Nginx开发第三方模块实例
Ø 序言
Nginxq显著体现在对第三方模块的支持，开发者能够将自己开发的模块Q按?/span>nginx框架中，以实现开发者定制功能的作用?/span>
开发第三方模块的同学，文章内容主要包括三个部分Q如何将HTTP和如何开发第三方模块以及(qing)一个例子?/span>
Ø 如何?/span>HTTP模块嵌入nginx
Nginx文g+configure 开发中Q基本上都是?/span>config脚本方式比较多。接下来分别介绍q两U方式?/span>
1) Config文g+configure脚本
l Config文gQƈ且在文g里面l出三个配置，可以达到效果。这三个配置Ҏ(gu)指：(x)
A. Ngx_addon_nameQ?/span>B. Nginx模块名：(x)对应的模块有Q?/span>$HTTP_MODULES(模块), $HTTP_FILTER_MODULES(HTTPQ?/span>$HTTP_HEAD_FILTER_MODULES(HTTP Q?/span> $CORE_MODULES(nginxQ?/span>$EVENT_MODULES(nginx{，然后在后面添加你所要添加的模块名字Q以I格隔开Q比如这ơ我们开发添加的模块:
HTTP_MODULES = “$HTTP_MODULES ngx_http_helloworld_module”
C. NGX_ADDON_SRCSQ?/span>的时候就指定?jin)上层\径，比如--add-modules=PATHNGX_ADDON_SRCS = “$NGX_ADDON_SRCS $ngx_addon_dir/ngx_http_helloworld_module.c”
l Configure脚本
第三方模块的\径加入，
rd_modules/
a) Auto/modules
里面?/span>
a)
b) 一?/span>下面是开?/span>http
二?/span>文g
b) Auto/make
1)
2) 
2) 手动方式
在里面增加对应的W三Ҏ(gu)块，W二Q修?/span>obj/Makefile,Ø 开发第三方模块注意事项
的模块是有顺序性的Q比如：(x) HTTP_MODULES = “$HTTP_MODULES ngx_http_helloworld_module”
是异步非d的，在编制代码时Q一定要注意不能dnginxØ CZ
见附?a href="/Files/jolleydtan/test.zip">/Files/jolleydtan/test.zip
参考：(x)
深入理解nginx.
]]>
133 {
134 int size = vec.size();
135 IntVec::iterator first = vec.begin();
136 IntVec::iterator last = vec.end();
137 for(IntVec::iterator i = first + 1; i != last; ++i)
138 {
139 std::iter_swap(i, first + rand()%((i - first)+ 1));
140 }
141 return vec;
142 }
q个呢，不会(x)影响源数l，q个
2Q取Mơ，每次做个事情是随机Z个数Q添加到目标数组里面Q然后从源数l中删除。?font color="#000000" face="Verdana">
143
144 IntVec shuffle(IntVec& vec, int count)
145 {
146 IntVec desVec;
147 for(int i = 0; i < count;i++)
148 {
149 int random = rand()%vec.size();
150 desVec.push_back(vec[random]);
151 vec.erase(remove(vec.begin(),vec.end(),vec[random]),vec.end());
152 }
153 return desVec;
154 }
q个呢，?x)?jing)响源数组?br />3Q取Mơ，借助一个辅助数l记录哪个数据被取过Q随机出一个数Q添加到目标数组里面Q然后将q个值添加到辅助数组里面.
IntVec shuffle2(IntVec& vec, int count)
158 {
159 IntVec desVec;
160 IntVec flagVec;
161 for(int j = 0; j < count; j++)
162 {
163 flagVec.push_back(0);
164 }
165 for(int i = 0; i < count;i++)
166 {
167
168 int random = rand()%vec.size();
169 if (flagVec[random] == 0)
170 {
171 desVec.push_back(vec[random]);
172 flagVec[random] = 1;
173 }
174 else
175 {
176 i--;// 如果已经取过?jin)，q次不算.
177 }
178
179 }
180 return desVec;
181 }
]]>
以及(qing)nginx_stream_lua模块?br />https://github.com/openresty/stream-lua-nginx-module
本来想着用这两个的，后来发现已经有支持了(jin)Q就先来试试看看q个?jin)?br />默认的话Qnginx是没有开启对tcp的支持的.如果要支持的话，需要用--with-streamL定?br />我这边只要需要三个IPQ一个是nginxq程所在的机器Q我q边?92.168.1.4表示Q?br />另外两台是TCP上游业务服务器，我们分别?92.168.1.5?92.168.1.21来表C，
Zq样的考虑是因为nginx充当tcp代理服务器后Q自w也?x)占用一个端口?br />q且从目前来_(d)貌似一个nginxq程也就只能listen一个端口?br />我用的版本是nginx 1.9.9.
我在/usr/local/nginx/conf下面新徏?jin)nginx_stream_tcp.conf作ؓ(f)本次的测试?br />worker_processes auto;
2 error_log /usr/local/nginx/logs/error.log info;
3 events {
4 worker_connections 1024;
5 }
6
7 stream {
8 upstream loadbalance {
9 # Specifies a load balancing method for a server group where client-server mapping is based on the hashed key value
10 hash $remote_addr consistent;
11 # sets the weight of the server, by default, 1.
