|
The CString class is not thread safe but using CAsyncSocket indicates that you are using multiple threads.
When using GetBuffer[Xxx], the string object is locked until ReleaseBuffer is called (which must be called from the same thread).
As Richard already noted, you need a byte buffer. While this can be achieved by using CStringA, I recommend to rethink your design. Don't use string classes for binary data and ensure thread safety.
|
|
|
|
|
Are you to
CByteArray if so how can I ensure the size
|
|
|
|
|
You should just use a simple array of bytes, which is what the socket method requires.
|
|
|
|
|
You probably don't need an array class if you just want to copy data and do not need inserting or removing of items. If you know the (max.) size in advance just use a plain array allocated with new. If not, you can still grow by allocating a new array, copying data from the old, and delete the old.
CByteArray is also an MFC class which is probably also not thread safe. But it is impossible to help without knowing what your code is doing.
The question are (you may ask and solve them yourself):
Does my array require special modification functions besides copying?
Which functions use my array under which conditions (read only, write only, both)?
Where is may array allocated (this might be at different code points if resizing is necessary)?
Where it is deleted?
Do I need thread safety?
|
|
|
|
|
I'm reading this excellent book. I considered to use steering behavior in our MMORPG game server project.
but if we use it all,it would consume a lot of cpu time which would be unacceptable for a server project.
does any one have any experience on this topic?How could I find a proper way to use steering behavior in MMORPG server?
Thank you very much.
|
|
|
|
|
You write the search algorthim as an iteration, with adjustable resolution and trigger(s).
If two objects are more than X apart the search doesn't run at all unless there is a special trigger. If you haven't been "spotted" there is no need for an enemy to search and so they usually just walk a defined set of waypoints called "pathing".
So movement especially for NPC's or bots usually have two different behaviours being "pathing" and "searching". Often there is also a way to get "unspotted" and NPC's/BOTs may be leashed to some co-ordinate. If they get more than some distance from the leash point they will drop "seaching" behaviour and resume "pathing".
The trigger whatever that is usually establishes the "target" and it is at that point the search AI kicks in. The game design thus far made sure searching is used sparingly.
Now when the search algorthim kicks in, it sets a resolution grid based upon how far the target is away. You don't need a fine grid for a target far away. So you will have a raycast which you run to each point on the grid. So when target is a long way away you still only get a small grid just they represent a big distance. That is the bit your code you currently have probably doesn't do and so you have a huge grid when the target is a long way away.
So now with your grid you run the AI search algorthim to work out what direction to move to the target allowing for walls etc. If the target is a long way away the direction will only be rough but that is all it needs it will fine up as it gets closer.
The problem you probably have with the AI search they have given you is the grid is massive and it gets bigger and bigger the further a target is away.
That would be how the vast majority of games not just MMORPG games work. Suggested start point for reading
Near-Optimal Hierarchical Pathfinding (HPA*) | AiGameDev.com[^]
In vino veritas
modified 20-Jul-17 3:34am.
|
|
|
|
|
thank you very much,it's really a great help.
there is a lot of work to do
|
|
|
|
|
Hi
I am getting a heap corruption as indicated by the message in quotes to my VS output debug window
At first I tried gflags from the command line and did a full enable for my application or my image
As I read up on gflags it puts a extra block of storage (that's is protected) between every allocation
I have what I think is a pretty big machine 64 gb of ram, While running the program I quickly got a message that I was out of storage
I am wondering if there is anything I can do programmatically to invoke gflags
As I have an idea in what part of the program this happening
I allocate storage for a dialog derived object via new
upon receiving a tcp/ip message in a modeless dialog I do a create for derived Cdialog for which I previously did the "new"
it is after do the create of the dialog and populate the control including allocating via new a Richedit and streaming data to it that something goes wrong all storage is allocated via
new
I am thinking maybe I can track this down with HeapValidate putting it various pots in this path of code and checking the return code
Thanks
|
|
|
|
|
I don't know how to do what you want with heap but there is an obvious way to make the program more deterministic which is to pre-create the dialogs just held in a storage array not in use. It is really dangerous to be creating anything so significant off an event.
When you get a message you call your little routine to pass you back the first free dialog which you load and use. When it's finished you put it back in the pool.
It's a lot easier to avoid such problems by changing behaviour to determinstic rather than chase the bug itself 
In vino veritas
|
|
|
|
|
Leon
I don't know if that's an option the dialog has a number of controls including a richedit which can be filled with a number of text lines
Would going down the path of using the functions in this article help me track down my problem
CRT Debug Heap Details[^]
|
|
|
|
|
This is where MFC obfuscating what is happening is annoying. You are building the dialog currently try not setting the modal flag and don't insert it into the desktop and tell me what happens .
You can make a dialog at any point in the lifecycle of your program, later on you can insert it into the desktop and make it visible and modal because that is how these things are designed to work.
The RichEdit control will happily take text, new or added or anything else you want via the message system. You can also talk it by it's control ID the sample below uses the handle.
