Remy Lebeau

FYI, your use of the TIdIOHandler.ReadFromSource() method is just mirroring what the public TIdIOHandler.CheckForDataOnSource() method does internally, so you don't really need your TIdIOHandlerHelper class at all, just use CheckForDataOnSource(1) instead of ReadFromStack(False, 1, False).

Unless the data is structured in such a way that you can definitively identify, from ANY point within the data, where the data ends and the next piece of communication begins, then "flushing" the existing data to resume communications is usually not a viable option. Especially if the next piece of communication has already been received and buffered and waiting for your code to read it from the buffer. Are you using a standard protocol like FTP for the transfer, or are you using a custom protocol? Unless the exception is a socket error, then you don't know the actual state of the socket, or how the data was buffered by the socket or by Indy. So, once the exception occurs, usually the only sane thing to do is to just disconnect and reconnect to restart communications fresh, yes. This is true for any TCP library, it is not limited to just Indy. Since you know how much data you transferred before the exception, you can resume the transfer from that position, if your system allows for resuming.

I don't really understand what you are asking for, but why would you ever need something like this? Please provide a use-case example. No (if such a flush were possible). By default, there is no write buffering beyond the send buffer that is built-in to the underlying socket at the OS layer. You can't "flush" that buffer, but you can reduce its size with the SO_SNDBUF socket option, or disable/avoid it with the TCP_NODELAY socket option (Indy's UseNagle property). On top of that, Indy does provide an application-level send buffer via the IOHandler.WriteBuffer...() methods. That buffer is flushed to the socket when the specified threshold is reached, or when WriteBufferFlush() or WriteBufferClose() is called. That buffer can be discarded with WriteBufferCancel(). Read buffering is a bit trickier. There is the receive buffer that is built-in to the underlying socket at the OS layer. You can't "flush" that buffer, only reduce its size with the SO_RCVBUF socket option. On top of that is the application-level buffering that Indy implements via the IOHandler.InputBuffer property, where the IOHandler.RecvBufferSize property controls how many bytes Indy attempts to read from the socket's receive buffer into the InputBuffer at a time. Once data is in the InputBuffer, you can "flush" that buffer by calling its Clear() or Remove() method (in the latter case, you could use the IOHandler.Discard() method instead). So, what is it that you really want to "flush", exactly? What is the scenario that you feel "flushing" is the solution for?

No (plus, that would require changing and recompiling the code, in which case you may as well just take out the offending code to begin with). But, you can accomplish something similar using breakpoints instead. Put a breakpoint at the beginning of the function, go into its properties, and disable the 'Break' option, and enable the 'Ignore subsequent exceptions' option. Then, put another breakpoint at the end of the function, go into its properties, and disable the 'Break' option, and enable the 'Handle subsequent exceptions' option.

You have the call to ShowDialogAfterTaskFinishes() in the wrong place. TTask runs asynchronously, but you are calling ShowDialogAfterTaskFinishes() immediately after the TTask object is created (and possibly before the TTask even begins to run). You need to move the call to ShowDialogAfterTaskFinishes() inside of the TTask procedure itself, similar to what you have for the call to UpdateMainWindow(), eg: procedure TfQBDailyInvoices.TransferEvent; begin TTask.Run( procedure var x: Integer; begin try for x := 0 to qInvoice.RecordCount - 1 do begin DidWorkorderTransfer := ProcessInvoice; TThread.Synchronize(nil, procedure begin { Update User Interface } UpdateMainWindow; end ); qInvoice.Next; end; finally TThread.Queue(nil, ShowDialogAfterTaskFinishes); end; end ); end; Alternatively, you can use TThread.CreateAnonymousThread() instead, so you can use the TThread.OnTerminate event (which is Synchronize()'d with the main UI thread), eg: procedure TfQBDailyInvoices.TransferEvent; var Thread: TThread; begin Thread := TThread.CreateAnonymousThread( procedure var x: Integer; begin for x := 0 to qInvoice.RecordCount - 1 do begin DidWorkorderTransfer := ProcessInvoice; TThread.Synchronize(nil, procedure begin { Update User Interface } UpdateMainWindow; end ); qInvoice.Next; end; end ); Thread.OnTerminate := TransferEventThreadTerminated; Thread.Start; end; procedure TfQBDailyInvoices.TransferEventThreadTerminated; begin ShowDialogAfterTaskFinishes; end; :

That is not what you described originally: Two different things. So, which one are you actually working with? Can you provide a code example that is crashing for you?

