File Sequencing and How Files are Located
The File table in a Windows Installer package has a Sequence column that begins with 1. Besides determining in what order files are to be installed, this column serves another, relatively more important function: determining where the source files are located.
In concert with the Media table, a range of sequence numbers identify in which cabinet – if that range of files is even compressed – the files are located. The Media table has a column named LastSequence that is inclusive of the maximum sequence number of a particular file range. All files in the File table with a sequence number greater than the previous Media table entry’s LastSequence and less than or equal to another Media table entry’s LastSequence are contained in the latter media. To put that a little more plainly, see the following color-coded tables.
The files a.dll and b.dll may be compressed into AB.cab, while c.dll and d.dll will be compressed into CD.cab. CD.cab begins with a hash symbol (#) because, as documented for the Cabinet data type, that cabinet will be embedded as a stream within the .msi package. AB.cab would identify a cabinet in the root of the source tree specified in the Directory table.
You might also notice that I said the files a.dll and b.dll may be compressed into AB.cab because, depending on the Word Count, those files aren’t explicitly attributed to be compressed. The Word Count Summary Property describes the default setting for whether files are compressed or uncompressed. As documented in Compressed and Uncompressed Sources, the Attributes column values of the File table override the Word Count property value. So, if the Word Count property specified that all sources were compressed then files a.dll and b.dll would be compressed into AB.cab; otherwise, if the Word Count property specified that all sources were uncompressed then only files c.dll and d.dll would be compressed into their respective cabinet.
Now, for example, b.dll needs to be patched. A patch must carry the new copy (or binary delta patch) of b.dll so it must change the file sequence of b.dll to match up with a new entry in the Media table to be added by one of the patch transforms for a particular product, as discussed in more detail in What’s in a Patch. The transform beginning with a hash symbol (#) would make the following changes from the previous tables.
Re-sequencing this file locates it in the the patch in the embedded stream
named P1.cab. The attributes are changed to
msidbFileAttributesCompressed (16384), the latter of which
to override whatever the default Word Count is set for the .msi package
being patched. The Word Count property value of the .msp file is used to
identify the minimum version of Windows Installer required to install the patch
and does not specify compression. Prior to Windows Installer 2.0 you had to
guarauntee uniqueness of file sequence numbers for both the installer package
for all patches, but starting with Windows Installer 2.0 you need only ensure
that file sequences numbers are correct for a given package since Windows
Installer 2.0 and newer will automatically adjust file sequence number as
When the transforms are applied from a patch and the files are re-sequenced, a reinstall of the affected features or all of the product is performed as documented in How Patching Works. The value in the Source column of the Media table for DiskId 3 is used to identify the location of the .msp file in order to locate the embedded cabinet.
Do note that the default schema of the File and Media tables allow for a maximum of 32,767 files including all changes made by any applicable patches. In the last set of tables, Windows Installer would view the product as now having 5 files. The reason is that the column data type is only a 2-byte signed integer. For more information about how to author a package with more initial files or to allow for many more patches, read Authoring a Large Package.