12 server 192.168.1.21:8001 weight=5;
13 # by default, the parameter is set to 10 seconds.
14 server 192.168.1.5:9001 max_fails=3 fail_timeout=30s;
15
16
17 }
18
19 server {
20 # listen 8001.
21 listen 8001;
22 # Defines a timeout for establishing a connection with a proxied server.
23 proxy_connect_timeout 1s;
24 # Sets the timeout between two successive read or write operations on client or proxied server connections
25 proxy_timeout 3s;
26 # Sets the address of a proxied server
27 proxy_pass loadbalance;
28 }
29
30 }
q个大部分是官网支持的，所以其实我也没有修改太多，p|了(jin)几个IP和端口?br />在编译好nginx后，可以?usr/local/nginx/sbin/nginx -c /usr/local/nginx/conf/nginx_stream_tcp.conf
可以v来了(jin)Q可以通过netstat -tlnp查看是否hQ大致显CL(fng)Q?br />tcp 0 0 0.0.0.0:8001 0.0.0.0:* LISTEN 7744/nginx_stream_t
当然也可以通过telnet来查看?br />接下来，写个单的csE序来验证下?
server端：(x)
if __name__ == '__main__':
import socket
import commands
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
sock.bind(('192.168.1.21', 8001))
sock.listen(5)
while True:
connection,address = sock.accept()
print "connected by",address
while 1:
buf = connection.recv(1024)
if buf != "" :
print "connection buffer", buf
connection.sendall(buf)
connection.close()
if __name__ == '__main__':
import socket
HOST='192.168.1.4'
PORT=8001
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
sock.connect((HOST, PORT))
while 1:
cmd = raw_input("please input cmd:")
sock.sendall(cmd)
data = sock.recv(1024)
print data
sock.close()
在client?92.168.1.4发v一个TCPq接Ӟ接下?92.168.1.4?x)连接?92.168.1.21Q?br />
后面的端口应该是随机的(f)时端口?br />其实它的vC(jin)转发作用Q这个nginxq程相当于一个通用|关Q如果有两个client向nginxq程发vq接Q一旦nginx那边调度到同一个上游TCP服务器，那么只需要从NGINXq程建立一个TCPq接C游TCP服务器即可?br />之前我自p计网关的时候，不是依赖于用L(fng)Q而是通过?rn)态配|来建立q接的，比如上游逻辑服务器跟|关有连接，那都是在启动服务器的时候，徏立v来了(jin)?br />q样有个好处是可以省事Q不q也可以采用q种方式Q有玩家q接q某个上游TCP服务器才让网兌q个TCP服务器徏立连接，否则无需处理。因Z时候还是会(x)有不服务器处于I闲?
]]>
{
int a;// 装备
int b;// 角色数据
char c;
};
void main( void )
{
//printf("hello world!");
char buffer[] = "This is a test of the memset function";
A aStruct;
// 发现数据相同?br /> memset(&aStruct,'*',sizeof(aStruct));
printf( "Before: %s\n", buffer );
memset( buffer, '*', 4 );
printf( "After: %s\n", buffer );
if (buffer[2] == '*')
{
printf("hello,world\n");
}
// 判断q个是否要保存?br /> if (aStruct.a == 0x2a2a2a2a)
{
// 发送给DBSERVER.
printf("hello char\n");
}
if (aStruct.c == 0x2a)
{
}
getchar();
]]>
Direct3D9: (INFO) :MemFini!
Direct3D9: (WARN) :Memory still allocated! Alloc count = 152
Direct3D9: (WARN) :Current Process (pid) = 00000fb8
Direct3D9: (WARN) :Memory Address: 00c95988 lAllocID=1 dwSize=000047f8, (pid=00000fb8)
q里是检出内存泄露?
一般是内存?sh)息什么的没有releasing和deleting?
之前遇到的一个问题lock is not supported when multi-sampling is enabled是因为shader debugging打开?jin)，之后关闭了(jin)，没有事情?jin)?br>刚才~译hlsl的时候，发现?jin)x3025错误Q后来的dx版本要用D3DXSHADER_ENABLE_BACKWARDS_COMPATIABLITY|D3DXSHADER_DEBUG才能~译,特别?008q的dx版本?br>
DX默认的是D3DFILL_SOLIDQ如果出现D3DFILL_WIREFRAME的情况一定要讄回去?
CD3DArcBall:允许你将用户的输入解释ؓ(f)旋{或者^U?我们所要做的是鼠标消息传递给q个c?q且视囄阵设为类的旋转或者^Uȝ阵就可以?
点乘: a.b = ||a|| *||b|| * cosA, 一般用来判断ؓ(f)0,那么表明正交,>0,方向基本相同,<0,方向基本相反.使用点乘来计投?
l定两向量v,n,其分解为vq_v?分别q于和垂直于n,q且满 v = vq?v?一般成q分量vqؓ(f)v在n 上的投媄(jing).
Vq?= n * ||Vq|| / ||n||.
cosA = ||Vq||/||V|| cosA||V|| = ||Vq||
Vq? n*n*v/( ||n||*||n||).
