This very good reference answer notes how the commands
\floatpagefraction can be used to fine-tune the placement of floats. Each of these defines a ratio whose denominator is ‘a fraction of
\textheight to determine how much of a page one or many floats may or must fill in order to allow placement.
When calculating whether a page is filled enough, it seems logical that LaTeX would use a box whose height corresponds to the height of the float, i.e. the figure or table plus caption plus separators. It is not so clear what LaTeX considers the width to be.
How the width of a float is defined is of interest. Consider a very high but very thin image or table (please ignore whether or not this would be good typography). Its width could be
0.2 * \textwidth while its height is
0.6 * \textheight. Will this float be considered to occupy
0.12 of the page or
0.6 of the page — i.e. using the standard settings (
\floatpagefraction = 0.5) would it be allowed on a float page or not?
Looking through the tables and figures in the thesis I’m writing suggests different treatments of tables and figures. Figure captions are always extended to
\textwidth while table captions only span the actual width of the table. If this makes a difference, please mention it.