I need to read the last line in some very large textfiles (to get the timestamp from the data). TStringlist would be a simple approach but it returns an out of memory error. I'm trying to use seek and blockread, but the characters in the buffer are all nonsense. Is this something to do with unicode?
Function TForm1.ReadLastLine2(FileName: String): String;
var
FileHandle: File;
s,line: string;
ok: 0..1;
Buf: array[1..8] of Char;
k: longword;
i,ReadCount: integer;
begin
AssignFile (FileHandle,FileName);
Reset (FileHandle); // or for binary files: Reset (FileHandle,1);
ok := 0;
k := FileSize (FileHandle);
Seek (FileHandle, k-1);
s := '';
while ok<>1 do begin
BlockRead (FileHandle, buf, SizeOf(Buf)-1, ReadCount); //BlockRead ( var FileHandle : File; var Buffer; RecordCount : Integer {; var RecordsRead : Integer} ) ;
if ord (buf[1]) <>13 then //Arg to integer
s := s + buf[1]
else
ok := ok + 1;
k := k-1;
seek (FileHandle,k);
end;
CloseFile (FileHandle);
// Reverse the order in the line read
setlength (line,length(s));
for i:=1 to length(s) do
line[length(s) - i+1 ] := s[i];
Result := Line;
end;
Based on www.delphipages.com/forum/showthread.php?t=102965
The testfile is a simple CSV I created in excel ( this is not the 100MB I ultimately need to read).
a,b,c,d,e,f,g,h,i,j,blank
A,B,C,D,E,F,G,H,I,J,blank
1,2,3,4,5,6,7,8,9,0,blank
Mary,had,a,little,lamb,His,fleece,was,white,as,snow
And,everywhere,that,Mary,went,The,lamb,was,sure,to,go
You really have to read the file in LARGE chunks from the tail to the head.
Since it is so large it does not fit the memory - then reading it line by line from start to end would be very slow. With ReadLn - twice slow.
You also has to be ready that the last line might end with EOL or may not.
Personally I would also account for three possible EOL sequences:
If you are sure your CSV files would only ever be generated by native Windows programs it would be safe to assume full CR/LF be used. But if there can be other Java programs, non-Windows platforms, mobile programs - I would be less sure. Of course pure CR without LF would be the least probable case of them all.
uses System.IOUtils, System.Math, System.Classes;
type FileChar = AnsiChar; FileString = AnsiString; // for non-Unicode files
// type FileChar = WideChar; FileString = UnicodeString;// for UTF16 and UCS-2 files
const FileCharSize = SizeOf(FileChar);
// somewhere later in the code add: Assert(FileCharSize = SizeOf(FileString[1]);
function ReadLastLine(const FileName: String): FileString; overload; forward;
const PageSize = 4*1024;
// the minimal read atom of most modern HDD and the memory allocation atom of Win32
// since the chances your file would have lines longer than 4Kb are very small - I would not increase it to several atoms.
function ReadLastLine(const Lines: TStringDynArray): FileString; overload;
var i: integer;
begin
Result := '';
i := High(Lines);
if i < Low(Lines) then exit; // empty array - empty file
Result := Lines[i];
if Result > '' then exit; // we got the line
Dec(i); // skip the empty ghost line, in case last line was CRLF-terminated
if i < Low(Lines) then exit; // that ghost was the only line in the empty file
Result := Lines[i];
end;
// scan for EOLs in not-yet-scanned part
function FindLastLine(buffer: TArray<FileChar>; const OldRead : Integer;
const LastChunk: Boolean; out Line: FileString): boolean;
var i, tailCRLF: integer; c: FileChar;
begin
Result := False;
if Length(Buffer) = 0 then exit;
i := High(Buffer);
tailCRLF := 0; // test for trailing CR/LF
if Buffer[i] = ^J then begin // LF - single, or after CR
Dec(i);
Inc(tailCRLF);
end;
if (i >= Low(Buffer)) and (Buffer[i] = ^M) then begin // CR, alone or before LF
Inc(tailCRLF);
end;
i := High(Buffer) - Max(OldRead, tailCRLF);
if i - Low(Buffer) < 0 then exit; // no new data to read - results would be like before
if OldRead > 0 then Inc(i); // the CR/LF pair could be sliced between new and previous buffer - so need to start a bit earlier
for i := i downto Low(Buffer) do begin
c := Buffer[i];
if (c=^J) or (c=^M) then begin // found EOL
SetString( Line, @Buffer[i+1], High(Buffer) - tailCRLF - i);
exit(True);
end;
end;
// we did not find non-terminating EOL in the buffer (except maybe trailing),
// now we should ask for more file content, if there is still left any
// or take the entire file (without trailing EOL if any)
if LastChunk then begin
SetString( Line, @Buffer[ Low(Buffer) ], Length(Buffer) - tailCRLF);
Result := true;
end;
end;
function ReadLastLine(const FileName: String): FileString; overload;
var Buffer, tmp: TArray<FileChar>;
// dynamic arrays - eases memory management and protect from stack corruption
FS: TFileStream; FSize, NewPos: Int64;
OldRead, NewLen : Integer; EndOfFile: boolean;
begin
Result := '';
FS := TFile.OpenRead(FileName);
try
FSize := FS.Size;
if FSize <= PageSize then begin // small file, we can be lazy!
FreeAndNil(FS); // free the handle and avoid double-free in finally
Result := ReadLastLine( TFile.ReadAllLines( FileName, TEncoding.ANSI ));
// or TEncoding.UTF16
// warning - TFIle is not share-aware, if the file is being written to by another app
exit;
end;
SetLength( Buffer, PageSize div FileCharSize);
OldRead := 0;
repeat
NewPos := FSize - Length(Buffer)*FileCharSize;
EndOfFile := NewPos <= 0;
if NewPos < 0 then NewPos := 0;
FS.Position := NewPos;
FS.ReadBuffer( Buffer[Low(Buffer)], (Length(Buffer) - OldRead)*FileCharSize);
if FindLastLine(Buffer, OldRead, EndOfFile, Result) then
exit; // done !
tmp := Buffer; Buffer := nil; // flip-flop: preparing to broaden our mouth
OldRead := Length(tmp); // need not to re-scan the tail again and again when expanding our scanning range
NewLen := Min( 2*Length(tmp), FSize div FileCharSize );
SetLength(Buffer, NewLen); // this may trigger EOutOfMemory...
Move( tmp[Low(tmp)], Buffer[High(Buffer)-OldRead+1], OldRead*FileCharSize);
tmp := nil; // free old buffer
until EndOfFile;
finally
FS.Free;
end;
end;
PS. Note one extra special case - if you would use Unicode chars (two-bytes ones) and would give odd-length file (3 bytes, 5 bytes, etc) - you would never be ble to scan the starting single byte (half-widechar). Maybe you should add the extra guard there, like Assert( 0 = FS.Size mod FileCharSize)
PPS. As a rule of thumb you better keep those functions out of the form class, - because WHY mixing them? In general you should separate concerns into small blocks. Reading file has nothing with user interaction - so should better be offloaded to an extra UNIT. Then you would be able to use functions from that unit in one form or 10 forms, in main thread or in multi-threaded application. Like LEGO parts - they give you flexibility by being small and separate.
PPPS. Another approach here would be using memory-mapped files. Google for MMF implementations for Delphi and articles about benefits and problems with MMF approach. Personally I think rewriting the code above to use MMF would greatly simplify it, removing several "special cases" and the troublesome and memory copying flip-flop. OTOH it would demand you to be very strict with pointers arithmetic.