叉乘得到的向量垂直于原来的两个向?
a *b指向该^面的正上?垂直于a和b.
||a*b|| = ||a|| * ||b|| * sinA.
||a*b|| 也等于以a和bZ边的q四边形的面积.
如果a,bq或Q意一个ؓ(f)0,则a*b = 0,叉乘寚w向量的解释是它^行于L其他向量,点乘的解释是和Q意其他向量垂?
L对角矩阵D,都有D转置矩阵= D,包括单位矩阵.
DirectX使用的行向量.
v = xp+yq+zr,向量vpC成p,q,r的线性变换了(jin),向量p,q,rU做基向?q里基向量是W卡?dng)坐标?br>变换物体相当于以相反的量变换描述q个物体的坐标系.
2d里面是这栯|的:
逆时针旋转经?不是必须)被认为是正方?时针方向是负方?
lx轴旋?
[ 1 0 0
0 cosA sinA
0 -sinA cosA]
ly轴旋?
[cosA 0 -sinA
0 1 0
sinA 0 cosA]
lz轴旋?
[ cosA sinA 0
-sinA cosA 0
0 0 1]
lun旋{角度A,那么获得的矩阉|:
[nx*nx*(1-cosA)+cosA nx*ny*(1-cosA)-nz * sinA nx*nz*(1-cosA) + ny*sinA
nx * ny *(1-cosA) - nz *sinA ny*ny*(1-cosA)+cosA ny*nz*(1-cosA) - nx* sinA
nx * nz *(1-cosA)+ny * sinA ny * nz *(1-cosA) + nx* sinA nz*nz*(1-cosA)+cosA]
~放:通过基向量构造矩?得到以单位向量n为羃放方?k为因子的~放矩阵:
s(n,k) = [ p] [ 1 + (k-1)nx2 (k-1)nxny (k-1)nxnz]
=
[ q] [ (k -1)nxny 1+(k-1)ny2 (k-1)nxnz]
[ r] [(k-1)nxnz (k-1)nzny 1+(k-1)nz2]
投媄(jing)意味着降维操作,有一U投影方法是在某个方向上用零做羃攑֛?q种情况?所有点都被拉^臛_直的?2D)或^?3D),q种cd的投q做正交投?或者^行投?因ؓ(f)从原来的点到投媄(jing)点的直线怺q.
n垂直于投qU?而不是^?3D?向垂直于n的^面投q矩阵如公?
p(N) = [1-nx2 -nxny -nxnz]
[-nxny 1-ny2 -nynz]
[-nxnz -nzny 1-nz2]
镜像: 物体沿直线(2D?或^?3D?L.
P(n) = [ 1-2nx2 -2nxny -2nxnz]
[-2nxny 1-2ny2 -2nynz]
[-2nxnz -2nzny 1-2nz2]
Hxy的意义是x,y坐标被坐标z改变
Hxy(s,t) = [1 0 0]
[0 1 0]
[s t 1 ]
Hxz(s,t) = [ 1 0 0]
[s 1 t]
[0 0 1]
Hyz(s,t) = [1 s t]
[0 1 0]
[0 0 1]
M物体->像机 = M物体->世界M世界->像机
P->像机 = P物体 M物体->像机
UŞ变换: 如果满下式,那么映射f(a)是U性的:
f(a+b) = f(a) + f(b)
以及(qing)F(ka) = kf(a).
仿射变换: U性变换后接着q移,因此仿射变换的集合是U性变换的集.MU性变换都是仿变?但不是所有仿变换都是线性变?
如果存在一个逆变换可?撤消"原变?那么该变换是可逆的.
如果变换前后两向量夹角的大小和方向都不改?该变换是{角.只有q移,旋{和均匀~放是等角变?
{角变换会(x)保持比例不变.
q移,旋{和镜像是仅有的正交变?长度,角度,面积,和体U都保持不变.
刚体变换只改变物体的位置和方?不包括Ş?所有长?角度,面积和体U都不变.q移和旋转是仅有的刚体变?镜像q不被认为是刚体变换.
假设矩阵M有r?c?记法Mij表示从M中除ȝi行和Wj列后剩下的矩?昄,该矩阉|r-1?c-1?矩阵MijUCؓ(f)M的余子式.
]]>
graphics adapter contains the hardware needed store and display images on a monitor or other display device.
all of these adapters stored a representation of the displayed image in a bank of dual-ported memeory called a frame buffer.
the system uses one of the ports to read and write images into the frame buffer.the second port is used by the video scan out circuitry of the adapter to create a signal for the monitor.Direct3D only supports VGA compatible adapters.
