DataSpell DataFrame Preview vs pandas‑profiling: Which Avoids Memory Leaks for Large Datasets?

When repeatedly inspecting large pandas DataFrames in DataSpell, the built‑in preview keeps a reference to the original object until the tab is closed, potentially causing a memory leak. An alternative is to generate a static report with an external library such as pandas-profiling, which does not rely on DataSpell’s in‑memory viewer.

The main constraint is the trade‑off between interactive exploration and memory consumption. The DataSpell viewer offers quick, in‑app previews but retains cached rows and metadata; the external profiling tool produces a disk‑based report and frees memory immediately after generation.

Currently, DataSpell 2025.1 lacks an API to force cache eviction, leaving the decision between closing the viewer tab or switching to an external report unresolved for teams that handle very large datasets.

Which approach provides lower memory usage when repeatedly opening large DataFrames?
Does DataSpell expose any programmatic way to clear the viewer cache without closing the tab?
What are the disk‑usage and reproducibility trade‑offs when using pandas-profiling instead of the built‑in preview?