1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
|
I just want to store some notes here in CVS as a compilation of
what I understood from some previous discussion on redesigning -- wanted
to do this before it is forgotten. It is very unlikely that I am
going to have the time/motivation to do this myself but hopefully
this can help give a start to whomever does the work. I don't think
anything here is definitive, and certainly if there are better ideas
than we should scrap these. For Norbert/Phillip/Ariya this is just
to help keep track of our ideas, and hopefully the future will see
many more KSpread hackers who can use this as a head-start in thinking
about KSpread design.
NOTE: when I say 'pointer' in this description, I'm thinking of some shared
object kind of class with a reference count -- not a literal pointer
that we would have to remeber to delete...
The problem needing solved is that the class KSpreadCell (which has
about a billion instantiations during a run of the program) is several
hundred bytes. This is a tremendous waste of space, and most of it
is spent holding information such as font type/size/color, which borders
to draw and line thickness, background color, and so on. Since this
information is going to be identical for a vast majority of cells, we
should find an efficient way to share data among cells.
The first idea is to break the format information into small classes,
such as one for font size/type, one for border information and so on.
The way to save memory is to use a 'flyweight' system in which cells
would have a pointer to the data, so cells with the same formatting have
the same pointer and the information itself has only a single instantiation.
At first, we can simply use the copy constructor of this class to implement
the sharing, and if it seems profitable in the long run these classes
can keep some kind of static mapping so that in the constructor a check
can be done to see if, for instance, helvetica font size 12 has already
been allocated in the past and use that pointer rather than allocating
a 2nd instance.
Next, these format objects would be collected objects I was calling 'styles'
A style would basically be one of every type of Format object and thus
would completely define the format of the cell. A style can be shared the
same way as a format object -- if two cells have all identical format
objects than they can share the same style object.
We had discussed two different ways of actually mapping these formats
and styles to particular cells.
One way is to simply have
each cell contain a pointer to its style. Rather than each cell using
200ish bytes to store the formatting, it has the single 4 byte pointer,
and then the the 200ish bytes is shared among all cells with that same
formatting information.
The other possibility is to map it by region. This involves storing
a map of some sort in KSpreadSheet to say, cells A2:E30 have this style,
column H has this style, etc. Here, the cell itself would store no
formatting.
If I remember correctly, we were leaning towards the second
method because of both the memory consumption, and because it is a simpler
way of handling setting formatting on a full column or row. However
this method will be much more complex to implement in a way that there
can be efficient lookup to retrieve the current style for a particular cell.
Some things to decide:
How fine grained to make the format objects?
- How much information to store in each format object. If there are a few,
large format objects, than each Style is very small, requiring only a
single pointer for each of these few format objects. However the data
sharing is not very efficient if between 2 cells the font color changes, and
there are 10 other pieces of data that are exactly the same
If there is too little in each format object, than we don't gain any
savings in memory because each additional type of format object results
in an extra pointer in each style object
There's probably much more that can be put in here.
BTW, I hope to stay involved at least a little with KSpread. It is unlikely
however that I will try to take on any large chunks of code unless I just get
in a random programming fit on a weekend :-)
-John
|