Frame Color Lookup
Buffer Table
2D Pixel D/A
Engine Converter
Host CPU CRT Monitor
Direct3D abstractions include :device, swap chains, surfaces, and resource .
an application enumerates the devices available on each adapter, examining their capablities and supported display modes to find an acceptable device.
device is the object that expose the rendering operations of the hardware. its properties control the rendering behavior or provide information about rendering, while the device's methods are used to perform the rendering itself.
devices always contain at least one swap chain and a collection of resources used for rendering.
resources are application specific data stored in or near the device hardware for use during rendering Direct3D provides resources for scene geometry(vertices and indices) and appearance(images, textures, and volumes).
a surface is a resource containing a rectangular collection of pixel data such as color, alpha, depth/stencil or texture information.
a swap chain contains one or more back buffer surfaces where scenes are rendered and presented for display on the monitor, is a collection of back buffers.
a device's render target is the back buffer surface, with an optional depth/stencil surface, in which rendering will occur.
while all back buffers are valid render targets, not all render targets are back buffers.
it is also possible to have a texture as a render target allowing dynamic rendering effects.
to obtain a device object, Direct3D provides an object for device enumeration and creation. all other objects are created through the device. an application first obtains the runtime interface, then selects and creates a device from those available, and using the device creates the necessary resources for rendering.
most COM methods return HRESULT to indicate success or failure.
in windowed mode: the results of graphic rendering are presented for display insidei the client area of a window on the desktop.
Direct3D cooperates with GDI to make the results of rendering visible, using a ::StretchBlt operation to present a back buffer in the window's client region
in exclusive mode: Direct3D communicates directly with the display driver avoiding GDI. when an exclusive mode application is running, no other applications have access to display hardware and no GDI is visible on the screen.
HAL(hardware abstraction layer): has hardware acceleration of graphics rendering, making it fastest device type.
reference device: useful for debugging, when developing software applications that target new features not yet common in hardware, the reference device maybe your obly choice for generating an image with those Direct3D features.
the null reference device: does nothing and all rendering will result in a black screen.useful for manipulating resources on a machine that provides no hardware or software implementation of the runtime.
the pluggable software device: RegisterSoftwareDevice method,
each resource has Type, Pool, Format, and Usage attributes, each of these attributes of a resource are specified at the time the resource is created and remain constant for the lifetime of the resource object.
type attributes describes the kind of resource and is defined by D3DRESOURCETYPE.
pool attribute of a resource describes how it is managed by the Direct3D runtime and is defined by the D3DPOOL enumeration.
resources in the default pool exist only in device memory.
resources in the managed pool exist in system memory and will be copied into the device's memory by the runtime when needed.
resource in the system memory pool exist only in system memory
resource in the scratch pool reside only in system memory and are not bound by format constraints of the device
when a device is lost, all resources in the default pool are lost and should be released and recreated by the application when the device is regained.
Format attribute of a resource describes the layout of the resources's data in memory and is defined by the D3DFORMAT enumeration.
all resources have a format, but most of the format enumerants define the memory layout for pixel data.
D3DFMT_D24S8: D:depth buffer, S:stencil buffer.
Usage attribute describes how the application will be use the resource and is defined by a collection of flag bits. static reources are typically loaded with data once and used repeatedly without change, while dynamic resources are repeatedly modified bu the application.
a direct3D application starts by obtaining the IDirect3D9 COM interface pointer by calling Direct3DCreate9.
GetAdapterCount is to determine the number of adapters in the system. each adapter provides a number of video display modes. each display mode contains a screen dimention, refresh rate and pixel format and is described by D3DDISPLAYMODE structure.
the back buffer surface format of a device must be compatible with the display mode's format. CheckDeviceFormat method can be used to discover compatible formats.
GetAdapterModeCount: the number of display modes.
EnumAdapterModes: the display mode information.
GetAdapterDisplayMode: the display mode currently in use by the adapter.
two display modes can have the same width, height and format values but different values for the RefreshRates.
the CheckDeviceType method tells us if a particular combination of display format and back buffer format are valid for a device of a given type operating in windowed or exclusive mode.
with a valid device type, we can check the device's capabilities for rendering required by our application with GetDeviceCaps.
we can validate all the resources required by our application with the CheckDeviceFormat, checkDeviceFormat should be used to validate all the format of all resources used by the application: back buffer surfaces, depth/stencil surfaces, texture surfaces, and volume texture formats. if the application requires a depth buffer for viaiblitiy determination, it should use CheckDepthStencilMatch to find a depth buffer that can be used with its render target formats in a given display mode.
CheckDeviceMultiSampleType: check multisampling needs.
GetAdapterIdentifier: allows an application to identify a specific brand of adapter from a specific vendor.
D3DCREATE_ADAPTERGROUP_DEVICE: allows an application to drive both video outputs through a single device interface, allowing resources to be shaed for both outputs.
Direct3D uses single precision floating point computations. if an application requires higher precision from FPU.there are two choices, either the application can ensure that the FPU is in single precision mode when calling into Direct3D, or ir can request that the device preserve the application's FPU precision before Direct3D performs any floating -point operations and restore the precision before returning to the application.
D3DCREATE_SOFTWARE_VERTEXPROCESSING: select software vertex processing, which is always available from the runtime.the runtime used an efficient implementation of software vertex processing that is optimized for the CPU.
D3DCREATE_MIXED_VEVERTEXPROCESSING: select a combination of software and hardware vertex processing selected by SetSoftwareVertexProcessing mixed vertex processing is incompatible with a pure device and will fail if both are requested together.
D3DPRESENTFLAG_DEVICECLIP: restricts the results of a present operation to the client area of the device window in windowed mode.
