Martino Ferrari
faef19ba60
The offset of a following group was clamped to `retained` -- measured from the newest sample -- so a pan (or a zoom-in anchored near an edge) could put the window's *right* edge exactly on the oldest retained sample. That leaves the window's entire span to its left over data the scope never had: an empty margin as wide as the view. The bound is about the whole window, so it is `retained - span` -- but only while the window *can* sit on the data. When the members retain less than a window's worth (a ring still filling, or one just trimmed) nothing avoids a margin, and the older placement is the honest one and stays: right edge on the newest sample, the shortfall on the left. Test: a_following_group_keeps_its_whole_window_on_the_data -- asks to scroll ten spans past the oldest sample and asserts the window is still a window, with its left edge on the data. The existing `a_following_log_group_holds_decades` codified the old bound (`max == edge/10` after asking for a decade of pan with far less than that retained); its log-space subject is unchanged and the offset assertion is now the invariant above.