It shouldn't work, though. This is just straight up undefined behavior, since the TMemo.Lines property does not point at a TStringList object, it points at a TMemoStrings object instead. And TSStringList overrides Add() whereas TMemoStrings does not, so even the vtable contents are different. Don't write code like this. The correct solution is to make FillStrings() take a TStrings instead: function FillStrings(aStrings: TStrings): boolean; begin aStrings.Add('a'); aStrings.Add('b'); end; ... FillStrings(Memo1.Lines); Correct, because FillStrings() expects a TStringList specifically, not a TStrings generally. Nope. Nope.

If only there was a system where component authors could update their GetIt submissions whenever they wanted..... oh wait, we were promised that years ago, but it never materialized.....

They are only just now working on this? When CE was first introduced several years ago, they didn't think this might be a problem down the line? They've had 3 years to fix it, what are they hoping to do in a few days?

The type of sublist is the plain Pointer-based TList, so it makes sense that these hoops have to be done manually to store strings in it, but it is really not optimal. If the code were updated to use TStringList or TList<String> instead, then you wouldn't have to worry about manually managing the string memory at all.

On the other hand, depending on the situation, it would probably be better not to perform the calculation at all if the input is invalid, in which case TryStrToFloat() would make more sense if you want to avoid the exception and skip the calculation at the same time.

I still think it is likely a bug in the program, not in Windows itself. Like if the program had opened the StorageDevicePolicies key for some unrelated purpose, and then accidentally used that opened HKEY handle as a base when writing its settings to its own subkey rather than using the predefined base HKLM handle.

You can't perform arithmetic with a Double and a String as operands. You have to first convert the String to a Double/Extended first. Use StrToFloat() or equivalent for that. And then you can convert the result back to a String using FloatToStr() or equivalent. Try something more like this: DBEdit43.Text := FloatToStr(0.000844 * StrToFloat(scGPDBEdit82.Text));

Yeah, like not using a TEdit for date/time input in the first place. Use a UI control that is specifically designed for date/time input (ie, TDateTimePicker, TDatePicker, TTimePicker). Same with integers, for instance (ie, TSpinEdit).

Yes, that is where I would expect them, too. That is absolutely wrong. Application settings do not belong there. That is likely a bug in the program. It is a system-managed key, and it holds system settings, hardware configurations, service setups, etc. Note that HKLM\SYSTEM likely contains other keys ControlSet001, ControlSet002, etc. Those are like backups/previous versions of CurrentControlSet. If something goes wrong with the current settings, Windows has something it can roll back to. Every computer has a CurrentControlSet key. But not every computer has a StorageDevicePolicies key (mine doesn't). Absolutely not. You should contact the program author and explain the issue so they can fix it.

You have 2 StrToDate() calls in that highlighted code (presumably more offscreen?), both are converting scGPDBDateEdit11.Text. Your earlier screenshot show the Text is ' / / ', which is obviously NOT a valid date string, which is why StrToDate() raises EConvertError. It should work just fine, eg: if TryStrToDate(scGPDBDateEdit11.Text, mydate) then begin // use mydate as needed... end else begin // do something else... end; Though, like I said, TDateTimePicker makes a lot more sense then a TEdit for date inputs, eg: mydate := IncYear(DateTimePicker11.Date, a); ... mydate := IncDay(DatePicker11.Date, a); ...