CHARRANGE cr;
cr.cpMin = -1;
cr.cpMax = -1;
SendMessage(hwnd, EM_EXSETSEL, 0, (LPARAM)&cr);
SendMessage(hwnd, EM_REPLACESEL, 0, (LPARAM)some_new_string); That is how windows is designed to work. The best bit if you don't use the HasString flag you can just send the pointer in to the text which saves copying and moving the text.
If you have a dialog coming up a lot you are expected to do it that way or via the dialog template system so you don't fragment the hell out of the memory. Common dialogs works that way you know FileOpen, FileSave etc.
CreateDialogIndirect function (Windows)[^]
Creates a modeless dialog box from a dialog box template in memory
All you have to do is insert your changes to the text etc via the message system using the control ID's, put it into the desktop or your app and then set the dialog modal and viola it looks just like what you are doing now.
I don't have a problem creating the odd dialog on the fly but if it comes up a lot try and at least use windows the way it was intended.
The debug thing uses a similar trick but into a multiline text window. They change the malloc stub to post the text messages to the multiline edit window.
I think that this is the sample Microsoft uses to show how to do it in MFC. Not much of a sample and doesn't explain why you do it. The template sample does however contain a multiline edit element to show you they are not anything special.
VCSamples/VC2010Samples/MFC/general/dlgtempl at master · Microsoft/VCSamples · GitHub[^]
In vino veritas
modified 20-Jul-17 10:48am.
|
|
|
|
|
Leon
I can try to Create the Dialog Box from a Template ::LoadResource however
The actual call DialogBoxIndirect has to be when the message is Received from the other computer
Via Sockets as that Contains the Data
Thanks
|
|
|
|
|
Yep in the sample these are the key lines
dlg.InitModalIndirect((DLGTEMPLATE*)pBuffer);
dlg.DoModal(); // tadaaa! this is the line everyone's been waiting for!!!
I loved the tadaa in the source comments and it's because at that moment the penny should drop that the template has just been sitting in memory and wont do anything until you execute those two lines. It's much kinder on windows especially in a threaded app and less chance of making mistakes.
So when you get the message from your ethernet port all you have to do is execute those two lines and dialog flies out of memory and up on screen. You can actaully do everything else before that any earlier time.
Not sure how MFC does this but there is an lParam value that is part of the indirect and usually you pass a struct with your data you want loaded in the dialog handler in as well. It saves having to carry a global with your data.In your case I would imagine it would be a pointer to your payload data recieved. Again saves copy or moving any data.
It stuns me dlg.InitModalIndirect doesn't take a second parameter which is what I was expecting and where you would pass in your data pointer. So bit of mystery to me how you do that under MFC. Oh I wonder does dlg.InitModalIndirect have overrides and used as a shim on the indirect call to cater for that. In which case there will be a form of that you can pass a second parameter on.
In vino veritas
modified 21-Jul-17 0:11am.
|
|
|
|
|
Leon
I would assume the same is True for Modeless Dialogs I pass all my parameters once I have allocated the Cdialog Object on the Heap I populated it before doing Create Or Indirect
I have another related question can the same HGLOBAL template be used to Create multiple instances of that object
Thanks
|
|
|
|
|
Yes any dialog or window can be created from templates. The window one usually catches even seasoned programmers out. If you look at the template call CreateDialogIndirectParam the DLGPROC is optional. What you do is pass in the window handler and use setwindowlong to put it onto the modeless dialog which just happens to have all the style flags of a window. Once it has the handler attached you can just insert it into any APP window or even the desktop because it's a normal window and so you fake CreateDialogIndirectParam returning a fail. Did you notice it's up to you to call DestroyWindow and that is why they don't do it automatically. Sometimes it's useful to lie to the API
Yes its a template you can launch multiple copies as it only gets read and all global alloc memory is thread safe. It can't write or move anything as you usually only allocate just enough space to hold the resource. I assume that all still holds true for MFC. Officially the GlobalAlloc etc is now "old" but they have not got around to giving us replacement template system.
Most windows programmers are familiar with LoadResource and loading bitmap/icons/menus etc because they are going to draw them on screen alot but using the system for dialogs and windows usually escapes them because they don't need the response of a window like in your example.
In vino veritas
modified 21-Jul-17 9:50am.
|
|
|
|
|
The heap corruption indicates that you have an out of bound memory write access (e.g. writing to an array beyond it's size). Running out of memory indicates that you are allocating a lot of memory but not freeing it.
For modeless dialogs follow these adviceses:
- Create a variable and initialise it (assigning
NULL or using new)
- When the dialog should be created the first time check if the variable is
NULL and call new if so
- Create the dialog
- When finished with the dialog, call
DestroyWindow()
- If the allocated memory is not needed anymore, delete it and assign
NULL to the variable
- When the variable goes out of scope, delete the memory (e.g. in the destructor of the class holding the variable)
I suggest to check your code for the above (each new must have a corresponding delete, and each CDialog::Create must have a corresponding CDialog::DestroyWindow).
Because the memory corruption is a severe error, I suggest to find the source and fix it first. Use a debug build. Then you don't need the GFlags. They are for debugging executable files where you don't have the sources.
If you know the code portions (or have a guess) where the heap corruption occurs, you can insert some AfxCheckMemory[^] calls to check the C++ heap.
