Creating/Using FileStream Thread Safe - multithreading

In my Application when I write text files (logs, traces, etc), I use TFileStream class.
There are cases that I write the data in multithreaded environment, those are the steps:
1- Write Cache Data
2- For each 1000 lines I save to File.
3- Clear Data.
This process is repeated during all processing.
Problem Description:
With 16 threads, the system throws the following exception:
Access Violation - file already in use by another application.
I guess this is happening because that the handle used by one thread is not closed yet, when another thread needs to open.
I changed the architecture to the following: (bellow is the NEW implementation)
In the previous way, the TFileStream was created with FileName and Mode parameters, and destroyed closing the handle (I wasn't using TMyFileStream)
TMyFileStream = class(TFileStream)
public
destructor Destroy; override;
end;
TLog = class(TStringList)
private
FFileHandle: Integer;
FirstTime: Boolean;
FName: String;
protected
procedure Flush;
constructor Create;
destructor Destroy;
end;
destructor TMyFileStream.Destroy;
begin
//Do Not Close the Handle, yet!
FHandle := -1;
inherited Destroy;
end;
procedure TLog.Flush;
var
StrBuf: PChar; LogFile: string;
F: TFileStream;
InternalHandle: Cardinal;
begin
if (Text <> '') then
begin
LogFile:= GetDir() + FName + '.txt';
ForceDirectories(ExtractFilePath(LogFile));
if FFileHandle < 0 then
begin
if FirstTime then
FirstTime := False;
if FileExists(LogFile) then
if not SysUtils.DeleteFile(LogFile) then
RaiseLastOSError;
InternalHandle := CreateFile(PChar(LogFile), GENERIC_READ or GENERIC_WRITE, FILE_SHARE_READ, nil, CREATE_NEW, 0,0);
if InternalHandle = INVALID_HANDLE_VALUE then
RaiseLastOSError
else if GetLastError = ERROR_ALREADY_EXISTS then
begin
InternalHandle := CreateFile(PChar(LogFile), GENERIC_READ or GENERIC_WRITE, FILE_SHARE_READ, nil, OPEN_EXISTING, 0,0);
if InternalHandle = INVALID_HANDLE_VALUE then
RaiseLastOSError
else
FFileHandle := InternalHandle;
end
else
FFileHandle := InternalHandle;
end;
F := TMyFileStream.Create(FFileHandle);
try
StrBuf := PChar(Text);
F.Position := F.Size;
F.Write(StrBuf^, StrLen(StrBuf));
finally
F.Free();
end;
Clear;
end;
end;
destructor TLog.Destroy;
begin
FUserList:= nil;
Flush;
if FFileHandle >= 0 then
CloseHandle(FFileHandle);
inherited;
end;
constructor TLog.Create;
begin
inherited;
FirstTime := True;
FFileHandle := -1;
end;
There is another better way?
Is this implementation correct?
May I improve this?
My guess about the Handle was right?
All theads use the same Log object.
There is no reentrance, i checked! there is something wrong with the TFileStream.
The Access to the Add is synchronized, I mean, I used critical session, and when it reaches 1000 lines, Flush procedure is called.
P.S: I do not want third-party component, i want to create my own.

Well, for a start, there's no point in TMyFileStream. What you are looking for is THandleStream. That class allows you to supply a file handle whose lifetime you control. And if you use THandleStream you'll be able to avoid the rather nasty hacks of your variant. That said, why are you even bothering with a stream? Replace the code that creates and uses the stream with a call to SetFilePointer to seek to the end of the file, and a call to WriteFile to write content.
However, even using that, your proposed solution requires further synchronization. A single windows file handle cannot be used concurrently from multiple threads without synchronisation. You hint in a comment (should be in the question) that you are serializing file writes. If so then you are just fine.

The threaded solution provided by Marko Paunovic quite nice, however while reviewing the code I noticed a small mistake, perhaps just an oversight in the example but I thought I'd mention it just the same in case someone actually tries to use it as-is.
There is a missing call to Flush in TLogger.Destroy, as a result any unflushed (buffered) data is disgarded when the TLogger object is destroyed.
destructor TLogger.Destroy;
begin
if FStrings.Count > 0 then
Flush;
FStrings.Free;
DeleteCriticalSection(FLock);
inherited;
end;

How about:
In each thread, add log lines to a TStringList instance until lines.count=1000. Then push the TStringList onto a blocking producer-consumer queue, immediately create a new TStringList and carry on logging to the new list.
Use one Logging thread that dequeues the TStringList instances, writes them to the file and then frees them.
This isolates the log writes from disk/network delays, removes any reliance on dodgy file-locking and will actually work reliably.

