Processing Modules often change over time. We want to classify the possible types of changes based on their effect on the produces output datasets. We need to consider three types of changes for modules:
Typically, these differences in result datasets stem from the following sourcecode changes for the module:
Each module installation must specify in a properties file the version number range of input datasets it requires. Each module installation must specify in a properties file the exact version number of the output datasets it will produce upon successfull execution.
Our proposed version numbering scheme is derived from a widely
used source code version numbering standard with three digits.
A version number consists of three digits, separated each by a dot. An example is 0.0.1 or 17.3.12. Numbers are not prefixed by zeros, so comparison must be done numerically not alphabetically.
The names of the three digits are: “major version”.“minor version”.“patch level”
We want to keep minor version number API compatibility. In detail:
Each of the three parts of the version number may be specified using a regular expression. The following syntax should be supported:
We need to track the version number of each module in the results storage, so that results can be reproduced.
Note: The first example is discouraged for normal use, because this example does not support newer patch levels of the same module. Explicit versioning should only be used to reprocude previous results. A workflow that can only use an exact version of ShadingCorrection:
Note: This is the most common example for typical normal use. A workflow that can use any minor version and patch level of Neighbor Tree Analysis:
Note: This is an example for modules that can support several different API-incompatible input formats: A workflow that can use several major versions and any minor version and patch level of CellProfiler features: