The way I am doing it is quite strange forward and involves 2 steps.
- Create base-line database.
- Create and maintain a single change script file which will be checked into source control.
Point A has no semantic difference.
For point B, it is not necessary because when you apply the script which has a version number in the source control (I am using SubVersion which will generate a unique repository-wide incremental number each time you make change(s) to the source repository and our build number is tied to this number), and therefore you know what is the database schema version after the script applies. If customer reports a problem with build 213, I know exactly what the database schema should be and all the source code that associated with this build.