D3DPRESENTFLAG_DISCARDDEPTHSTENCIL: instructs the runtime to discard the contents of the depth stencil surface after a call to Present, or when a new depth stencil surface is set on the device.
D3DPRESENTFLAG_LOCKABLEBACKBUFFER: requests a default swap chain with back buffer surfaces that can be directly accessed by the application.
in windowed mode: hDeviceWindow specifies the window whose client area will be used for presentation in windowed operation if hDeviceWindow is zero, then the focus window willbe used for presentation.
in exclusive mode, hDeviceWindow specifies the top-level window used by the application
D3DPRESENT_RATE_DEFAULT: instructs the runtime to choose a suitable refresh rate in exclusive mode, and uses the current refresh rate in windowed mode.
an application may wish to create a full-screen display on a specific monitor.
GetAdapterMonitor: return an HMONITOR handle for an adapter, once you have the handle to the device 's montior, u can determine what part of the virtual desktop is covered by the monitor.
each object in a scene is described by a collection of geometric primitives, such as points, lines, and triangles, with the device's action methods.
the pipeline converts geometric descriptions into pixels on the render target surface through the process of rasterization. graphic primitives are rendered in the order in which they are described to the devices similar to the way a CPU executes binary instructions sequentially through memory.
the entire graphics pipeline is controlled through the properties of the device.
the properites are manipulated through the Get/Set methods of the device.
every device has a distinct set of capabilities. Direct3D specifies an abstract machine interface, but does not provide a software emulation for a feature not provided directly by the hardware's driver.
a device provides specific information on its capabilities to allow an application to adapt to the capabilities of the device.
much of the behavior of the graphics pipeline is controlled by render states.
groups of device properties, including render states, can be cached and set as a group with state blocks.
the device object represents the rendering pipeline, we get images from the pipeline by supplying it with scene data and instructing it to render scenes into images.
scene data consists of geometric data defining shapes in space and data defining the appearance of the shapes.
we can break the pipeline up into large sections consisting of vertex data assembly, vertex processing, promitive assembly and rasterization, pixel processing, the frame buffer,a dn video scan out.
vertex data assembly section gathers together vertex components from a collection of data streams to assemble a compute vertex and its associated data.
vertex processing performs computations on each vertex such as the transformation and lighting of vertices.the processed vertex data is then assembled into graphic primitives and rasterized into a stream of pixels.
pixel processing performs a computations on each rasterized pixel to determine the final color of the pixel that will be written into the frame buffer.
frame buffer performs a read/modify/write operations combining processed pixels from the rasterizer with the existing pixels in the render ttarget.
video scan out reads the pixels out of the frame buffer for conversion into video signals displayed by the monitor.
scenes are rendered to the render target selected on the device. if the render target is part of a swap chain, the render target can be made viaible on the device through presentation.
the render target may not be part of a swap chain if the render target is a texture resource.
you present renderings for display through a swap chain as the last thing you do when rendering a scene.
a swap chain consists of one or more back color buffers into which images are rendered. a device is always associated with at least one swap chain and in windowed mode additional swap chains can be created to obtain multiple presentable surfaces.
the philosophy of the Direct3D object is to expose the capabilities of the hardware to the application programmer and let the application adapt to the hardware.
GetDeviceCaps: return the capabilities of an actual device, while GetDeviceCaps method on the Direct3D object returns the generic capabilities of a device.
IUnknown
IDirect3DVolume9
IDirect3DResource9
IDirect3DIndexBuffer9
IDirect3DVertexBuffer9
IDirect3DSurface9
IDirect3DBaseTexture9
IDirect3DTexture9
IDirect3DCubeTexture9
IDirect3DVolumeTexture9
resource objects are used as containers for the scene data rendered by the device such as primitive vertices, indices intto vertex arrays, surface textures and volumes.
two and three dimensional textures expose their contents as collections of sutfaces and volumes, respectively. the back buffer, depth/stencil buffer and render target properties of the device are also exposed as surfaces.
volume objects do not inherit from IDirect3DResoutce9 and therefore do not participate in the resource management exposed by this interface.
managed resources are assigned an unsigned integer priority, with higher priority taking precedence so that resources with a lower priority are discarded from device memory first.
non-managed resources always return a priority of zero. within the same priority, Direct3D uses a least-recently used strategy to discard old resoucrs in preference to newly created resources.
when the application attempts to use more resources than the card can hold while rendering a scene, Direct3D switches to a most-recently used first strategy for discarding memory.
resources in D3DPOOL_DEFAULT are not managed and are never discarded from device memory by the resource manager.
if u need to allocate new resources in the default pool after managed resources have been loaded, you should evict all managed resources before allocating new resources in the default pool.
resources in pool D3DPOOL_SYSTEMMEM are never located in device memory, so that they do not participate in resource management.
GetAvailableTextureMem: obtain an estimate of the available texture memory.
the GetDevice method returns the device with which this resource is associated. resources cannot be shared across devices.
GetPrivateData/SetPrivateData: allow an application to associate its own arbitrary chunks of data with any Direct3D erouces.
each distinct item of private data is identified by a GUID. all private data associated with a resource is freed when the associated resource itself is freed.
methods and functions in Direct3D that create COM objects, such as CreateDevice, add a reference to the object for the caller before they return the interface pointer, the application must release these objects when they are no longer needed to avoid a memory leak.
