Change-Id: I632d617cad76485f7e1f57daa7db4d4cfa775e8b

Change-Id: I9f200377c1f5bf44eb4ba60c272b02ae3d632a19

Change-Id: Iee1b3ff637d10c6bd2d2342db518a2fc986327b7

This speeds up the following scenario:

1) type =B1 in A1. Leave A2 empty.
2) Select A1:A2 and Ctrl-C to copy.
3) Select A3:A50000 (or longer if you so wish), and ctrl-V to paste.

This causes the broadcaster storage array in column B to be heavily
partitioned due to the empty cells interspersed between formula cells
in column A.  Without tracking the column position this would cause a
O(n^2) complexity algorithm.

Change-Id: Ic2f23c2c2bea3353c517faa73fe5412c7528bd95

Change-Id: I3627a83669b6a69c299aef96b8b2ead1352eabe2

Change-Id: I112a4be56910c07007b28715336fcd82d56bb313

Change-Id: I6f358f89348a23b3c70f8b7db8a1f01d659393ba

The use case that would previously make Calc freeze.

1) type in "test" in A1.
2) Select A1:A2 and Ctrl-C to copy to the clipboard.
3) select the whole column B.
4) paste special, select "Add" operation and click OK to paste.

With this change, the above scenario no longer freezes Calc.

Change-Id: I98b1f1e6c25b853aa884e7598c04b1457fab4636

Change-Id: Iaa7003cabee4630e4fbb8fb1cd114d04b4e9e33c

This currently only affects document import via orcus, but it'll be
good to put this in place.

Change-Id: I8cc6d54aba6fab1f2774127f92c2a764f7b690fb

With the removal of existing cells. So, this change is not enough to
make this operation fly.

Change-Id: Ic468375f6d0c28e2cc7d5391fb0565d53ee7fb4e

Change-Id: Ib9b734ac9800b1ecf005dcd3a66beac6fcfc3a4f

UndoToDocument and its nested methods are tightly coupled with the CopyTo
counterparts, and they both are used in the same undo / redo code....

Perhaps we can clean this up to reduce redundancy in the future.

Change-Id: Id3aec60914ac3a87acbb53f889896a5850047c59

This removes O(n^2) slowness from undo and redo of a large array of
cells with empty cells in between.

Change-Id: Id87c8ac8986c97ddf305f849a4752f0f92cc5899

To reduce the number of function parameters lower.

Change-Id: I816bd7c27070597f5fcdf69bdc7d981279390ca6

The basic idea is the same as the one I did for CopyFromClip.  This prevents
the O(n^2) algorithm from being invoked during undo object creation.

Change-Id: Iac46663b0c25db2f00ffa3ab95a4dcf2857252ce

This alone makes AdjustRowHeight() fly on large ranges, by eliminating
the silly O(n^2) algorithm.

Change-Id: Id2693aab7fb9b2a5575c6390fcaae5a404549962

This change reduces the duration of this method from somewhere in the
ballpark of 10 seconds down to a tiny fraction of a second.

Change-Id: I0278dc06a4f134b43cd08bd94693b6dec4893f1f

Change-Id: Iad2e69e4eb61167ad85581a83b19e3798c5edfd0

Use block position hint to avoid re-starting search in the cell text
attribute array.  This change alone cuts additional 10 seconds off of
the previously mentioned use case, by reducing the duration of
HasEditCells() call from the previous ~10 seconds to a tiny fraction
of a second.

Change-Id: Id9f951cd235a88928c900619b6b66d7b8a057e0c

This avoids re-starting the search for insertion position in the cell
text attribute array, on every single cell insertion. Instead of always
searching from the first position, it starts the seaarch from the previous
insertion position.

This reduces the paste time from the previous 3.5 seconds to 1.7 seconds
on my machine.

Here is the scenario.

1) Put =A1 in B1, and leave B2 empty.
2) Select B1:B2 and copy it via Ctrl-C.
3) Select B3:B50000, and paste via Ctrl-V.

Because of the empty cells interspersing the destination range, this makes
the data array highly partitioned, which makes the position lookup very
very expensive without the position hint.

Still, I was hoping the duration would become a fraction of a second. 1.7
seconds still seems a bit too slow...

Change-Id: I7742ce7e22935b6d0e082e4569d53dbd2072c4e5

so that I can reference it in another class.

Change-Id: I035cc1d37ec219c3e3da2e180a0337fb3c6fba8a

Change-Id: I9f56c7a54405a805cb843321bdab660a3ab22fd6

This moderately improves the performance of pasting a large array
of formula cells.

Change-Id: I27a59ff348cb715df15db80693cb7d193e67ec2e

Change-Id: I0b4deec7f4b10c7b14e482f61c667972caa30bc8

Change-Id: Ida0542583242ebcdd7f1448931e31656aa542457

It's not used yet, and is currently empty.

Change-Id: Ia377994ae042fc8b19d269adc2f94830422770f1

Change-Id: I5e108efa0219620696d15a304d2f4e3566d58d57

Change-Id: I2fdf38b23e74a9f4cad7abd9a943e8817f39474c

Change-Id: I90e419d19f4565f7f20603358308fd7aeba83190

Change-Id: I945c7ae834393c760d8a552043c391ff0eb9ce22

Change-Id: I615c9f9e95a44149c8079754e9aa6497f34c76cf

This massively speeds up the deletion of a large group of adjacent formula
cells. As an example, on machine, deletion of formula cells over B2:B109101
(109100 cells in total) got reduced from 4.7 seconds to 0.09 seconds).

Change-Id: Ib72da42a6644421601111907cf7c899d828c2996

This is to avoid repeated calls to set_empty() on the broadcaster array
which causes array reallocation on each call. Instead, we can store the
segments of broadcasters to be deleted later, and call set_empty() on
those segments to reduce the number of calls to set_empty().

The meat of this is not implemented yet, as EndListeningContext is still
empty.

Change-Id: I914bc2881acee3902c4fec0f6c21aaf4d2489df8

Change-Id: I4c34cd4b352dfafe7f423ab8e85f0d93c0368349

I screwed this up earlier which made this operation take almost 50 seconds
to complete. It should finish in 0.00001 seconds or less.  The test checks
against 1 second, which should be enough of a buffer for slower machines.

Change-Id: I9923033045111c75a0740b6bb30a518fe93e01d2

Since they are always in sync...

Change-Id: Ic07f57e1804d76ae801c7947d18df5fd0d729632

This block will be used to merge the text width and script type vectors
that are currently separate.  Text widths and script types should really
be stored together.

Change-Id: I6783769cc03180b513319e0a98a0773bd20ba941

This header hosts custom element types and blocks used in
multi_type_vector.

Change-Id: I32481a2354d8c66eb877b4005a260a79183cd714

Change-Id: I2c20a52e7f7688359172c9ccbd46b4b012bd4c02