I figured MY MISTAKE.
In first place, I want to apologize for posting this stupid question without a proper way to reproduce the exception. In other words, without a SSCCE.
The problem was a control flag that my TLog class used internally.
This flag was created, when we started to evolve our product a parallel architecture.
As we needed to keep the previous form working (at least until everything was in the new architecture).
We created some flags to identify if the object was either the new or old version.
One of that flags was named CheckMaxSize.
If CheckMaxSize was enabled, at a certain moment, every data inside the instance of this object in each thread, would be thrown to the main instance, which was in the "main" thread (not the GUI one, because it was a background work). Furthermore, when CheckMaxSize is enabled, TLog should never ever call "flush".
Finally, as you can see, in TLog.Destroy there is no check to CheckMaxSize. Therefore, the problem would happen because the name of the file created by this class was always the same, since it was processing the same task, and when One object created the file and another one tried to create another file with the same name, inside the same folder, the OS (Windows) rose an Exception.
Solution:
Rewrite the destructor to:
destructor TLog.Destroy;
begin
if CheckMaxSize then
Flush;
if FFileHandle >= 0 then
CloseHandle(FFileHandle);
inherited;
end;

If you have multithreaded code that needs to write to single file, it's best to have as much control as you can in your hands. And that means, avoid classes which you are not 100% sure how they work.
I suggest that you use multiple threads > single logger architecture, where each thread will have reference to logger object, and add strings to it. Once 1000 lines are reached, logger would flush the collected data in file.
There is no need to use TFileStream to write data to file, you can
go with CreateFile()/SetFilePointer()/WriteFile(), as David already suggested
TStringList is not thread-safe, so you have to use locks on it
main.dpr:
{$APPTYPE CONSOLE}
uses
uLogger,
uWorker;
const
WORKER_COUNT = 16;
var
worker: array[0..WORKER_COUNT - 1] of TWorker;
logger: TLogger;
C1 : Integer;
begin
Write('Creating logger...');
logger := TLogger.Create('test.txt');
try
WriteLn(' OK');
Write('Creating threads...');
for C1 := Low(worker) to High(worker) do
begin
worker[C1] := TWorker.Create(logger);
worker[C1].Start;
end;
WriteLn(' OK');
Write('Press ENTER to terminate...');
ReadLn;
Write('Destroying threads...');
for C1 := Low(worker) to High(worker) do
begin
worker[C1].Terminate;
worker[C1].WaitFor;
worker[C1].Free;
end;
WriteLn(' OK');
finally
Write('Destroying logger...');
logger.Free;
WriteLn(' OK');
end;
end.
uWorker.pas:
unit uWorker;
interface
uses
System.Classes, uLogger;
type
TWorker = class(TThread)
private
FLogger: TLogger;
protected
procedure Execute; override;
public
constructor Create(const ALogger: TLogger);
destructor Destroy; override;
end;
implementation
function RandomStr: String;
var
C1: Integer;
begin
result := '';
for C1 := 10 to 20 + Random(50) do
result := result + Chr(Random(91) + 32);
end;
constructor TWorker.Create(const ALogger: TLogger);
begin
inherited Create(TRUE);
FLogger := ALogger;
end;
destructor TWorker.Destroy;
begin
inherited;
end;
procedure TWorker.Execute;
begin
while not Terminated do
FLogger.Add(RandomStr);
end;
end.
uLogger.pas:
unit uLogger;
interface
uses
Winapi.Windows, System.Classes;
type
TLogger = class
private
FStrings : TStringList;
FFileName : String;
FFlushThreshhold: Integer;
FLock : TRTLCriticalSection;
procedure LockList;
procedure UnlockList;
procedure Flush;
public
constructor Create(const AFile: String; const AFlushThreshhold: Integer = 1000);
destructor Destroy; override;
procedure Add(const AString: String);
property FlushThreshhold: Integer read FFlushThreshhold write FFlushThreshhold;
end;
implementation
uses
System.SysUtils;
constructor TLogger.Create(const AFile: String; const AFlushThreshhold: Integer = 1000);
begin
FFileName := AFile;
FFlushThreshhold := AFlushThreshhold;
FStrings := TStringList.Create;
InitializeCriticalSection(FLock);
end;
destructor TLogger.Destroy;
begin
FStrings.Free;
DeleteCriticalSection(FLock);
inherited;
end;
procedure TLogger.LockList;
begin
EnterCriticalSection(FLock);
end;
procedure TLogger.UnlockList;
begin
LeaveCriticalSection(FLock);
end;
procedure TLogger.Add(const AString: String);
begin
LockList;
try
FStrings.Add(AString);
if FStrings.Count >= FFlushThreshhold then
Flush;
finally
UnlockList;
end;
end;
procedure TLogger.Flush;
var
strbuf : PChar;
hFile : THandle;
bWritten: DWORD;
begin
hFile := CreateFile(PChar(FFileName), GENERIC_WRITE, FILE_SHARE_READ, nil, OPEN_ALWAYS, FILE_ATTRIBUTE_NORMAL, 0);
try
strbuf := PChar(FStrings.Text);
SetFilePointer(hFile, 0, nil, FILE_END);
WriteFile(hFile, strbuf^, StrLen(strbuf), bWritten, nil);
FStrings.Clear;
finally
CloseHandle(hFile);
end;
end;
end.

