UICollectionViewLayout with dynamic heights - but NOT using a flow layout - uicollectionview

Say you have a UICollectionView with a normal custom UICollectionViewLayout.
So that is >>> NOT <<< a flow layout - it's a normal custom layout.
Custom layouts are trivial, in the prepare call you simply walk down the data and lay out each rectangle. So say it's a vertical scrolling collection...
override func prepare() {
cache = []
var y: CGFloat = 0
let k = collectionView?.numberOfItems(inSection: 0) ?? 0
// or indeed, just get that direct from your data
for i in 0 ..< k {
// say you have three cell types ...
let h = ... depending on the cell type, say 100, 200 or 300
let f = CGRect(
origin: CGPoint(x: 0, y: y ),
size: CGSize(width: screen width, height: h)
)
y += thatHeight
y += your gap between cells
cache.append( .. that one)
}
}
In the example the cell height is just fixed for each of the say three cell types - all no problem.
Handling dynamic cell heights if you are using a flow layout is well-explored and indeed relatively simple. (Example, also see many explanations on the www.)
However, what if you want dynamic cell heights with a (NON-flow) completely normal everyday UICollectionViewLayout?
Where's the estimatedItemSize ?
As far as I can tell, there is NO estimatedItemSize concept in UICollectionViewLayout?
So what the heck do you do?
You could naively just - in the code above - simply calculate the final heights of each cell one way or the other (so for example calculating the height of any text blocks, etc). But that seems perfectly inefficient: nothing at all of the collection view, can be drawn, until the entire 100s of cell sizes are calculated. You would not at all be using any of iOS's dynamic heights power and nothing would be just-in-time.
I guess, you could program an entire just-in-time system from scratch. (So, something like .. make the table size actually only 1, calculate manually that height, send it along to the collection view; calculate item 2 height, send that along, and so on.) But that's pretty lame.
Is there any way to achieve dynamic height cells with a custom UICollectionViewLayout - NOT a flow layout?
(Again, of course obviously you could just do it manually, so in the code above calculate all at once all 1000 heights, and you're done, but that would be pretty lame.)
Like I say above the first puzzle is, where the hell is the "estimated size" concept in (normal, non-flow) UICollectionViewLayout?

Just a warning: custom layouts are FAR from trivial, they may deserve a research paper on their own ;)
You can implement size estimation and dynamic sizing in your own layouts. Actually, estimated sizes are nothing special; rather, dynamic sizes are. Because custom layouts give you a total control of everything, however, this involves many steps. You will need to implement three methods in your layout subclass and one method in your cells.
First, you need to implement preferredLayoutAttributesFitting(_:) in your cells (or, more generally, reusable views subclass). Here you can use whatever calculations you want. Chances are that you will use auto layout with your cells: if so, you will need to add all cell's subviews to its contentView, constrain them to the edges and then call systemLayoutSizeFitting(_:withHorizontalFittingPriority:verticalFittingPriority:) within this "preferred attributes" method. For example, if you want your cell to resize vertically, while being constrained horizontally, you would write:
override func preferredLayoutAttributesFitting(_ layoutAttributes: UICollectionViewLayoutAttributes) -> UICollectionViewLayoutAttributes {
// Ensures that cell expands horizontally while adjusting itself vertically.
let preferredSize = systemLayoutSizeFitting(layoutAttributes.size, withHorizontalFittingPriority: .required, verticalFittingPriority: .fittingSizeLevel)
layoutAttributes.size = preferredSize
return layoutAttributes
}
After the cell is asked for its preferred attributes, the shouldInvalidateLayout(forPreferredLayoutAttributes:withOriginalAttributes:) on the layout object will be called. What's important, you can't just simply type return true, since the system will reask the cell indefinitely. This is actually very clever, since many cells may react to each other's changes, so it's the layout who ultimately decides if it's done satisfying the cells' wishes. Usually, for resizing, you would write something like this:
override func shouldInvalidateLayout(forPreferredLayoutAttributes preferredAttributes: UICollectionViewLayoutAttributes, withOriginalAttributes originalAttributes: UICollectionViewLayoutAttributes) -> Bool {
if preferredAttributes.size.height.rounded() != originalAttributes.size.height.rounded() {
return true
}
return false
}
Just after that, invalidationContext(forPreferredLayoutAttributes:withOriginalAttributes:) will be called. You usually would want to customize the context class to store the information specific to your layout. One important, rather unintuitive, caveat though is that you should not call context.invalidateItems(at:) because this will cause the layout to invalidate only those items among the provided index paths that are actually visible. Just skip this method, so the layout will requery the visible rectangle.
However! You need to thoroughly think if you need to set contentOffsetAdjustment and contentSizeAdjustment: if something resizes, your collection view as a whole probably will shrink or expand. If you do not account for those, you will have jump-reloads when scrolling.
Lastly, invalidateLayout(with:) will be called. This is the step that's intended for you to actually adjust your sections/rows heights, move something that's been affected by the resizing cell etc. If you override, you will need to call super.
PS: This is really a hard topic, I just scratched the surface. You can look here how complicated it gets (but this repo is also a very rich learning tool).

