm:Size/binary>> ->
Term = Fun(BinTerm),
NL = [?OBJ(Term, [SizeB | BinTerm]) | L],
{0, NL, <<>>, 0};
_ ->
{0, L, B, Size}
end.
key_loop(KeyPos, I, L, Size, SizeB, B, Fun, HL) ->
case B of
<<BinTerm:Size/binary, NSizeB:HL/binary, R/binary>> ->
<<NSize:HL/unit:8>> = NSizeB,
BT = [SizeB | BinTerm],
UniqueKey = make_key(KeyPos, Fun(BinTerm)),
E = ?OBJ(UniqueKey, BT),
key_loop(KeyPos, I+1, [E | L], NSize, NSizeB, R, Fun, HL);
<<BinTerm:Size/binary>> ->
UniqueKey = make_key(KeyPos, Fun(BinTerm)),
NL = [?OBJ(UniqueKey, [SizeB | BinTerm]) | L],
{I+1, NL, <<>>, 0};
_ ->
{I, L, B, Size}
end.
merge_loop(KeyPos, I, L, Size, SizeB, B, Fun, HL) ->
case B of
<<BinTerm:Size/binary, NSizeB:HL/binary, R/binary>> ->
<<NSize:HL/unit:8>> = NSizeB,
BT = [SizeB | BinTerm],
UniqueKey = make_stable_key(KeyPos, I, Fun(BinTerm)),
E = ?OBJ(UniqueKey, BT),
merge_loop(KeyPos, I, [E | L], NSize, NSizeB, R, Fun, HL);
<<BinTerm:Size/binary>> ->
UniqueKey = make_stable_key(KeyPos, I, Fun(BinTerm)),
NL = [?OBJ(UniqueKey, [SizeB | BinTerm]) | L],
{{merge, I}, NL, <<>>, 0};
_ ->
{{merge, I}, L, B, Size}
end.
fun_objs(Objs, L, LSz, NoBytes, I, W) ->
#w{keypos = Keypos, format = Format, hdlen = HL} = W,
case catch fun_loop(Objs, L, LSz, NoBytes, I, Keypos, Format, HL) of
{'EXIT', _R} ->
error({error, bad_object}, W);
Reply ->
Reply
end.
fun_loop(Objs, L, LSz, RunSize, _I, 0, binary, HdLen) ->
fun_binloop(Objs, L, LSz, RunSize, HdLen);
fun_loop(Objs, L, LSz, RunSize, _I, 0, Fun, HdLen) ->
fun_loop(Objs, L, LSz, RunSize, Fun, HdLen);
fun_loop(Objs, L, LSz, RunSize, {merge, I}, Keypos, Fun, HdLen) ->
fun_mergeloop(Objs, L, LSz, RunSize, I, Keypos, Fun, HdLen);
fun_loop(Objs, L, LSz, RunSize, I, Keypos, Fun, HdLen) when is_integer(I) ->
fun_keyloop(Objs, L, LSz, RunSize, I, Keypos, Fun, HdLen).
fun_binloop([B | Bs], L, LSz, RunSize, HL) when LSz < RunSize ->
Size = byte_size(B),
Obj = ?OBJ(B, [<<Size:HL/unit:8>> | B]),
fun_binloop(Bs, [Obj | L], LSz+Size, RunSize, HL);
fun_binloop(Bs, L, LSz, _RunSize, _HL) ->
{0, Bs, L, LSz}.
fun_loop([B | Bs], L, LSz, RunSize, Fun, HL) when LSz < RunSize ->
Size = byte_size(B),
Obj = ?OBJ(Fun(B), [<<Size:HL/unit:8>> | B]),
fun_loop(Bs, [Obj | L], LSz+Size, RunSize, Fun, HL);
fun_loop(Bs, L, LSz, _RunSize, _Fun, _HL) ->
{0, Bs, L, LSz}.
fun_keyloop([B | Bs], L, LSz, RunSize, I, Kp, Fun, HL) when LSz < RunSize ->
Size = byte_size(B),
UniqueKey = make_key(Kp, Fun(B)),
E = ?OBJ(UniqueKey, [<<Size:HL/unit:8>> | B]),
fun_keyloop(Bs, [E | L], LSz+Size, RunSize, I+1, Kp, Fun, HL);
fun_keyloop(Bs, L, LSz, _RunSize, I, _Kp, _Fun, _HL) ->
{I, Bs, L, LSz}.
fun_mergeloop([B | Bs], L, LSz, RunSize, I, Kp, Fun, HL) when LSz < RunSize ->
Size = byte_size(B),
UniqueKey = make_stable_key(Kp, I, Fun(B)),
E = ?OBJ(UniqueKey, [<<Size:HL/unit:8>> | B]),
fun_mergeloop(Bs, [E | L], LSz+Size, RunSize, I, Kp, Fun, HL);
fun_mergeloop(Bs, L, LSz, _RunSize, I, _Kp, _Fun, _HL) ->
{{merge, I}, Bs, L, LSz}. % any I would do
%% Inlined.
make_key(Kp, T) when is_integer(Kp) ->
element(Kp, T);
make_key([Kp1, Kp2], T) ->
[element(Kp1, T), element(Kp2, T)];
make_key([Kp1, Kp2 | Kps], T) ->
[element(Kp1, T), element(Kp2, T) | make_key2(Kps, T)].