Related

Delphi TMonitor.Wait multi-threading problem

We have run into this multi-threading problem in our backend services:
In a multi-threading Windows service app, with 30+ threads, problem in SysUtils.EventCache arise. The problem is that NewWaitObj function sometimes return NIL instead of Event object. This function is used in TMonitor sync methods Wait. TMonitor.Wait stops working when it get NIL for event object. That affects many VCL and RTL thread sync source code and it cause different side problems in multi-threading apps, for example TThreadedQueue.PopItem doesnt wait for new item to arrive in Queue and returns immediately with timeout result.
Problem occurs in NewWaitObj function:
function NewWaitObj: Pointer;
var
EventItem: PEventItemHolder;
begin
EventItem := Pop(EventCache);
if EventItem <> nil then
begin
Result := EventItem.Event;
EventItem.Event := nil;
Push(EventItemHolders, EventItem);
end else
Result := NewSyncWaitObj;
ResetSyncWaitObj(Result);
end;
Looks like Pop function is not well protected in heavy multi-threaded app and at some number of concurrent threads it starts to return one and the same EventItem instance to two (or more) threads. Then race conditions are happening in NewWaitObj:
One thread takes EventItem.Event and return it as Result and zero it with NIL, the racing parallel thread is getting the same EventItem.Event but it is already cleared by first thread.
That cause one of racing threads to return valid Event handle and the other(s) racing threads return NIL.
TMonitor.Wait function doesnt work, because it get NIL as Event handle.
TThreadedQueue.PopItem doesnt wait, other sync methods also doesnt work correctly.
For some reason thread sync in Pop method doesnt work when app have many concurrent threads:
function Pop(var Stack: PEventItemHolder): PEventItemHolder;
begin
repeat
Result := Stack;
if Result = nil then
Exit;
until AtomicCmpExchange(Pointer(Stack), Result.Next, Result) = Result;
end;
In test app on 60 test threads problem arise in about 10-20 secs, with 30 threads its much harder to happens, usually 5-10 mins are needed. Once problem occurs - it never stop until restart of App. In test app after thread sync get broken - about one of each 5 operations with EventCache return NIL. Looks like something get broken in AtomicCmpExchange, I've checked the generated code - it's just one CMPXCHG instruction and few more to setup registers. I am not quite sure what cause the problem - can one thread get intervention from other thread for example while it setups registers to call CMPXCHG or after the call while it process the results?
Trying to understand what cause the problem, so I can find best workaround. For now I am planning to replace original NewWaitObj with my own, which will just call the original version till it return valid object. This problem occurs constantly in our dev, test and prod environments, for real middle-ware services on production servers it's needed few hours (sometimes couple of days) for problem to arise, after that only restart fix the problem.
Test app can be downloaded from issue in Embarcadero JIRA: https://quality.embarcadero.com/browse/RSP-31154
EDIT: TestApp: https://quality.embarcadero.com/secure/attachment/31605/EventCacheBug.zip
Example Delphi source code:
unit FormMainEventCacheBugU;
interface
uses
Winapi.Windows, Winapi.Messages, System.SysUtils, System.Variants, System.Classes, Vcl.Graphics,
Vcl.Controls, Vcl.Forms, Vcl.Dialogs, Math, Vcl.StdCtrls;
const
MaxProducers = 60;