Related

GTK+3 render rectangle with selection background color

How is it possible to render a rectangle with the background color of selections in GTK+3. I cannot find any API to do that:
static gboolean draw_callback (GtkWidget *widget, cairo_t *cr, gpointer data)
{
auto state=reinterpret_cast<State*>(data);
auto width = gtk_widget_get_allocated_width (widget);
auto height = gtk_widget_get_allocated_height (widget);
auto context = gtk_widget_get_style_context (widget);
gtk_render_background(context,cr,0,0,width,height);
cairo_rectangle(cr,0,height*(1.0 - state->max),width,height*(state->max - state->min));
cairo_set_source_rgb(cr, 0.05,0.6,0.15); //What color should be used here?
cairo_fill (cr);
cairo_set_source_rgb(cr,0.01,0.3,0.07); //And here
auto mid=height*(1.0 - 0.5*(state->min + state->max));
cairo_move_to(cr,0, mid);
cairo_line_to(cr,width,mid);
cairo_stroke(cr);
return FALSE;
}
Use gtk_render_frame() and gtk_render_background(), and set up the GtkStyleContext you obtain from the GtkWidget instance with the CSS state you want to replicate.
If you want to adhere to the theme, then you cannot draw yourself; and CSS does not have "colors": each CSS state can have multiple layers that include images, gradients, and complex blend modes.
Well, here is my hack:
ColorRGBA get_ambient_color(GtkWidget* widget)
{
auto surface=cairo_image_surface_create(CAIRO_FORMAT_ARGB32,4,4);
auto cr=cairo_create(surface);
while(widge!=NULL)
{
auto context=gtk_widget_get_style_context(widget));
gtk_render_background(context,cr,0,0,1,1);
cairo_surface_flush(surface);
auto content=cairo_image_surface_get_data(surface);
if(content[3]==255)
{
auto ret=ColorRGBA{content[2]/255.0f,content[1]/255.0f,content[0]/255.0f,content[3]/255.0f};
cairo_destroy(cr);
cairo_surface_destroy(surface);
return ret;
}
// Surface is not opaque yet. Continue to parent container.
widget_handle=gtk_widget_get_parent(GTK_WIDGET(widget_handle));
}
cairo_destroy(cr);
cairo_surface_destroy(surface);
return ColorRGBA{1.0f,1.0f,1.0f,1.0f};
}
It seams that I have failed to convince people, why you need the ambient colour, so here are two use-cases:
Determine if we are using a dark/light theme. For some applications, this is sufficient. Querying the state only works if the theme supports dark/light modes. This proves the actual result.
Use as input colour for simulating global illumination. The shading of widgets should be affected by the ambient, hence the name. Another good name would be get_average_background. Themers: please don't use gradients with high contrast.
Case 1: A plot
Now you say that the colour of cursors and function graphs should be themable. That is simply not possible: The user of this plot widget can add as many curves and cursors as he wishes, and the easiest way to differentiate them is to use distinct colours.
What about curve and cursor lightness? If the background is dark, then the curve should be light and vice versa. And what background should be chosen? Ideally, something close the the background of the parent widget, but if the theme is regular, white for light, and black for dark would work. Do you notice that the curves are darker in the second figure?
Case 2: A checkbox that looks like a metallic toggle switch button
With the following technique, I have created a switch that looks exactly as if it were rendered through the Cycles path tracer. This is implemented in Gtk+2, but the algorithm is the same.
The two input images
The code
GtkAllocation alloc;
gtk_widget_get_allocation(widget,&alloc);
auto width=alloc.width;
auto context=CairoContext( gdk_cairo_create(gtk_widget_get_window(widget)) );
auto w_in=cairo_image_surface_get_width(light);
auto h_in=cairo_image_surface_get_height(light);
// Render direct lighting
auto surf_temp=CairoSurface( cairo_image_surface_create(CAIRO_FORMAT_ARGB32,w_in,h_in) );
auto context_temp=CairoContext( cairo_create(surf_temp) );
cairo_set_source_surface(context_temp,light,0,0);
cairo_set_operator(context_temp,CAIRO_OPERATOR_OVER);
cairo_paint(context_temp);
//Render ambient reflections
auto surf_temp_2=CairoSurface( cairo_image_surface_create(CAIRO_FORMAT_ARGB32,w_in,h_in) );
auto context_temp_2=CairoContext( cairo_create(surf_temp_2) );
cairo_set_source_surface(context_temp_2,background,0,0);
cairo_set_operator(context_temp_2,CAIRO_OPERATOR_OVER);
cairo_paint(context_temp_2);
cairo_set_operator(context_temp_2,CAIRO_OPERATOR_MULTIPLY);
//Multiply reflections with the background color
cairo_set_source_rgb(context_temp_2, color_bg.r, color_bg.g, color_bg.b);
cairo_rectangle(context_temp_2, 0, 0, w_in, h_in);
cairo_mask_surface(context_temp_2,surf_temp,0,0);
//Add the results
cairo_set_source_surface(context_temp,surf_temp_2,0,0);
cairo_set_operator(context_temp,CAIRO_OPERATOR_ADD);
cairo_mask_surface(context_temp,surf_temp,0,0);
//Scale and move things into place
auto s=static_cast<double>(width)/static_cast<double>(w_in);
cairo_translate(context,alloc.x,alloc.y);
cairo_scale(context,s,s);
cairo_set_source_surface(context,surf_temp,0,0);
cairo_set_operator(context,CAIRO_OPERATOR_OVER);
cairo_paint(context);
Thoughts
The first example boils down to a light/dark query which is currently missing. Maybe querying colours is not required for this to work, but then there has to be an API controlling the shape and blending mode when rendering the background. For example, to render the ambient reflection, I use multiply rather than over. Also, gtk_render_background appears to be a no-op, since GtkDrawingArea has zero opacity (that's why I needed the loop). To be useful, it must use the background as it appears on screen, not the background of the current widget.

