Packaging component code for automatic download

ISVs and authors of COM Objects for the internet should package their implementations so that they may be downloaded automatically by web browsers such as the Microsoft Internet Explorer. Such objects will be downloaded, for instance, when parsing the OBJECT tag in HTML(3). For details, see the OBJECT tag specification.

Interpreting the "CODEBASE" URL

The "CODEBASE" attribute in an OBJECT tag contains a URL pointing to the implementation of a given COM object. This URL is of critical importance for component download, because it must specify all files necessary to implement a particular COM object. HTML authors can author "CODEBASE" URL to point to one of three file types. Component developers should choose one of the packaging schemes below for their COM Objects:

  1. A single PE (portable executable, e.g. an .OCX, a .DLL, or a .EXE): This single executable is downloaded, installed, and registered in one fell swoop. This is the simplest way to package a single-file OLE control, but (a) it will not use file compression, (b) it will not be platform independent except with HTTP.
  2. A .CAB (cabinet) file: This file contains one or more files, all of which are downloaded together in a compressed cabinet.<Footnote Code Text> Care must be taken so that the cabinet file contains only those files that must necessarily be downloaded (e.g. the OCX executable itself). Any additional helper DLLs (e.g. MFC) may have already been installed and if so should not be bundled into the cabinet(4). Exactly one file in the cabinet is an .INF file providing further installation information. This .INF file may refer to files in the .CAB as well to files at other URLs. This mechanism requires authoring of a .INF and packaging of a .CAB file, but in return it provides file compression. It will not be platform independent, however, except with HTTP format negotiation.
  3. A stand-alone .INF file: This file specifies various files that need to be downloaded and setup for the OCX to run. The syntax of the .INF file allows (a) URLs pointing to files to download, and (b) platform independence (by enumerating files for various platforms). This mechanism provides platform independence for non-HTTP servers.

Registry settings and self-extracting .exes

It's recommended to use self-registering code for Internet Component Download, because the .INF format used by Internet Component Download (see below) does not provide syntax for changing registry information (for security reasons). For .OCXs, .DLLs, and .EXEs marked as "OleSelfRegister" in the Version resource, Internet Component Download will try to run self-registration. For .DLLs, this means loading the .DLL library and calling the DllRegisterServer entry point, if available. For .EXEs, this means running the .EXE with the run-time parameter of "/RegServer". This ensures that COM Objects implemented as local servers (e.g. winword.exe) are registered correctly. If an object is not marked as "OleSelfRegister" but registration is necessary, or if it is desired to over-ride the "OleSelfRegister" flag, one can add the following to an .INF file (see .INF setup-script format below):


[foo.ocx]
RegisterServer=no ; don't register even if marked OleSelfRegister
	or
RegsiterServer=yes ; register this even if not marked OleSelfRegister. This is the typical workaround for getting old
			; controls to register

Code that is downloaded via Internet Component Download may be a self-extracting .EXE because Internet Component Download ignores the "OleSelfRegister" flag if the main URL for code download points directly at a .EXE file. In this case it is assumed that this is a self-registering .EXE, and this enables self-extracting .EXEs to work correctly as long as they ignore the "/regsvr" command-line parameter. Supporting self-extracting .EXEs enables very complex setup mechanisms to be launched automatically via Internet Component Download. However, if a self-extracting .EXE is called via this mechanism, then any components that it installs will not be automatically tracked by Internet Component Download (see Appendix on ModuleUsage section in registry). Such components are permanently installed and are not marked by Internet Component Download for future cleanup.

Including version number in the "CODEBASE" URL

Besides the actual address of code, the "CODEBASE" URL may also include an optional version number using the following syntax: "CODEBASE=http://www.foo.com/bar.ocx#Version=a,b,c,d". The Internet Component Download mechanism will download and install the file only if the specified version number is more recent than any existing version of the same file currently installed in the system. (see Appendix on registry details for more information). If a version number is not specified with a file, it is assumed that any version installed on the system is recent enough(5).

If the version number specified in the CODEBASE attribute is "-1,-1,-1,-1", then Internet Component Download will always try to download the latest version of the desired component. Note that this can be a costly effort involving many network transactions, especially if the Internet Search Path is searched for newest versions of an object (see below for details on Internet Search Path). Note also that because of the Internet Search Path, it is possible for the Component Download service to try to download code in the absence of a CODEBASE attribute. In fact, if the CODEBASE attribute is the URL fragment "#Version=-1,-1,-1,-1", then there is no specific location to download code from, but the Internet Search Path will still be searched to find the latest version of an object(6).

Platform independence and HTTP

When code to downloaded is on an HTTP server, the HTTP Accept header MIME request type may be used to specify which platform the code is to run on, thus allowing platform independence of the "CODEBASE" URL.

The following MIME types will be used to describe PEs (portable executables - .EXE, .DLL, .OCX), cabinet files (.CAB), and setup scripts (.INF):(7)
File description MIME Type
PE (portable executable) - .EXE, .DLL, .OCX application/x-pe-%opersys%-%cpu%
Cabinet files - .CAB application/x-cabinet-%opersys%-%cpu%
Setup scripts - .INF (platform independent) application/x-setupscript

%opersys% and %cpu% above will specify the operating system and CPU for the desired platform downloaded components will be executed on. For example, the MIME type for a Win32 cabinet file running on an Intel® x86-architecture processor would be application/x-cabinet-win32-x86.