type
TFormEventCacheBug = class(TForm)
BtnMaxProducers: TButton;
BtnRemoveProducer: TButton;
BtnAddProducer: TButton;
procedure BtnMaxProducersClick(Sender: TObject);
procedure BtnRemoveProducerClick(Sender: TObject);
procedure BtnAddProducerClick(Sender: TObject);
private
{ Private declarations }
public
{ Public declarations }
end;
TEventEater = class(TThread)
private
SleepTime: Integer;
SMsg, EMsg, NMsg: PChar;
procedure EatEvent;
protected
procedure Execute; override;
public
constructor Create;
end;
var
FormEventCacheBug: TFormEventCacheBug;
Producers: array[1..MaxProducers] of TThread;
ProdCount: Integer;
implementation
{$R *.dfm}
procedure AddProducer;
begin
if ProdCount < MaxProducers then
begin
Inc(ProdCount);
Producers[ProdCount] := TEventEater.Create;
Producers[ProdCount].FreeOnTerminate := True;
end;
end;
procedure RemoveProducer;
begin
if ProdCount > 0 then
begin
Producers[ProdCount].Terminate;
Dec(ProdCount);
end;
end;
{ TEventEater }
constructor TEventEater.Create;
begin
inherited Create(False);
SleepTime := RandomRange(1, 3);
end;
procedure TEventEater.EatEvent;
var
EventHandle: Pointer;
begin
//OutputDebugString(SMsg);
EventHandle := System.MonitorSupport.NewWaitObject;
try
if EventHandle = nil then
OutputDebugString('NIL');
Sleep(SleepTime);
finally
if EventHandle <> nil then
System.MonitorSupport.FreeWaitObject(EventHandle);
// OutputDebugString(EMsg);
end;
end;
procedure TEventEater.Execute;
begin
SMsg := PChar('S:' + GetCurrentThreadId.ToString);
EMsg := PChar('E:' + GetCurrentThreadId.ToString);
NMsg := PChar('NIL:' + GetCurrentThreadId.ToString);
while not Terminated do
begin
EatEvent;
Sleep(SleepTime);
end;
end;
procedure TFormEventCacheBug.BtnAddProducerClick(Sender: TObject);
begin
AddProducer;
end;
procedure TFormEventCacheBug.BtnRemoveProducerClick(Sender: TObject);
begin
RemoveProducer;
end;
procedure TFormEventCacheBug.BtnMaxProducersClick(Sender: TObject);
var
i: Integer;
begin
for i := ProdCount + 1 to MaxProducers do
AddProducer;
end;
end.
Thanks for any ideas,
#MiroslavPenchev, thank you for the post!
Working in XE2 and had similar issue.
Delphi 10.4.1 got TMonitor ABA problem solved using the linked list head with a counter and 128-bit Compare Exchange.
Unfortunately this is not an easy option for XE2.
Again, thanks to your suggestion to override some of MonitorSupport methods calling original ones.
The following is the solution that I'm using. It is not 100% perfect as involves locking, but for less concurrent environment it at least makes system stable and free from 100% CPU issue.
var
MonitorSupportFix: TMonitorSupport;
OldMonitorSupport: PMonitorSupport;
NewWaitObjCS: TCriticalSection;
function NewWaitObjFix: Pointer;
begin
if Assigned(NewWaitObjCS) then
NewWaitObjCS.Enter;
try
Result := OldMonitorSupport.NewWaitObject;
finally
if Assigned(NewWaitObjCS) then
NewWaitObjCS.Leave;
end;
end;
procedure FreeWaitObjFix(WaitObject: Pointer);
begin
if Assigned(NewWaitObjCS) then
NewWaitObjCS.Enter;
try
OldMonitorSupport.FreeWaitObject(WaitObject);
finally
if Assigned(NewWaitObjCS) then
NewWaitObjCS.Leave;
end;
end;
procedure InitMonitorSupportFix;
begin
OldMonitorSupport := System.MonitorSupport;
MonitorSupportFix := OldMonitorSupport^;
MonitorSupportFix.NewWaitObject := NewWaitObjFix;
MonitorSupportFix.FreeWaitObject := FreeWaitObjFix;
System.MonitorSupport := #MonitorSupportFix;
end;
initialization
NewWaitObjCS := TCriticalSection.Create;
InitMonitorSupportFix;
finalization
FreeAndNil(NewWaitObjCS);
end.