%% Inlined.
%% A sequence number (I) is used for making the internal sort stable.
%% I is ordering number of the file from which T was read.
make_stable_key(Kp, I, T) when is_integer(Kp) ->
?SK(element(Kp, T), I);
make_stable_key([Kp1, Kp2], I, T) ->
?SK([element(Kp1, T) | element(Kp2, T)], I);
make_stable_key([Kp1, Kp2 | Kps], I, T) ->
?SK([element(Kp1, T), element(Kp2, T) | make_key2(Kps, T)], I).
make_key2([Kp], T) ->
[element(Kp, T)];
make_key2([Kp | Kps], T) ->
[element(Kp, T) | make_key2(Kps, T)].
infun(W) ->
W1 = W#w{in = undefined},
try (W#w.in)(read) of
end_of_input ->
{end_of_input, W1};
{end_of_input, Value} ->
{end_of_input, W1#w{inout_value = {value, Value}}};
{Objs, NFun} when is_function(NFun),
is_function(NFun, 1),
is_list(Objs) ->
{cont, W#w{in = NFun}, Objs};
Error ->
error(Error, W1)
catch Class:Reason ->
cleanup(W1),
erlang:raise(Class, Reason, erlang:get_stacktrace())
end.
outfun(A, #w{inout_value = Val} = W) when Val =/= no_value ->
W1 = W#w{inout_value = no_value},
W2 = if
W1#w.fun_out ->
outfun(Val, W1);
true -> W1
end,
outfun(A, W2);
outfun(A, W) ->
W1 = W#w{out = undefined},
try (W#w.out)(A) of
Reply when A =:= close ->
Reply;
NF when is_function(NF), is_function(NF, 1) ->
W#w{out = NF};
Error ->
error(Error, W1)
catch Class:Reason ->
cleanup(W1),
erlang:raise(Class, Reason, erlang:get_stacktrace())
end.
is_keypos(Keypos) when is_integer(Keypos), Keypos > 0 ->
true;
is_keypos([]) ->
{badarg, []};
is_keypos(L) ->
is_keyposs(L).
is_keyposs([Kp | Kps]) when is_integer(Kp), Kp > 0 ->
is_keyposs(Kps);
is_keyposs([]) ->
true;
is_keyposs([Bad | _]) ->
{badarg, Bad};
is_keyposs(Bad) ->
{badarg, Bad}.
is_input(Fun) when is_function(Fun), is_function(Fun, 1) ->
{true, Fun};
is_input(Files) ->
is_files(Files).
is_files(Fs) ->
is_files(Fs, []).
is_files([F | Fs], L) ->
case read_file_info(F) of
{ok, File, _FI} ->
is_files(Fs, [File | L]);
Error ->
Error
end;
is_files([], L) ->
{true, lists:reverse(L)};
is_files(Bad, _L) ->
{badarg, Bad}.
maybe_output(Fun) when is_function(Fun), is_function(Fun, 1) ->
{true, Fun};
maybe_output(File) ->
case read_file_info(File) of
{badarg, _File} = Badarg ->
Badarg;
{ok, FileName, _FileInfo} ->
{true, FileName};
{error, {file_error, FileName, _Reason}} ->
{true, FileName}
end.
read_file_info(File) ->
%% Absolute names in case some process should call file:set_cwd/1.
case catch filename:absname(File) of
{'EXIT', _} ->
{badarg, File};
FileName ->
case file:read_file_info(FileName) of
{ok, FileInfo} ->
{ok, FileName, FileInfo};
{error, einval} ->
{badarg, File};
{error, Reason} ->
{error, {file_error, FileName, Reason}}
end
end.
%% No attempt is made to avoid overwriting existing files.
next_temp(W) ->
Seq = W#w.seq,
NW = W#w{seq = Seq + 1},
Temp = lists:concat([W#w.prefix, Seq]),
{NW, Temp}.
%% Would use the temporary directory (TMP|TEMP|TMPDIR), were it
%% readily accessible.
tmp_prefix(F, TmpDirOpt) when is_function(F); F =:= undefined ->
{ok, CurDir} = file:get_cwd(),
tmp_prefix1(CurDir, TmpDirOpt);
tmp_prefix(OutFile, TmpDirOpt) ->
Dir = filename:dirname(OutFile),
tmp_prefix1(Dir, TmpDirOpt).
tmp_prefix1(Dir, TmpDirOpt) ->
U = "_",
Node = node(),
Pid = os:getpid(),
Unique = erlang:unique_integer([positive]),
F = lists:concat(["fs_",Node,U,Pid,U,Unique,"."]),
TmpDir = case TmpDirOpt of
default ->
Dir;
{dir, TDir} ->
TDir
end,
filename:join(filename:absname(TmpDir), F).