The following are valid values for %opersys% and %cpu%:
Valid values for %opersys% Meaning
win32 32-bit Windows® operating systems (Windows95 or Windows NT)
mac Macintosh® operating system
<other> will be defined as necessary
Valid values for %cpu% Meaning
x86 Intel® x86 family of processors
ppc Motorola® PowerPC architecture
mips MIPS® architecture processors
alpha DEC® Alpha architecture

When the code is on a non-HTTP server (e.g. at a local LAN location), a .INF file can be used to achieve platform independence by specifying different URLs for files to be downloaded for different platforms. (see the section below on platform independence in .INF files)

.CAB format

The .CAB format used for Internet Component Download is a non-proprietary format based on Lempel-Ziv compression. The Microsoft Internet SDK includes a free tool called "diantz.exe" that will package cabinet files into this non-proprietary format. There no specification of this .CAB format publicly available, although such a specification will be distributed as soon as possible.

Use of the DIANTZ.EXE tool for creating .CAB cabinet files

The DIANTZ.EXE tool takes a .DDF "directive file" specifying which files to combine into a cabinet. The syntax for using this tool from a command line is:


DIANTZ.EXE /f <directive file.ddf>

The example directive file below, CIRC3Z.DDF, would be used for creating a cabinet file containing two files - circ3.inf and circ3.ocx. It should be straightforward to add to this list of files...


; DIAMOND directive file for CIRC3.OCX+CIRC3.INF
.OPTION EXPLICIT							; Generate errors on variable typos
.Set CabinetNameTemplate=CIRC3Z.CAB 
;** The files specified below are stored, compressed, in cabinet files
.Set Cabinet=on
.Set Compress=on
circ3.INF
circ3.OCX

Note It is possible to use the "code-signing" utilities to sign entire cabinet files using a digital certificate. However, in order to do this, it is necessary to add the following linesto the .DDF file before the list of files for inclusion in the cabinet.


;Reserve space for PKS#7 Code Signature
.Set ReservePerCabinetSize=2048

If a cabinet file is signed, it is assumed that every file inside the cabinet is trusted, including .INF and .INI files. This has two advantages:

  1. It is now possible to include powerful .INFs inside a trusted cabinet
  2. By signing an entire cabinet the time for verifying digital certificates can be sped up due to the cabinet compression

.INF setup-script format

Here is a sample .INF file that demonstrates the syntax that is understood by the Component Download service. Note: Only the .INF syntax below may be used to write setup scripts for Internet Component Download. Due to security reasons the system standard (SetupX) .INF setup is not called to install components with setup scripts, and instead the limited INF syntax below is the only legal format for Internet Component Download. See Future Directions below for plans for eventually supporting other .INF formats.


;Sample INF file for CIRC3.OCX
[Add.Code]
circ3.ocx=circ3.ocx
random.dll=random.dll
mfc40.dll=mfc40.dll
foo.ocx=foo.ocx

[circ3.ocx]
; lines below specify that the specified circ3.ocx (clsid, version) needs to be installed on
; the system. If doesn't exist already, can be downloaded from the given location (a .CAB)
; note: if "thiscab" is specified instead of the file location, it is assumed that the
; desired file is present in the same .CAB cabinet that the INF originated from
; otherwise, if the location pointed to is a different .CAB, the new cabinet is also downloaded and
; unpacked in order to extract the desired file
file=http://www.code.com/circ3/circ3.cab
clsid={9DBAFCCF-592F-101B-85CE-00608CEC297B}
FileVersion=1,0,0,143

[random.dll]
; lines below specify that the random.dll needs to be installed in the system
; if this doesn't exist already, it can be downloaded from the given location.
file=http:// www.code.com/circ3/random.dll
; Note that the FileVersion is option, and it may also be left empty, meaning that any version is ok.
FileVersion=
DestDir=10


; DestDir can be set to 10 or 11 ( LDID_WIN or LDID_SYS by INF convention)
; this places files in \windows or \windows\system, respectively
; if no dest dir specified (typical case), code is installed in the fixed occache directory.

[mfc40.dll]
; leaving the file location empty specifies that the installation
; needs mfc40 (version 4,0,0,5), but it should not be downloaded.
; if this file is not already present on the client machine, component download fails
file=
FileVersion=4,0,0,5

[foo.ocx]
; leaving the file location empty specifies that the installation
; needs the specified foo.ocx (clsid, version), but it should not be downloaded.
; if this file is not already present on the client machine, component download fails
file=
clsid={DEADBEEF-592F-101B-85CE-00608CEC297B}
FileVersion=1,0,0,143

Platform independence in .INF files

It is possible to create platform-independent setup scripts that pull files from different locations depending on the desired platform. Internet Component Download .INF files will use a scheme similar to the one described above under "Platform Independence and HTTP". Specifically, a sample platform-independent .INF file would include a text such as the following:


[circ3.ocx]
; lines below specify that the specified circ3.ocx (clsid, version) needs to be installed on
; the system. If doesn't exist already, can be downloaded from the given location (a .CAB)
file-win32-x86=file://products/release/circ3/x86/circ3.cab
file-win32-mips=file://products/release/circ3/mips/circ3.cab
file-mac-ppc=ignore
	; the 'ignore' keyword means that this file is not needed for this platform

clsid={9DBAFCCF-592F-101B-85CE-00608CEC297B}
FileVersion=1,0,0,143

Thus the "file=" syntax used in the .INF file is expanded to "file-%opersys%-%cpu =", allowing the .INF file to specify multiple locations where various platform-dependent modules can be found and downloaded. See the section above for valid values for %opersys% and %cpu%.

© 1996 Microsoft Corporation