| 1 |
=========== |
| 2 |
Buici NOTES $Id: NOTES,v 1.3 1997/10/15 22:33:28 elf Exp $ |
| 3 |
=========== |
| 4 |
|
| 5 |
by Marc Singer, [email protected] |
| 6 |
12 October 1997 (date of last revision) |
| 7 |
|
| 8 |
Why write another clock program? Have you looked at the published |
| 9 |
ones? At the time I started on this track, the best one I found |
| 10 |
doesn't draw a round face. A clock must |
| 11 |
|
| 12 |
1) Draw a correct representation of the clock face |
| 13 |
2) Support configurable clock face elements |
| 14 |
3) Respond to some sort of mousing request to display today's date |
| 15 |
|
| 16 |
Nothing came close, so I wrote Buici. The rest of this document |
| 17 |
describes some of the interesting design issues and notes about |
| 18 |
enhancements. |
| 19 |
|
| 20 |
- Fetching the current time |
| 21 |
I'm using gettimeofday () which returns the seconds since |
| 22 |
midnight, January 1, 1970, do the timezone and daylight savings |
| 23 |
corrections, and use this number of seconds to position the clock |
| 24 |
hands. I need to figure out how to do daylight savings |
| 25 |
corrections for non-US zones. I could use another call that |
| 26 |
already does the conversion, but this may add some calculation |
| 27 |
that I can omit if I do the math myself. |
| 28 |
|
| 29 |
- Rendering timer |
| 30 |
We set an interval timer to interrupt us every 100ms, or so. At |
| 31 |
this point we fetch the current value for seconds and decide if |
| 32 |
the time has changed. If so we draw into the m_pixmap and copy it |
| 33 |
to the screen. |
| 34 |
|
| 35 |
- Exposure |
| 36 |
Since we draw to the screen when the timer alarm fires we need to |
| 37 |
be concerned about race conditions. The simple solution is to set |
| 38 |
a flag when we receive an exposure and check this when we next |
| 39 |
receive a timer alarm. Thus, we draw the clock face when it |
| 40 |
changes or when we are exposed. |
| 41 |
|
| 42 |
- SIN COS and correct face representation |
| 43 |
The rendering routine uses floating point math to locate the |
| 44 |
positions of the clock face elements. I spent a few hours trying |
| 45 |
to make everything look good, but found that the hands were not |
| 46 |
always pointing straight at the hour marks. All of the default |
| 47 |
hands are longer than face radius, extending away from the mark |
| 48 |
that the hand indicates. For some reason not yet understood, the |
| 49 |
angle of this short line segment is not the same as the segment |
| 50 |
extending toward the mark. We solved the display problem by |
| 51 |
drawing lines that cross the face center in two pieces. |
| 52 |
|
| 53 |
- Efficiency |
| 54 |
There is something very inefficient about the drawing. For some |
| 55 |
reason the NCD pauses a lot when we use the Buici clock. I'm |
| 56 |
thinking we either need to batch the strokes, use more backing |
| 57 |
storage for the fixed face elements, or check that we are updating |
| 58 |
only once a second. |
| 59 |
* This has been resolved. On the NCD, the problem appears to be |
| 60 |
drawing. Batching the line segments into a fewer number of |
| 61 |
requests did not affect the clock performance. Using a backing |
| 62 |
pixmap for the clock marks does help. We note that this |
| 63 |
requires that the server has memory for the backing pixmap. |
| 64 |
Should the clock be very large, the backing pixmap may be |
| 65 |
excessively large. |
| 66 |
* We need to evaluate the drawing performance of other servers. |
| 67 |
It may be OK to fallback to drawing marks every second because |
| 68 |
more contemporary servers are more efficient. |
| 69 |
* An alternative solution to the size of the backing store is to |
| 70 |
use color mapped bitmaps for marks and BLT them by color instead |
| 71 |
of using XCopyArea. As long as we have fewer than depth layers, |
| 72 |
the extra work may be justified. |
| 73 |
|
| 74 |
|