%% -> {Fd, NewW} | throw(Error)
open_file(FileName, W) ->
case file:open(FileName, W#w.z ++ [raw, binary, write]) of
{ok, Fd} ->
{Fd, W#w{temp = [{Fd,FileName} | W#w.temp]}};
Error ->
file_error(FileName, Error, W)
end.
read(Fd, FileName, N, W) ->
case file:read(Fd, N) of
{ok, Bin} ->
{ok, Bin};
eof ->
eof;
{error, enomem} ->
%% Bad N
error({error, {bad_object, FileName}}, W);
{error, einval} ->
%% Bad N
error({error, {bad_object, FileName}}, W);
Error ->
file_error(FileName, Error, W)
end.
write(Fd, FileName, B, W) ->
case file:write(Fd, B) of
ok ->
ok;
Error ->
file_error(FileName, Error, W)
end.
-spec file_error(_, {'error',atom()}, #w{}) -> no_return().
file_error(File, {error, Reason}, W) ->
error({error, {file_error, File, Reason}}, W).
error(Error, W) ->
cleanup(W),
throw({W#w.ref, Error}).
cleanup(W) ->
close_out(W),
W1 = close_input(W),
F = fun(IFun) when is_function(IFun) ->
IFun(close);
({Fd,FileName}) ->
_ = file:close(Fd),
_= file:delete(FileName);
(FileName) ->
_= file:delete(FileName)
end,
lists:foreach(F, W1#w.temp).
close_input(#w{in = In}=W) when is_function(In) ->
catch In(close),
W#w{in = undefined};
close_input(#w{in = undefined}=W) ->
W.
close_out(#w{out = Out}) when is_function(Out) ->
catch Out(close);
close_out(_) ->
ok.
close_file(Fd, W) ->
{Fd, FileName} = lists:keyfind(Fd, 1, W#w.temp),
?DEBUG("closing ~tp~n", [FileName]),
case file:close(Fd) of
ok ->
W#w{temp = [FileName | lists:keydelete(Fd, 1, W#w.temp)]};
Error ->
file_error(FileName, Error, W)
end.
%%%
%%% Format 'term'.
%%%
file_rterms(no_file, Files) ->
fun(close) ->
ok;
(read) when Files =:= [] ->
end_of_input;
(read) ->
[F | Fs] = Files,
case file:open(F, [read, compressed]) of
{ok, Fd} ->
file_rterms2(Fd, [], 0, F, Fs);
{error, Reason} ->
{error, {file_error, F, Reason}}
end
end;
file_rterms({Fd, FileName}, Files) ->
fun(close) ->
file:close(Fd);
(read) ->
file_rterms2(Fd, [], 0, FileName, Files)
end.
file_rterms2(Fd, L, LSz, FileName, Files) when LSz < ?CHUNKSIZE ->
case io:read(Fd, '') of
{ok, Term} ->
B = term_to_binary(Term),
file_rterms2(Fd, [B | L], LSz + byte_size(B), FileName, Files);
eof ->
_ = file:close(Fd),
{lists:reverse(L), file_rterms(no_file, Files)};
_Error ->
_ = file:close(Fd),
{error, {bad_term, FileName}}
end;
file_rterms2(Fd, L, _LSz, FileName, Files) ->
{lists:reverse(L), file_rterms({Fd, FileName}, Files)}.
file_wterms(W, F, Args) ->
fun(close) when W =:= name ->
ok;
(close) ->
{fd, Fd} = W,
file:close(Fd);
(L) when W =:= name ->
case file:open(F, Args) of
{ok, Fd} ->
write_terms(Fd, F, L, Args);
{error, Reason} ->
{error, {file_error, F, Reason}}
end;
(L) ->
{fd, Fd} = W,
write_terms(Fd, F, L, Args)
end.
write_terms(Fd, F, [B | Bs], Args) ->
case io:request(Fd, {format, "~p.~n", [binary_to_term(B)]}) of
ok ->
write_terms(Fd, F, Bs, Args);
{error, Reason} ->
_ = file:close(Fd),
{error, {file_error, F, Reason}}
end;
write_terms(Fd, F, [], Args) ->
file_wterms({fd, Fd}, F, Args).
fun_rterms(InFun) ->
fun(close) ->
InFun(close);
(read) ->
case InFun(read) of
{Ts, NInFun} when is_list(Ts),
is_function(NInFun),
is_function(NInFun, 1) ->
{to_bin(Ts, []), fun_rterms(NInFun)};
Else ->
Else
end
end.
fun_wterms(OutFun) ->
fun(close) ->
OutFun(close);
(L) ->
case OutFun(wterms_arg(L)) of
NOutFun when is_function(NOutFun), is_function(NOutFun, 1) ->
fun_wterms(NOutFun);
Else ->
Else
end
end.
to_bin([E | Es], L) ->
to_bin(Es, [term_to_binary(E) | L]);
to_bin([], L) ->
lists:reverse(L).
wterms_arg(L) when is_list(L) ->
to_term(L, []);
wterms_arg(Value) ->
Value.
to_term([B | Bs], L) ->
to_term(Bs, [binary_to_term(B) | L]);
to_term([], L) ->
lists:reverse(L).
