To try and clarify. In 1.3, only the DTMF and Noise Generators remember their duration from one Audacity session to another when generating at a point. The others will default to 30 seconds. If you drag a region, remembering a length is irrelevant.
None of the generators remember the selection format. The design decision was made that when generating at a point (inserting, so increasing the length of the track), the time format is in whole seconds irrespective of your time selection format in
Selection Toolbar. If selecting a region and generating into that (so keeping the track length the same), the selection format is always hh:mm:ss + samples. The rationale was that you can almost never drag out a region that is a complete second (unless you enable "Snap to" in Selection Toolbar), so show the length you have actually dragged out down to the finest granularity, which is samples.
1.3 has "TimeTextControls" for time selection in generators - those spinning boxes for each character. This makes it much easier for visually impaired users to manipulate these time values. There is a slight penalty for sighted users that you cannot now simply overtype an arbitrary value unless it's the selected (white) time value.
In 1.2.6, if you wanted to work in samples selection format or something other than seconds (View > Set Selection Format), then having the generators in seconds was a pain. Unfortunately, 1.3 still has the problem that the generators do not default to the selection format you chose in Selection Toolbar. A number of users have suggested they should. I think on balance that when generating at a point, generators should default to the current Selection Format. When generating at a region, they should default as now to hh:mm:ss + samples. However if the user wants to change each generator to some other format, that format should be remembered. So in your case you might want to default them to hh:mm:ss + hundredths. None of this is a bug though - it's "design decisions".
So Ed, pending such a change, there isn't any easy one step way to generate ".8" seconds. I assume you want to overwrite a region with silence as you are seeing "samples" in the Generate Silence dialogue. You may find it easier to select the region you want to generate into in Selection Toolbar. Right-click over Selection Toolbar and choose "hh:mm:ss + hundredths". You only have to make this change once in Selection Toolbar, as opposed to having to make it every time in the Generate dialogues. Set the End/Length button in Selection Toolbar to "Length". Click the last digit but one from right and type "8". Then just Generate > Silence and hit OK, disregarding what it says about "samples".
Had you wanted to generate whole seconds, you could check "Snap to" in Selection Toolbar. Then whatever region you dragged, it would always start at a whole second and its length would be a whole second. You could again ignore the hh:mm:ss + samples display in the generator.
Gale