Quote: it is after do the create of the dialog and populate the control including allocating via new a Richedit and streaming data to it that something goes wrong all storage is allocated via new
A common error for such cases is forgetting to append a NULL char to the received data before passing them as string and/or forgetting to allocate the additional memory for the NULL char.
|
|
|
|
|
how to work with cards.dll
|
|
|
|
|
|
|
The article is out of date there is also a 32bit version of the DLL kicking around from Win98 and XP.
ReactOS basically duplicated most of the DLL as it is reasonably trivial
ReactOS: dll/win32/cards/cards.c Source File[^]
Find and download free deck of cards images, bind it all in Visual Studio and you have your own
In vino veritas
|
|
|
|
|
Yes, I am well aware of that, but it's the only one I know.
|
|
|
|
|
Can someone explain to me why for the following code
memcpy(&PSW+8,holdptr,8);
memcpy(&PSW,holdptr,8);
00007FF655264C1D mov rax,qword ptr [rbx+4]
00007FF655264C21 mov qword ptr [rbp+3F0h],rax
memcpy(&PSW+8,holdptr,8);
00007FF655264C28 mov rax,qword ptr [rbx+0Dh]
00007FF655264C2C mov qword ptr [rbp+530h],rax
The compiler generate a displacement of Hex 140 from the structure PSW instead of 8
struct {
BYTE sysmask;
BYTE pkey;
BYTE states;
BYTE asc;
BYTE cc;
BYTE progmask;
BYTE zerobyte;
u_int
amode64:1,
amode:1,
zeroilc:1;
char zeroword[4];
char ia[16];
char amask[4];
char intcode[2];
BYTE ilc;
BYTE unused;
} PSW;
|
|
|
|
|
You need to go and review C/C++ pointer arithmetic 
Your pointer arithmetic asks it to do that .... you must have your warnings turn way down because really it should give a warning.
Since it is has escaped you, specifically what you have asked it to do is add 8 times the size of PSW to the pointer to PSW which is what it did. For some reason your expect byte operation on a pointer that isn't a pointer to a byte. If you wanted a byte operation then cast the pointer to a byte.
I can't make head nor tails of what appears to be bitcount numbers. I warn you that you need pack(1) for that, and even if packed those bitcounts don't match. It looks like some of the bit fields need unions because there seems to be variants.
I assume you were trying to point at the 8th byte in the struct but without packing it isn't guaranteed you will get what you expect. I strongly recommend you put the bitpacks in a struct so it's named and then set the pointer to the named element in the struct.
It would be a whole lot easier to pack that structure exactly as whatever those bitcounts you are writing because it seems like it's a real thing ... So make the struct match reality. This is my best guess what your struct is supposed to be and I have used the proper standard C/C++ types, it is ill advised to do such precise packing on user defined types like BYTE. C/C++ standard file stdint.h is your friend use it when not on Windows/Linux API. The uint8_t will cast to everything properly as C knows exactly what it is.
#include <assert.h>
#include <stdbool.h>
#include <stdint.h>
#pragma pack(push, 1)
struct psw_s {
union {
struct variant1 {
uint8_t sysmask;
uint8_t pkey;
uint8_t states;
struct byte3_fields {
uint8_t asc : 2;
uint8_t cc : 2;
uint8_t progmask : 4;
};
union {
uint8_t zerobyte;
struct byte4_fields {
uint8_t zerobyte4 : 7;
bool amode64 : 1;
};
};
struct byte5_fields {
bool amode : 1;
bool zeroilc : 1;
uint8_t zerobyte5 : 6;
} AddrMode;
uint8_t variant1_unused[11];
};
char ia[16];
uint32_t ia32[4];
};
char amask[4];
char intcode[2];
BYTE ilc;
BYTE unused;
} PSW;
static_assert(offsetof(struct psw_s, amask) == 17, "Structure pack failed amask should be byte 17 in struct");
#pragma pack(pop)
So ia and the struct totally overlap each other they are different ways to get to same data being the 16 bytes (128 bits at the top of the struct). Array amask is at byte 17 in struct and it is checked at compile time to make dam sure it is. Compile time asserts were added in C++11 spec both GCC and VS have them as should Intel.
Anyhow now you can set the pointer to what you want by name. You can also access the bool fields. PW.AddrMode.amode is the named struct one, amode64 is in an anonymous struct so its just PW.amode64 to access. They are both just bools you can set them true/false or if test them like any bool. That shows you a named and unnamed struct bitfield access.
I think all that gets rid of the need for your pointer arithmetic as there is no field you can't hit by name.
It looks a lot like a hardware register definition I would do, but they get even worse than that
In vino veritas
modified 14-Jul-17 3:28am.
|
|
|
|
|
leon de boer wrote: you must have your warnings turn way down because really it should give a warning.
Or ignoring them. Either way a bad idea.
|
|
|
|
|
&PSW is of type "pointer to PSW", so &PSW+8 would have the same value as (PSW *)((byte *)&PSW + 8 * sizeof(PSW))
The old C pointer rules live on!
Cheers,
Peter
Software rusts. Simon Stephenson, ca 1994. So does this signature. me, 2012
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