Device queries allow you to obtain information from the driver layer of the device, the two main uses for driver queries are for obtaining rendering statistics and event notifications from the device.
D3DQUERYTYPE enumeration gives the possible kinds of queries: vertex cache description queries, resource manager statistics queries, vertex statistics queries, event queries, occlusion queries, timestamp queries, time queries and cache utilization queries.
a query exists in one of three state: signaled, building, or issued.
Issue: used by the application to signal a state transition on the query.
the device driver can also change the state of a query when it has returned the data requested by the query.
the query will report the results for primitives issued between the begining of the query and the end of the query.
occlusion queries return the number of pixels that passed the depth test for primitives rendered between the begin and end of the query.
the resource manager statistics query returns a D3DDEVINFO_RESOURCEMANAGER structure, which contains an array of D3DRESOURCESTATS structure, one for each resource type.
the vertex cache is a memory cache close to the GPU that avoids access to vertex memory for a small number of recently used vertices.
PIX: is a performance measurement tool for DirectX applications. the remaining query types return data for performance measurements with tools like PIX.
the device object's properties control the behavior of the rendering pipeline while its methods supply data for the pipeline to render.
state blocks are COM objects that provide a way for your application to cache a group of device properties for later use.
each state block is associated with a device, returned by the GetDevice method.once a state block has been created, the state block can be applied to the device by calling apply on the state block, calling capture on an existing state block captures the current values of the device properties into the state block.
there are two ways to create a state block object and fill it with specific device property values.
the first way: call CreateStateBlock with D3DSTATEBLOCKTYPE value identifying the kind of state you want recorded in the block.
the second way of obtaining a state block object is to call BeginStateBlock, set device properties and then call EndStateBlock.
when EndStateBlock is called, each device property marked for capture is recorded into the state block. if a device property is set multiple times between BeginStateBlock and EndStateBlock, only the last value set in the property is captureed into the state block.
release the state block COM object when you are finished with a state block.
a pure device has a performance advantage because the runtime and driver do not have to keep a copy of the non-queryable state for the application.
scenes contain one or more objects that are positioned relative to each other and to the virtual camera that views the scene. such objects are called "models", and contain data that define their shape and apperance. the shape of a model is described by a collection of a simple geometric shapes, called graphic primitives.
when we draw a three dimensional scene, we need to solve the problem of visibility: objects in the foreground should occlude objects in the background. Two- dimensional applications use a painter's algorithm to determine visibility. Direct3D can use depth/stencil surfaces to resolve visibility.
using either a flexible vertex format or a vertex shader declaration, Direct3D describes colors, textures corrdinates,vertex blending weights and arbitrary shader data at each vertex.the vertex data is supplied to the device through a collection of streams, each associated with a vertex buffer.
the streams can be driven through a level of indirection with an index buffer.the streams can be driven though a level of indirection with an index buffer.
the scene is described to the device one primitive at a time. each primitive is rasterized into a collection of pixels written to the render target property of the device.
All Direct3D rendering is done within a scene. each successive frame draws the models at successive moments in time, as in cel animation used for cartoons.
frame rates of greater than 15 fps can be preceived as "real-time", the z-buffer algorithm solution to the visibility problem works at each pixels on the render target instead of on models, as models are rasterized, each pixel that is associated with z value is used to determine which pixel is cloest to the camera.the closer pixels are stored into the render target, while pixels farther away are discarded. the depth/stencil buffer holds each pixel's depth from the camera. to use the Z-buffer for resolving visibility, set RS_Z Enable to D3DZB_TRUE, RS_Z Write Enable to TRUE, and RS_Z Func to D3DCMP_LESS.
if the D3DPRASTERCAPS_ZBUFFERLESSHSR bit of D3DCAPS9:RasterCaps is set, it indicates the device has an alternative visivility algorithm for hidden surface removal that doesn't use a Z-buffer. how visibility is determined is hardware dependent and application transparent.
all back buffers on swap chains are valid render targets, but render targets are not restricted to back buffer surfaces, when the device is created or reset, the render target is back buffer zero of the device's default swap chain. when present is called on the device, the render target advances to the next back buffer so that after present returns, render target is back buffer zero again.
setting the render target to a surface other than a back buffer surface allows the device to render directly into an imanage surface instead of rendering into a swap chain's back buffer and using stretchrect to obtain the results of the rendering, which could stall the pipeline.
D3DPT_POINTLIST draws a collection of points,
D3DPT_LINELIST draws a squence of possibly disjoint line segments.
]]>
static bool ChangeChannelAndSetFocus(char* username,Channel channel);
q段代码看v来是没有事情?但是在接口设|上面存在不好的做法,如果是私密频道的?是有用username做私聊对象的,如果是非U聊频道的话,没有这个私聊对象了(jin).但是如果要调用这个接口的?q得l个q样的?
比如q样的用?->ChangeChannelAndSetFocus("",public_channel),
其实如果E微改变?sh)下接口设|的?好?
static bool ChangeChannelAndSetFocus(Channel channelchar* username = "");
q样使用默认参数好?
