Page Comparison

...

Hopefully this illustrates the problem.  Here, with target-file-name specified, file refers to the expected data file, what etlunit calls the target, and target-file-name refers to the actual file, what etlunit refers to as the source.  Beyond that, though, since the file assertion shares it's implementation with the database assertion, the two file types, source and target, refer to target-file-name and file, respectively.

So, to address this, we scuttled the existing file assertion and made it match database assertions so a single definition fits both.  The minimal assertion above is represented like this:

assert(source-file: 'name');

Pretty simple.  Just copy and paste, right?  Not quite.  In the case when the source and target names differ, it has to be handled like this:

assert(
	source-file: 'target',
	target: 'name',
	source-reference-file-type: 'src',
	target-reference-file-type: 'tgt'
);

In this case, the attribute that was named file is renamed to target, and the attribute that was named target-file-name is now source-file.  The fact that it isn't a simple search and replace illustrates why it was inconsistent in the first place.

Assertion files

Beyond the operations themselves, the target files will have to change any time the columns in the assertion don't match the target fml.  In < 3.2, the assertion data file (the expected file - stored in the local data folder) had to match the final definition for the assertion - since the column specs were completely determined by the source before ever opening the data file.  Now, however, the expected data file must 100% match the target file type, and any columns specs, etc, are processed on the target afterwards.