change CAMetalLayer background color

My CAMetalLayer background color is black, even if i'm assigning new color as the backgroundColor property.
Am i missing something? Thanks!
Link to the original project :
https://github.com/audiokit/MetalParticles
This project takes a rather unconventional approach to clearing the drawable's texture each frame: it replaces the textures contents with an array of zeros that is the same size as the texture (width * height * 4). Subsequently, it encodes some compute work that actually draws the particles. This is almost certainly not the most efficient way to achieve this effect, but if you want to make the smallest change that could possibly work (as opposed to experimenting with making the code as efficient as possible), Just fill the blankBitmapRawData array with your desired clear color (near line 82 of ParticleLab.swift).
I have gone through your code and can not see a place where you are setting background Color.
The metal layer is added as a sublayer to it, so you have to set it explicitly.
Add this line at the end of your init method in ParticialLab class and see if it works.
self.backgroundColor = UIColor.redColor().CGColor
I found that self.isOpaque = false was needed on the layer.

looping a sprite vertically objective C sprite builder

Note: for this I am using a program called spritebuilder, which allows me to create a game with less code than would normally be needed. If you know a solution that's just all code, then by all means feel free to share it :)
Also, for this question, I followed a tutorial at this link: Build Your Own Flappy Bird Clone. Just scroll down to the part that says: "Loop the Ground"
So here's my problem. Im currently working on a game, and I created a camera which scrolls vertically long with the character sprite i created, however i need a certain image to loop. When the image leaves the bottom part of the screen I would like it to loop around to the top of the screen, infinitely. For this i created two identical images (in this case its the bark of a tree). One will be on screen, while the other will be offscreen, so as the first image leaves the screen, the second will replace it (seamlessly). I created two objects for the images, and assigned them the name _ground1, and _ground2, and I also created an NSArray in which to store them in. (Please refer to the link above if it is somewhat confusing)
Here is the code that I have:
CCNode *_ground1;
CCNode *_ground2;
NSArray *_grounds;
for (CCNode *ground in _grounds) {
// get the world position of the ground
CGPoint groundWorldPosition = [_physicsNode convertToWorldSpace:ground.position];
// get the screen position of the ground
CGPoint groundScreenPosition = [self convertToNodeSpace:groundWorldPosition];
// if the left corner is one complete width off the screen, move it to the right
if (groundScreenPosition.y <(-1 * ground.contentSize.height)) {
ground.position = ccp(ground.position.x , ground.position.y + 2 * ground.contentSize.height);
}
For some reason when I try this, it doesnt seem to work. what happens is that, the camera will travel vertically as it is meant to do, but the images do not loop. Once the two images leave the bottom of the screen, no new images replace them.
i also done this project as above tutorials. it work fine but you have some mistake to set variable in spritebuilder. in your above code replce code as and try it. you only put less than may be it issue.
if (groundScreenPosition.y <=(-1 * ground.contentSize.height)) {
ground.position = ccp(ground.position.x , ground.position.y + 2 * ground.contentSize.height);
}
You are using CCNode objects as _ground1and _ground2.
CCNode objects usually do not have a contentSize, they will return 0 unless you explicitly set them inSpriteBuilder`.
Make sure that you are using CCSprite objects in SpriteBuilder and in your code.