之前的调用就可以q样使用?
:->ChangeChannelAndSetFocus(public_channel),看v来舒服多?哎感觉编E功底还有待加强,q么单的问题,拿出来想?q有q么多学问呢.
q几天写?jin)一D函C?后来被一个家伙调用了(jin),他直接在我的函数里面修改,q样造成?jin)原来我函数里面的逻辑出现错误,现在x(chng),如果以后是这L(fng)情况,׃能允总在我的函数里面修Ҏ(gu)的逻辑和相兌?但是要留出接口给它调?至于具体的操?要我自己允许才能d,当然我可以帮他看看原来他d的代?但是q样也就没有什么意义了(jin).q种合作性的工作其要讲I这?不然很麻?ch)?
H然惌v来要记录点什?调试引擎的时?可以采用q样的方?打开上层逻辑的工E?然后引擎相关的代码拖拽C层逻辑所在的工程里面?然后在这个代码里面可以设|相x(chng)?以前M知道怎么找上层逻辑与底层引擎的入口,最q终于知道了(jin).呵呵.通过q样的方?可以在上层逻辑做某些操?然后兌q些逻辑相关的引擎代?
今天一直在被单态困扰着,事情是这L(fng):大厅和战斗场景里面都有两个类似的chatroom,q且他们各自使用的协议是不一L(fng).现在x(chng)他合qv来做成类似singleton的东? 之前遇C个问题，大厅和战斗场景不能在怺之间发送信息，后来我的处理Ҏ(gu)是将战斗场景里面d大厅里面的响应事Ӟq样的话Q在大厅里面发送私聊信息到战斗场景里面Q就可以接受C(jin)Q但是后来发现在战斗场景里面发送信息到大厅里面Q却接受不到Qƈ且这个时候，在大厅和战斗场景里面已经存在很多冗余代码?jin)，q样的处理让我感觉很不爽Q我在想有没有其它办法，后来h?jin)老大Q老大帮忙解释?jin)一下，后来Q在协议处理方式上面修改?jin)一下，才可以的。chatroom在不同的时候发送不同的协议好?jin)，q且在接受的时候分开大厅和战斗场景处理就好了(jin)Q这栯是用?jin)接受、和发送这两段废代码。对于singleton的了(jin)解还是不够好?nbsp;
之前遇到的问题，是聊天发送消息ƈ不立x(chng)C在本地Q而要先将信息转到服务端，服务端进行查询，如果能够扑ֈq个U聊对象Q那么就q回l信息，否则显C该用户不存在Q在战斗场景里面向大厅的某个玩家发送私聊信息以后，退出战斗场景以后，发现大厅里面q显C着在战斗场景里面发送的聊天信息。这样即使我按照常规思\L|络接受端的相关代码注释掉也不能辑ֈ消除大厅里面的聊天信息的目的Q感觉很郁闷。后来查看私聊信息方面的代码Q发现没有将信息有效地拦截，U聊信息在本C面还可以被显C，没有l过|络斚w的验证就直接发送过来了(jin)。在U聊信息斚w做一些处理ƈ且在退出战斗场景的时候做一些信息删除处理，q样可以达到目的了(jin)?
因ؓ(f)玩家对象位置数据是从服务器那边发送过来的,玩家从服务器获得数据需要一D|?所以需要加入一个插?来将现在的位|过渡到下一个的位置.q里可以采用多种插值方?
今天写了(jin)代码L写强制动作的处理Q结果发现很多处是有问题的，其中原因是没有将原来的代码作用看仔细造成?jin)，感到愧呢?br>
另外在编写交互性比较强的代码时Q一定要考虑自己的代码可能带来的影响?br>
更新某个文g比较ȝ(ch)的话,那么可以先采用原始文件删?q且重新全部更新的做法去?而不是用更新某个字段的方法去?
如果没有初始化声韌备成功的?那么应该可以弹出提示信息?q且原始声音音量设|ؓ(f)0.
之前在修改一个bug(在h物控刉板打开以后,一直克隆object,造成?jin)在引擎底层无法锁定vertexBuffer的问?,开始以为是内存泄露?在打开控制面板10分钟以后,游戏q然出错了(jin),q(sh)度想用一些辅助工h查找内存的信息是否出?后来发现在h物控刉杉K面克隆了(jin)q多的h物object,而这些object却没有及(qing)时地删除,造成占用q多的资?到最后连锁住vertex buffer也不行了(jin).在分析这个问题(sh)面的时?我还是一个劲地挖底层而没有考虑太多的逻辑层面的东?实在有点南辕北辙的味?值得反省.
在输入的使用上面,一般的游戏都采用directInput,即是U2,上次遇到的问题是,在一同事机器上面q行输入的时候就q入不了(jin),但是在别人的机器上面却可以运行好好的,后来l常查访发现:1)用配|文件写入的,之后再绑定到某个特定键的快捷键都不能?因ؓ(f)q些是读入内存(sh)?然后用add_binding来绑定到directInput讑֤上面?2) 在游戏里面用windows message的却可以正常使用.q入游戏的时?你不能输?而只能回?按tab键以?可以在登陆界面输入,如果不能按tab?则不能输?q入游戏以后却可以按不是从配|文仉面读取的信息,比如技?人物面板,仓库,以及(qing)地图{?
