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
|
KSpread column/row limits
=========================
Restrictions by sizeof(int); i <= 2^31-1:
-----------------------------------------
- PointStorage.h:
QVector<T>'s index is an int; Provides access to all cells.
Therefore, sizeof(col)*sizeof(row) <= sizeof(int); sizeof(col/row) = 2^16;
As QVector and QRect depend on (signed) int: max(col/row) = 2^15-1
-> Global.h: KS_colMax = 2^15-1, KS_rowMax = 2^15-1
-> Cell::Private: restriction by bit fields; 16 bits per column/row
-> Region.h:
uint qHash(QPoint); column coded in first 16 bits, row in the last
Can be extended by just considering the last 16 bits of the col/row,
even if bigger than 16 bits. Leads to hash collisions, but works.
-> Cell.h:
uint qHash(Cell); column coded in first 16 bits, row in the last
Can be extended by just considering the last 16 bits of the col/row,
even if bigger than 16 bits. Leads to hash collisions, but works.
-> Column-/RowCluster: restricted to 2^15
Can be extended easily by increasing one of the cluster levels,
preferably the first.
The limitation could be lifted, if a clustered array is used in PointStorage
and PointStorage::data(int), ::col(int), ::row(int) get ported to qint64.
- QRect: based on int
Minor issues, that would need adjustment:
-----------------------------------------
- Headers.h: The RowHeader's width should display all row numbers correctly.
- Sheet: QHash<QString, QRegion> *StyleRegion; QRegion is restricted (on Windows 95/98/ME)
Other restrictions:
-------------------
- Rect-/StyleStorage.h:
Based on QRectF, which is based on qreal (double or float). Numbers range
from 0..2^52 or 0..2^23, resp., with an exponent of 0, i.e. mantissa * 2^0.
Other exponents lead to "integer gaps"; step size != 1 (Actually, if a
number could be expressed with an other exponent without inaccurrancy, this
will be done, but that does not matter here. It's possible to encode it with
e=0.)
Performance:
------------
- Copying cell contents is done cell-by-cell. PointStorage is arranged in rows,
i.e. copying of rows could be optimized, but for columns a traversal of all
occupied columns is necessary.
Go the same way as in Cluster and follow a two-levelled approach? Could be an
option for further optimizations for copying areas, but would give headaches
at the cluster boundaries.
|