Delphi set thread to nil after starting it

Today I have missed a lesson and I have found a really weird line of code that I cannot understand. This is the class:
type
TMemoMessageThread = class(TThread)
strict private
FMemo: TMemo;
FMemoMessage: string;
protected
procedure Execute; override;
public
constructor Create(aMemo: TMemo);
property MemoMessage: string read FMemoMessage write FMemoMessage;
end;
Here I have created a class descending from TThread because I want to simulate a heavy computation. Look at the very simple UI.
You click the Button2"Create suspended" which creates a Suspended thread and then use this thread pressing Button3. This is the code:
constructor TMemoMessageThread.Create(aMemo: TMemo);
begin
if (aMemo = nil) then
raise Exception.Create('tMemo non valid!!!');
inherited Create(True);
FreeOnTerminate := True;
FMemo := aMemo;
end;
procedure TMemoMessageThread.Execute;
begin
Synchronize(procedure
begin
FMemo.Lines.Add('... process in parallel thread ...');
end);
Sleep(7000); //simulate something heavy to compute
Synchronize(procedure
begin
FMemo.Lines.Add(FMemoMessage + '. Done! :)');
end);
end;
procedure TForm1.Button2Click(Sender: TObject);
begin
MemoMessageThread := TMemoMessageThread.Create(Memo1);
MemoMessageThread.MemoMessage := 'Hello from TMemoMessageThread';
Button3.Enabled := true;
end;
procedure TForm1.Button3Click(Sender: TObject);
begin
MemoMessageThread.Start;
MemoMessageThread := nil; //why nil after start?
end;
As you can see, why is there a nil after the start? I really cannot understand. Since it is FreeOnTerminate shouldn't I ignore that line?
Setting MemoMessageThread to nil removes the reference to the instantiated object. The object still lives and executes, but the contact to the object is lost.
Since the thread is created with FreeOnTerminate, you should not use a reference to the object anyway. This is the only purpose of setting it to nil as I can see.
In the scope of this limited example, there indeed is no point in nil-ling the reference. It doesn't do any harm either though.
The thread is created with FreeOnTerminate set, which means that once it has started, you must not access any of its properties again, because the thread might have already been terminated and freed. Nil-ling the reference makes sure that you cannot do that, and as such is not a bad idea, but isn't necessary as long as you remember to not access the started thread instance.