其中我个人徏议是,量不要用directInput.1)首先是directInput依赖于机?如果机器好的?那么没事?如果有事情的?那么也很难解?2)directInput里面采用?jin)hook,来拦截消?q样,q(sh)如直接用windows message来的?
http://www.gamedev.net/community/forums/topic.asp?topic_id=520366
详细的讨上面.
在调试逻辑没有发现什么错误提C的时?试着去调试引?可能问题发生得更加深,不在逻辑层面?而在引擎内部,q样p直接调试引擎?
在涉?qing)到渲染斚w的bug,量借用d3ddebugging来辅助查?q样可以方便解决问题.
q接错误QIID_IDirectSound3DBufferQ其实加上dxguid.lib好?jin)?br>U理d模式包括Q?br>wrap, border color, clamp,和mirror以及(qing)mirror once.
pDrawPrimInternal->SetTexture(hTex);
之前遇到的一个问题是Q显C玩家在地图上面到地图边界?jin)，q能重复l制地图Q给人的感觉好像是走不到边界Q走地图好像是在卯u里面一栗?br>后来才知道一般默认的U理d模式是wrap方式?br> get_device()->SetSamplerState(0,D3DSAMP_BORDERCOLOR,0x00ffffff);
get_device()->SetSamplerState(0, D3DSAMP_ADDRESSU, D3DTADDRESS_BORDER);
get_device()->SetSamplerState(0, D3DSAMP_ADDRESSV, D3DTADDRESS_BORDER);
解决bug有时候不如重?br>量让策划去理逻辑Q逻辑E序员尽量将重心(j)攑֜具体技术问题(sh)面，不懂的地方应该主动跟{划沟通，而不是自己想?br>代码可以先模拟出来，然后再放到具体的环境里面试Q独立写个demo之后再去试Q这样往往?x)比较快捷和省事情?br>学会(x)模拟机器执行代码的次序，q且提高阅读代码的问题，之前的时候，我劲去调试Q每ơ调试只Z(jin)解决一个问题，但是事实上，如果从阅M码开始的话，那么我可以省?天的旉?br>学会(x)调节~写代码时候的?j)情和?j)态，q样有助于进行高效编E，当心(j)情不好的时候，一定要集中全部的注意力Qƈ且将思\和流E作出图记蝲在纸上面Q这样可以强q自己全诏注?br>
今天遇到的文件打开问题是文件名路径有空格造成的，值得指出的是对于文g打开报错Q一般很难获得返回g息，但是可以通过GetLastError或者@Err,以及(qing)hr{来获得一些错误返回g息?br>
对字W串操作Q尽量要(g)查是否有I格Q如果字W串长度前面有空格的话，那么被认ؓ(f)所提供的字W串是空的。要避免q种情况的发生。对于字W串匚w要注意部分匹配的情况Q比?Tex", "Text",q里如果只匹配前面三个字W的话，那么两者是相等的，之前遇C个这L(fng)问题Q就是texture的信息被text信息所覆盖?jin)，D控g找不到正的texture而不能显C出来?nbsp;
今天遇到的问题是q样的，之前遇到的状态切换问题，是与状态切换无养I而与插值有养I在切换状态以后，收到一个插g|，q样又给角色一个新的位|，l别人看的效果就是角色先落地?jin)，之后又上升?jin)Q像坐电(sh)梯一栗不q从q里说明?jin)一个问?我在思考问题的时候，有欠周全的因素?br>
今天跟踪一个bug,在战斗的时候，出现卡死的情况，后来发现状态切换出C(jin)问题Q后来一直去查看战斗斚w的状态，但是q没有发C么结果，现在在想在战斗中的标志或者状态的时候应该设立一些标志集合，q样便于在某个时候检查一些状态的信息。现在很多的标志在游戏里面，局部的Q全局的，标志在角色上面的Q标志在物体上面的，q些都需要好好地理?br>
全方?多角度地思考问?而不能将思维局限在某处.学会(x)从大局或者小处去分析问题,大处不行的话,小?相关的信息串v?所有的信息应该是个串型或者链型的.之前在处理问题的时?把思\限制在太的范围内了(jin).以至于花?jin)很多时间才扑ֈ问题所?
Ҏ(gu)正常情况和非正常情况来获得数据比较，以确定正的程和结构。如果遇到错误的乱|那么q看正常的数|然后Ҏ(gu)q个比较来获得一些提CZ息?br>
今天犯了(jin)错误Q自己有最新版本，但是没有上传Q结果vss上面的版本把自己的最新版本给覆盖?jin)?br>
在发送消息的时候，有时候ؓ(f)?jin)防止频J发送消息，要加入一个cd旉Q防止因为网l问题，服务端收到玩家连l的信息?br>
物理pȝ, v = v0 - gt. 玩家上蟩的速度表示,而在下降,则是v = gt,按照自由落体q动来进?q且速度是个向量,q样如果在X,Z轴上面有初速度的话,那么可能出现抛物线的情?
]]>