You can use the 'is' operator for that purpose, eg: procedure GetString(param1: string; param2: string; myControl: TControl); begin if myControl is TLabel then TLabel(myControl).Caption := GetDescription() else if myControl is TMemo then TMemo(myControl).Lines.Add(GetDescription()) else ...; end; Alternatively, every TControl has a Caption (and Text) property, even if it is not published. You can take advantage of that, eg: type TControlAccess = class(TControl) end; procedure GetString(param1: string; param2: string; myControl: TControl); begin if myControl is TMemo then TMemo(myControl).Lines.Add(GetDescription()) else TControlAccess(myControl).Caption := GetDescription(); end; Or, you can use RTTI instead, eg: procedure GetString(param1: string; param2: string; myControl: TControl); var Ctx: TRttiContext; Typ: TRttiType; Prop: TRttiProperty; begin Ctx := TRttiContext.Create; try Typ := Ctx.GetType(myControl.ClassType); Prop := Typ.GetProperty('Caption'); if (Prop <> nil) and (Prop.Visibility = mvPublished) then begin Prop.SetValue(myControl, GetDescription()); Exit; end; Prop := Typ.GetProperty('Text'); if (Prop <> nil) and (Prop.Visibility = mvPublished) then begin Prop.SetValue(myControl, GetDescription()); Exit; end; Prop := Typ.GetProperty('Lines'); if (Prop <> nil) and (Prop.Visibility = mvPublished) then begin (Prop.GetValue(myControl).AsObject() as TStrings).Add(GetDescription()); Exit; end; ... finally Ctx.Free; end; end;

If you are converting the Text of the "né" control to a TDate/TDateTime using StrToDate(), consider using TryStrToDate() instead. That would avoid the EConvertError being raised. Just make sure you pay attention to its return value to know if the conversion was successful or not. Personally, I would use a TDateTimePicker or TDatePicker instead. For instance, TDateTimePicker as a built-in CheckBox that can be used to indicate whether a date has been entered.

Indy comes pre-installed in the IDE. If you want to use your own compiled version of Indy, you should remove the pre-installed one to avoid conflicts.

Then you are likely not building with the DCUs you just compiled. Did you remove Indy from the IDE first? Did you update the IDE's search paths? https://github.com/IndySockets/Indy/wiki/Updating-Indy Also, which platform are you running your code on? The leaks you have showed are the intentional Indy leaks that should not be appearing on Windows, as they are "registered by pointer", as the dialog says.

You did not explain what kind of leaks you are actually experiencing (can you provide the actual leak report?), but I suspect the leaks have nothing to do with the TIdHTTP object itself. Are you referring to the intentional leaks in Indy's IdStack.pas and IdThread.pas units? If so, that is a completely separate matter, and one that is not affected by the Owner you assign to the TIdHTTP object.

And? The problem is ...? Wow, that is a really dirty and ugly hack. That code is reading the machine instructions of TIdStack.DecUsage() at runtime trying to find where it accesses the global GStackCriticalSection variable, and then it dereferences that variable to get the address of the TIdCriticalSection object, and then register it with FastMM so it won't appear in leak reports. I don't see why that hack is necessary at all when Indy already does that registration during unit initialization when REGISTER_EXPECTED_MEMORY_LEAK is defined, which it is by default when FREE_ON_FINAL is not defined, and either HAS_System_RegisterExpectedMemoryLeak or USE_FASTMM4 are defined. Yes, that is the correct solution. You have to recompile Indy for it to take effect. The simplest way would be to have your project refer to Indy's source files directly so they compile directly into the executable. If you have installed Indy into the IDE, you will have to recompile the installed BPLs instead.

That is a bad idea in general. It prevents users from being able to set their own working directory when running your app from a command-line window or a shortcut.

Check with TDirectory.GetCurrentDirectory() or equivalent to make sure that '.' refers to the folder you think it does.

A read timeout error has nothing to do with authentication. Sounds more like the data connection is probably being blocked. Remember, FTP uses multiple TCP connections. I notice in your earlier log that a PORT command is being issued. That means your FTP client is acting in ACTIVE mode, where the server opens a data connection to the client. That is not very friendly to routers/proxies. You should use PASSIVE mode instead, where the client opens a data connection to the server instead. That is less likely to cause problems.