Dynamically initialize and call LoadLibrary from a TThread by demand only once

I have a Delphi DLL, which needs to be called from my main UI application or worker threads.
I do not want to call LoadLibrary/FreeLibrary each time I call the DLL. But, I also don't want to load it in my application initialization section. because I might not use the DLL at all during the lifetime of the application.
So what I need is the first caller (thread or main UI) to initialize and load the DLL.
the DLL will be unloaded in the finalization section. I realize I need some synchronization. so I have used a critical section BUT I can't seem to make it work.
Only one thread should attempt and load the DLL. if it fails other threads should not attempt to load the DLL again and again.
The synchronization is not working as expected!
Can someone suggest why?
MCVE:
program Project1;
{$APPTYPE CONSOLE}
uses
Windows,
SysUtils,
Classes;
const
MyDLL = 'MyDLL.dll';
type
TDLLProcessProc = function(A: Integer): Integer; stdcall;
var
DLLProc: TDLLProcessProc = nil;
DLLModule: HMODULE = 0;
DLLInitialized: Boolean = False;
DLLInitialized_OK: Boolean = False;
CS: TRTLCriticalSection;
procedure InitDLLByFirstCall;
begin
if DLLModule = 0 then
begin
if DLLInitialized then Exit;
EnterCriticalSection(CS);
try
if DLLInitialized then Exit;
DLLInitialized := True;
DLLModule := LoadLibrary(MyDLL);
if DLLModule = 0 then RaiseLastWin32Error;
DLLProc := GetProcAddress(DLLModule, 'Process');
if #DLLProc = nil then RaiseLastWin32Error;
DLLInitialized_OK := True;
finally
LeaveCriticalSection(CS);
end;
end;
end;
function DLLProcess(A: Integer): Integer;
begin
InitDLLByFirstCall;
if not DLLInitialized_OK then
raise Exception.Create('DLL was not initialized OK');
Result := DLLProc(A);
end;
type
TDLLThread = class(TThread)
private
FNum: Integer;
public
constructor Create(CreateSuspended: Boolean; ANum: Integer);
procedure Execute; override;
end;
constructor TDLLThread.Create(CreateSuspended: Boolean; ANum: Integer);
begin
FreeOnTerminate := True;
FNum := ANum;
inherited Create(CreateSuspended);
end;
procedure TDLLThread.Execute;
var
RetValue: Integer;
begin
try
RetValue := DLLProcess(FNum);
Sleep(0);
Writeln('TDLLThread Result=> ' + IntToStr(RetValue));
except
on E: Exception do
begin
Writeln('TDLLThread Error: ' + E.Message);
end;
end;
end;
var
I: Integer;
begin
InitializeCriticalSection(CS);
try
// First 10 thread always fail!
for I := 1 to 10 do
TDLLThread.Create(False, I);
Readln;
for I := 1 to 10 do
TDLLThread.Create(False, I);
Readln;
finally
DeleteCriticalSection(CS);
end;
end.
DLL:
library MyDLL;
uses
Windows;
{$R *.res}
function Process(A: Integer): Integer; stdcall;
begin
Result := A;
end;
exports
Process;
begin
IsMultiThread := True;
end.
You need to modify your code in a way that the condition variable that is checked at the start of InitDLLByFirstCall is set only after all initialization has been completed. The DLL handle is therefore a bad choice.
Second you need to use the same condition variable outside and inside of the critical section - if you use DLLInitialized for that, then there is not really a use for either DLLInitialized_OK nor DLLModule.
And to make things easier to reason about you should try to get away with the minimum number of variables. Something like the following should work:
var
DLLProc: TDLLProcessProc = nil;
DLLInitialized: Boolean = False;
CS: TRTLCriticalSection;
procedure InitDLLByFirstCall;
var
DLLModule: HMODULE;
begin
if DLLInitialized then
Exit;
EnterCriticalSection(CS);
try
if not DLLInitialized then
try
DLLModule := LoadLibrary(MyDLL);
Win32Check(DLLModule <> 0);
DLLProc := GetProcAddress(DLLModule, 'Process');
Win32Check(Assigned(DLLProc));
finally
DLLInitialized := True;
end;
finally
LeaveCriticalSection(CS);
end;
end;
function DLLProcess(A: Integer): Integer;
begin
InitDLLByFirstCall;
if #DLLProc = nil then
raise Exception.Create('DLL was not initialized OK');
Result := DLLProc(A);
end;
If you don't want to check for the function address inside of DLLProcess then you could also use an integer or enumeration for the DLLInitialized variable, with different values for not initialized, failed and success.
You've got your double checked locking implemented incorrectly. You assign to DLLModule before assigning to DLLProc. So DLLModule can be non-zero whilst DLLProc is still null.
The variable that you test outside the lock must be modified after all the initialization is complete.
The pattern is like this:
if not Initialised then begin
Lock.Enter;
if not Initialised then begin
// Do initialisation
Initialised := True; // after initialisation complete
end;
Lock.Leave;
end;
Remember that double checked locking, as implemented here, only works because of the strong x86 memory model. If you ever move this code onto hardware with a weak memory model, it won't work as implemented. You'd need to implement barriers. Possible to do, but not entirely trivial.
Double checked locking is pointless here though. Remove it and protect everything with a single critical section. You are spinning up a thread, a very expensive task. The potential contention on a critical section is negligible.