Also, as a friendly hint you should also consider refactoring (renaming) your sprites with more meaningful names for your use case like _treeBark1 and treeBark2 for example.

Two NSTextFields with interdependent widths in autolayout

I’m trying to put together what seems to be a simple case of two NSTextFields with dynamic width and fixed spacing in between. I cannot figure out an effective way to do so though.
I’m looking to get something like this:
The blue boxes are the NSTextFields. When more text is entered into one, it should grow and thus make the other one shrink, maintaining the lead space, trailing space and the spacing in between the fields. The first one should take the priority if both of the fields have too much text. Each field will also clearly have a maximum and a minimum possible width it can reach.
How would I go around handling this, preferably utilising IB autolayout as much as possible?
It seems to me that all of constraints you mentioned directly translate into interface builder --
First view has width >= something.
First view has width <= something
Same for Second view.
Space between views is fixed.
Second view wants to be as small as possible (have its width at 0) but this has lower lower priority than the previous constraints and lower priority than inner content size constraints.
The code I had to add to my view controller, after applying the constraints as per the ilya’s answer:
In controlTextDidChange (_controlWidthConstraint refers to the fixed width constraint of the input; it’s probably 0 by default for the second input):
// Get the new width that fits
float oldWidth = textControl.frame.size.width;
[input sizeToFit];
float controlWidth = textControl.frame.size.width;
// Don’t let the sizeToFit method modify the frame though
NSRect controlRect = textControl.frame;
controlRect.size.width = oldWidth;
textControl.frame = controlRect;
_controlWidthConstraint.constant = controlWidth;
The key lies in invalidating the intrinsicContentSize for the text field when text is input.
You can check a sample project here, to get you on the right track.

Rotate UITableViewCell inside a rotated UITableView

Rather than implementing a clunky, potentially bug-ridden custom table, I went with the much simpler rotate table option. The problem comes in when I decide that, rather than initializing rotated contents, I want to rotate the cell itself and cut down on the amount of code in complex cells.
The following lines are immediately after cell configuration.
This causes every cell to rotate 90° on load, regardless of orientation:
cell.transform =
(CGAffineTransform)CGAffineTransformRotate(cell.transform, (M_PI / 2.0));
cellRotated = YES;
And this option only rotates the first cell once, but preserves the rotation:
if (!cellRotated) {
cell.transform =
(CGAffineTransform)CGAffineTransformRotate(cell.transform, (M_PI / 2.0));
cellRotated = YES;
}
Can cell orientation be tracked with an existing function (or set thereof)?
I can't find anything about this with Google. There are related questions, but mostly about tables and Portrait/Landscape UI orientations, so the counter-rotation implementation is quite a bit different.
EDIT
If I move cellRotated = YES to viewDidAppear: every cell except one gets rotated. (And then cell reuse makes it so that, in this case, every sixth cell is left alone.)
1-5 is good, 6 is bad, 7-11 is good, 12 is bad, etc (and then it changes in a perfectly logical but entirely unwanted pattern when I hit the end of the table)
Halfway there or a step back, I don't know, but that's what I have now.
If rotation is the only transform you do then why don't you just do this,
cell.transform = CGAffineTransformMakeRotation(M_PI/2.0);
and then,
if ( !CGAffineTransformIsIdentity (cell.transform) ) {
// Rotated.
}