There are a number of operations that are made much easier and/or more efficient by simply having data be passed around a file on disk instead of as FME features. When dealing with large amounts of features going into a PythonCaller, for instance, it can be preferable to dump them all into a Parquet or Arrow file and decode that file with Pandas or some other dataframe library, rather than extracting attribute data feature by feature in Python. Some other use-cases, like Emailer, also often make use of temporary files.
The issue is that having FeatureWriter recieve both a path from TempPathnameCreator and the features it needs to actually write doesn’t work, that path needs to be joined on each feature in any of various clumsy methods. Given most uses of TempPathnameCreator are intended to feed into FeatureWriter, I would suggest that “Temporary local file” be a valid dataset option for FeatureWriter, which would simply generate a temporary path in FME_TEMP, write all features to it and allow the file name to be retrieved as the _dataset
attribute under the Summary tag, as is already the case. It would otherwise work in exactly the same way as the combination of TempPathnameCreator and FeatureWriter, but simpler to setup.
Any scenario where it is more practical to pass features between transformers as single files rather than large collections of features might benefit from this.