DLL and class in multithreaded application

I have a class inside my DLL. And this DLL provides an "interface" to create objects of this class and call their methods.
Class code (simplified):
TLogger = class
private
//
public
function AddToLog(sBuf: PWideChar): Word;
constructor Create(Name: PWideChar);
destructor Destroy; override;
end;
constructor Create(Name: PWideChar);
begin
//
end;
destructor TLogger.Destroy;
begin
//
end;
function TLogger.AddToLog(sBuf: PWideChar): Word;
var
Temp1 : WideString;
Temp2 : AnsiString;
begin
Result := NO_ERRORS;
WaitForSingleObject( FMutex, INFINITE );
Temp1 := 'a';
Temp2 := Temp1;
//
ReleaseMutex( FMutex );
end;
DLL code
function CreateLogger( LogFileName: PWideChar; PLogger: PCardinal): Word; stdcall;
begin
try
PLogger^ := Cardinal(TLogger.Create(LogFileName));
Result := NO_ERRORS;
except
Result := ERR_CREATE_OBJECT;
end;
end;
function DestroyLogger( PLogger: Cardinal ): Word; stdcall;
begin
try
TLogger(PLogger).Free;
Result := NO_ERRORS;
except
Result := ERR_OBJECT_NOT_FOUND;
end;
end;
function AddToLog( PLogger: Cardinal; BufStr: PWideChar ): Word; stdcall;
begin
try
Result := TLogger(PLogger).AddToLog( BufStr );
except
Result := ERR_OBJECT_NOT_FOUND;
end;
end;
When I'm trying to use this library from 1 thread - everything is OK. The problems start when I create many threads that call function AddToLog with random periods (each thread has it's own object of the class). In some time I catch Access Violation or Invalid pointer operation.
I've made some research and pointed out that if variable Temp2 has WideString type everything is OK. Another solution is to move mutex to the library code (it's just a "research" code):
function AddToLog( PLogger: Cardinal; BufStr: PWideChar ): Word; stdcall;
begin
WaitForSingleObject( TLogger(PLogger).FMutex, INFINITE );
Result := TLogger(PLogger).AddToLog( BufStr );
ReleaseMutex( TLogger(PLogger).FMutex );
end;
The second solution is bad for me because each object has it's own mutex (idea is that if two objects must work with one file, they have the same mutex to wait each other; if two objects must work with different files, they have different mutexes and work in parallel).
I'm trying to solve this problem for 2 days but I can't understand what goes wrong. How string cast can cause such problem?
Put the following line:
IsMultiThread := True;
as the first line in your DLL project's main code block. This will instruct the memory manager to switch to a thread-safe mode.
This would explain a behaviour difference between AnsiString and WideString because AnsiString is allocated by the Delphi memory manager, and WideString is allocated on the COM heap. When IsMultiThread is False, the Delphi memory manager is not thread-safe, but the COM heap is always thread-safe.

How can I access the members of a class from a thread procedure without TThread?

