Extending the selection with Ctrl-arrows is different from all the other
cases, as we need to combine both the selection anchor and the current
cell coordinates when doing it.
This means that we can't reuse the same PrepareForSelectionExpansion()
helper for this case, so this function is not useful finally and this
commit removes it entirely. It also replaces GetCurrentBlockCornerRow()
and GetCurrentBlockCornerCol() functions with GetExtensionAnchor() which
combines both of them.
Finally, it adds wxGridDirectionOperations::TryToAdvance() helper to
avoid repeating the IsAtBoundary() check which was previously part of
PrepareForSelectionExpansion() in multiple places.
And because the "extending" and normal parts of DoMoveCursorByBlock()
are so different now, it also factors out AdvanceByBlock() helper which
can be used to keep these parts well separate from each other instead of
intermixing them together.
With all these preparatory changes, it's finally possible to implement
the "extending selection by block" logic relatively easily, with the
bulk of this branch actually taken by comments explaining why do we have
to do what we do.
Add unit tests verifying that the functions used by Shift-Ctrl-arrow
work as expected.
The selBlock reference could be used after the vector invalidation,
which resulted in using the wrong vector element at best and could
probably lead to a crash at worst.
Don't use it after invalidating the vector any longer.
This notably fixes a bug in row-or-column selection mode where the grid
wasn't updated visually after modifying its selection by deselecting a
single cell as soon as there were more than one selected row or column.
Mouse events in row/column headers should be just ignored when the
current selection mode disallows row/column selection.
This wasn't the case before, and while actually selecting the line
corresponding to the header was disallowed further down the call chain,
clicking it still unexpectedly cleared the existing selection and
dragging mouse in the header window selected the entire grid.
Fix this by just never entering the corresponding cursor mode in the
first place if it's incompatible with the current selection mode.
This function finally doesn't ever create a new block, except for the
trivial case when there is no current block, so rename it to a simpler
and more clear name.
No real changes.
This seems to be more consistent with the existing functions and doesn't
create ambiguity with a grid range.
Also rename wxGridSelectionRange to just wxGridBlocks as, in principle,
this class could be used for iterating over any blocks, not just the
selected ones.
No changes in functionality, this is just a renaming.
We should never adjust the fixed coordinate of the selection blocks in
this mode, so ensure we never end up with a partially selected line in
this case by preserving the -- correct by construction -- current block
coordinates in this direction.
Extend the current block to the entire line when the corresponding
header is Shift-clicked and, importantly, keep the full-line selection
when using Shift-arrows later to make the selection behave in the
expected way.
This made the logic of this function unnecessarily more complicated.
Instead, just fall back to the current cell coordinates in the only
place where this could happen before.
Doing this still preserves the correct behaviour of Shift-arrow
selection when entire rows/columns are selected and the current cell is
not the leftmost/topmost cell (due to scrolling), but the code is
simpler.
Remove the now always true condition check and assert that it's indeed
always true.
Note that the changes to gridsel.cpp in this commit are best viewed
ignoring whitespace changes.
The difference between calling SelectAll() and SelectBlock() with a
block covering the entire grid is that the former discards any
previously selected blocks, which become clearly redundant.
As a consequence, clicking on the grid corner 10 times in a row still
results in a selection with a single block, not 10 (identical) blocks.
These functions always returned the first partially visible line, not
the first fully visible one as intended because IsXXXShown() always
returned true when it applied to that line and we need to increment the
index first, before calling it -- even if this requires writing the loop
in a slightly uglier way.
This fixes bugs in commits 0920a1646b (Make wxGrid row selecting more
user friendly, 2020-03-04) and 89dd47edee (Make wxGrid column selecting
more user friendly, 2020-03-04).
Show the selected cells/rows/columns/blocks instead of just showing
their number, as we now have an efficient (i.e. taking time proportional
to the number of selected blocks and not to the number of the individual
cells, rows or columns) way of doing this.
See #2576.
Don't use long-lived non-const "block" reference, this makes it
difficult to see that it's never changed until the very end of the
function, when it's updated.
Compare "blockStart" with "block" itself instead of comparing it with
"newBlock" which is initially its copy but gets modified later (however
these modifications don't affect the comparisons involving it).
Add a comment explaining why are we doing what we do here.
Expanding the selection from keyboard with Ctrl pressed should move in
the same way Ctrl-cursor does, but use the same selection anchor as
Shift-cursor does instead of always using the current cell.
This makes the expansion work much more intuitively in the grid, e.g.
pressing Shift-Ctrl-Down in
1 2
3 4
grid when 1 and 2 are selected now selects all the cells instead of
selecting 1 and 3 as it did before.
Switch from using just "bool expandSelection" in the grid functions
(possibly) extending the current selection to using the full
wxKeyboardState.
This allows to pass it to ExtendOrCreateCurrentBlock() and slightly
simplify the code by using DoMoveCursorFromKeyboard().
Don't try to extend the existing selected blocks to rows/columns, this
contradicts the documented behaviour which is to discard the selected
blocks that become invalid in the new mode.
Do handle switching to wxGridSelectRowsOrColumns mode, as there doesn't
seem to be any reason not to.
Update the tests to check for the expected selection update behaviour.
Change the return type of this function to a simple and clear bool
instead of 3-valued int requiring a special explanation. This is simpler
and not any less efficient as checking for whether one block contains
another or the other one contains this one are separate operations
anyhow.
Rename the function to a more grammatically correct name.
Also move it inline as it's now trivial enough for this to be worth it.
This class was a strange hybrid of a container/view/range and iterator,
as it both provided begin()/end() container-like methods and
iterator-like methods for dereferencing/advancing.
Simplify this by removing the latter part and making this class really
just a range, with its own iterator class in order to avoid leaking the
exact type of the iterator used in the API.
Note that while it's now completely trivial, it is still useful as it
isolates the application code from the vector used to store the selected
blocks currently and will allow to change internal representation in the
future without breaking the existing code.
Really edit the current selection block instead of storing the temporary
information about the current selection and applying it on releasing Shift
key or LKM.
VZ: [...] this is not used by wxGrid itself and SendEvent() is not part
of the public API, so we never made any promises about being able to call
it with wxEVT_GRID_RANGE_SELECT.
Store all types of selection with an array of blocks instead of arrays of
cells, blocks, rows and columns.
It (hopefully) simplifies the code and allows us to implement editing of
the last selection block much easier.
Line height calculation based on the font size is not right for wxGTK3
because native buttons (used as in-line editor buttons) don't scale down
well and they are not displayed properly within the line.
Minimal button size when the label is displayed properly is 35 so this
has to be minimal line height under wxGTK3.
Closes#18715.