I have a unit something like this
type
TMyClass = Class(TObject)
private
AnInteger : Integer;
MyThreadHandle : DWORD;
procedure MyPrivateProcedure;
public
procedure MyPublicProcedure;
end;
procedure TMyClass.MyPrivateProcedure;
procedure MyThread; stdcall;
begin
if AnInteger <> 0 then MyPublicProcedure;
end;
var
DummyID: DWORD;
begin
MyThreadHandle := CreateThread(NIL,0,#MyThread,NIL,0, DummyID);
end;
procedure TMyClass.MyPublicProcedure;
begin
AnInteger := 0;
end;
My goal is to have a Thread (no TTthread please.) that can "access" the vars/functions/procedures just like it's part of the class. This Example fails because it doesn't have access to the vars nor to the procedure. This is just an example, I am aware that the Integer can't change just like that. To me it's just important to have a thread that is part of the class. I also tried to pass the integer as a pointer (which worked) to the thread but I still can't access a procedure/function of the class. any ideas?
You can use TThread and keep filesize small. I think you are going into a difficult path: reinvent the wheel is time consuming, I can tell you that! :)
Here is some working code to initialize the thread:
function ThreadProc(Thread: TThread): Integer;
var FreeThread: Boolean;
begin
if not Thread.FTerminated then
try
result := 0; // default ExitCode
try
Thread.Execute;
except
on Exception do
result := -1;
end;
finally
FreeThread := Thread.FFreeOnTerminate;
Thread.FFinished := True;
if Assigned(Thread.OnTerminate) then
Thread.OnTerminate(Thread);
if FreeThread then
Thread.Free;
EndThread(result);
end;
end;
constructor TThread.Create(CreateSuspended: Boolean);
begin
IsMultiThread := true; // for FastMM4 locking, e.g.
inherited Create;
FSuspended := CreateSuspended;
FCreateSuspended := CreateSuspended;
FHandle := BeginThread(nil, 0, #ThreadProc, Pointer(Self), CREATE_SUSPENDED, FThreadID);
if FHandle = 0 then
raise Exception.Create(SysErrorMessage(GetLastError));
SetThreadPriority(FHandle, THREAD_PRIORITY_NORMAL);
end;
That is, you pass the object as pointer() to the thread creation API, which will be passed as the unique parameter of the ThreadProc.
The ThreadProc should NOT be part of any method, but global to the unit.
Here is another piece of code directly calling the APIs to handle multi-thread compression, with no overhead, and synchronization:
type
TThreadParams = record
bIn, bOut: pAESBlock;
BlockCount: integer;
Encrypt: boolean;
ID: DWORD;
AES: TAES;
end;
{ we use direct Windows threads, since we don't need any exception handling
nor memory usage inside the Thread handler
-> avoid classes.TThread and system.BeginThread() use
-> application is still "officialy" mono-threaded (i.e. IsMultiThread=false),
for faster System.pas and FastMM4 (no locking)
-> code is even shorter then original one using TThread }
function ThreadWrapper(var P: TThreadParams): Integer; stdcall;
begin
with P do
AES.DoBlocks(bIn,bOut,bIn,bOut,BlockCount,Encrypt);
ExitThread(0);
result := 0; // make the compiler happy, but won't never be called
end;
procedure TAES.DoBlocksThread(var bIn, bOut: PAESBlock; Count: integer; doEncrypt: boolean);
var Thread: array[0..3] of TThreadParams; // faster than dynamic array
Handle: array[0..3] of THandle; // high(Thread) is not compiled by XE2
nThread, i, nOne: integer;
pIn, pOut: PAESBlock;
begin
if Count=0 then exit;
if {$ifdef USEPADLOCK} padlock_available or {$endif}
(SystemInfo.dwNumberOfProcessors<=1) or // (DebugHook<>0) or
(Count<((512*1024) div AESBlockSize)) then begin // not needed below 512 KB
DoBlocks(bIn,bOut,bIn,bOut,Count,doEncrypt);
exit;
end;
nThread := SystemInfo.dwNumberOfProcessors;
if nThread>length(Thread) then // a quad-core is enough ;)
nThread := length(Thread);
nOne := Count div nThread;
pIn := bIn;
pOut := bOut;
for i := 0 to nThread-1 do
with Thread[i] do begin // create threads parameters
bIn := pIn;
bOut := pOut;
BlockCount := nOne;
Encrypt := doEncrypt;
AES := self; // local copy of the AES context for every thread
Handle[i] := CreateThread(nil,0,#ThreadWrapper,#Thread[i],0,ID);
inc(pIn,nOne);
inc(pOut,nOne);
dec(Count,nOne);
end;
if Count>0 then
DoBlocks(pIn,pOut,pIn,pOut,Count,doEncrypt); // remaining blocks
inc(Count,nOne*nThread);
assert(integer(pIn)-integer(bIn)=Count*AESBlockSize);
assert(integer(pOut)-integer(bOut)=Count*AESBlockSize);
bIn := pIn;
bOut := pOut;
WaitForMultipleObjects(nThread,#Handle[0],True,INFINITE);
for i := 0 to nThread-1 do
CloseHandle(Handle[i]);
end;
A thread has its own stack pointer, so you can't access local variables or parameters (like the hidden Self parameter) in you MyThread local procedure (which BTW is declared wrong). Furthermore you can't use local procedures for threads if they access variables (including Self) from the outer function. And if you want to use the 64bit compiler in the future, you can't use local procedures for any callback.
In your case you just have to fix the declaration of your procedure and move it into the unit scope (make it a "stand alone" procedure. This allows you to use the thread-callback parameter for "Self".
function MyThread(MyObj: TMyClass): DWORD; stdcall;
begin
if MyObj.AnInteger <> 0 then
MyObj.MyPublicProcedure;
Result := 0;
end;
procedure TMyClass.MyPrivateProcedure;
var
DummyID: DWORD;
begin
MyThreadHandle := CreateThread(nil, 0, #MyThread, Self, 0, DummyID);
end;

Resources