Commit f07cd320 authored by Daniel Kochmański's avatar Daniel Kochmański
Browse files

sheets as ideal forms

parent afbb11ec
title: Sheets as ideal forms
date: 2018-03-05 17:30
format: md
tags: port, sheet, graft, medium, drawing, graphics, clim, lisp
author: Daniel Kochmański
CLIM operates on various kinds of objects. Some are connected by an inheritance,
other by a composition and some are similar in a different sense.
As programmers, we often deal with the inheritance and the composition,
especially since OOP is a dominating paradigm of programming (no matter if the
central concept is the object or the function). Not so often we deal with the
third type of connection, that is the Form and the phenomena which are merely a
shadow mimicking it[^1].
Let us talk about sheets. The sheet is a region[^2] with an infinite resolution
and potentially infinite extent on which we can draw. Sheets may be arranged
into hierarchies creating a windowing system with a child-parent relation. Sheet
is the Form with no visual appearance. What we observe is an approximation of
the sheet which may be hard to recognize when compared to other approximations.
[^1]: [Theory of forms](
[^2]: And many other things.
## Physical devices
Sheet hierarchies may be manipulated without a physical medium but to make them
visible we need to draw them. CLIM defines `ports` and `grafts` to allow rooting
sheets to a display server. `medium` is defined to draw sheet approximation on
the screen[^3].
How should look a square with side length 10 when drawn on a sheet? Since it
doesn't have a unit we can't tell. Before we decide on a unit we choose we need
to take into consideration at least the following scenarios:
1. If we assume device-specific unit results may drastically differ (density may
vary from say 1200dpi on a printer down to 160dpi on a desktop[^4]!. The very
same square could have 1cm on a paper sheet, 7cm and 10cm on different displays
and 200cm on a terminal. From the perspective of a person who uses the
application, this may be confusing because physical objects don't change size
without a reason.
2. Another possible approach is to assume the physical world distances measured
in millimeters (or inches if you must). Thanks to that object will have a
similar size on different mediums. This is better but still not good. We have to
acknowledge that most computer displays are pixel based. Specifying distances in
millimeters will inevitably lead to worse drawing quality[^5] (compared to the
situation when we use pixels as the base unit). Moreover, conversion from
millimeter values to the pixel will most inevitably lead to work on floats.
3. Some applications may have specific demands. For instance, application is
meant to run on a bus stop showing arrival times. Space of the display is very
limited and we can't afford approximation from the high-density specification
(in pixels or millimeters) to 80x40 screen (5 lines of 8 characters with 5x8
dots each). We need to be very precise and the ability to request a particular
unit size for a sheet is essential. Of course, we want to test such application
on a computer screen.
I will try to answer this question in a moment. First, we have to talk about the
`CLIM` specification and limitations imposed by McCLIM's implementation of
[^3]: See some [general recommendations](
[^4]: Technically it should be [PPI, not DPI]( (pixels per inch).
[^5]: Given programmer specifies sheet size in integers (like 100x100).
## Ports, grafts, and McCLIM limitations
If a port is a physical connection to a display server then graft is its screen
representation. The following picture illustrates how the same physical screen
may be perceived depending on its settings and the way we look at it.
![Graft drawing](
As we can see graft has an orientation (`:default` starts at the top-left corner
like a paper sheet and `:graphics` should start at the bottom left corner like
on the chart). Moreover, it has units. Currently, McCLIM will recognize
`:device`, `:inches`, `:millimeters` and `:screen-sized`[^11].
That said McCLIM doesn't implement graft in a useful way. Everything is measured
in pixels (which `:device` units are assumed to be) and only the `:default`
orientation is implemented. By now we should already know that pixels are a not
a good choice for the default unit. Also, programmer doesn't have means to
request unit for a sheet (this is the API problem).
[^11]: Screen size unit means that the coordinate [1/2, 1/2] is exactly in the
middle of the screens.
## Physical size and pixel size compromise
We will skip the third situation, for now, to decide what unit should we default
to. There are cognitive arguments for units based on a real-world distance but
there are also and practical reasons against using millimeters – poor mapping to
pixels and the fact that CLIM software which is already written is defined with
the assumption that we operate on something comparable to a pixel.
Having all that in mind the default unit should
[device-independent pixel](
One of McCLIM long-term goals is to adhere
to [Material Design]( guidelines – that's why we will use
dip[^6] unit. 100dp has absolute value 15.875mm. On 160dpi display it is 100px,
on 96dpi display it is 60px, on 240dpi display it is 150px etc. This unit gives
us some nice characteristics: we are rather compatible with the existing
applications and we preserving absolute distances across different screens.
[^6]: [Density-independent pixel](
## How to draw a rectangle on the medium
Application medium may be a pixel-based screen, paper sheet or even a text
terminal. When the programmer writes the application he operates on dip units
which have absolute value 0.15875mm. It is the McCLIM responsibility to map
these units onto the device. To be precise each graft needs to hold an extra
transformation which is applied before sending content to the display server.
Now we will go through a few example mappings of two rectangle borders[^7] drawn
on a sheet. The violet rectangle coordinates are `[5,5], [22,35]` and the cyan
rectangle coordinates are `[25,10], [30,15]`.
* MDPI display device units are dip and they match native units of our
choosing. No transformation is required.
![Graft drawing](
* Some old displays have density 72PPI. Not all coordinates map exactly to
pixels - we need to round them[^3]. Notice that border is thicker and that
proportions are a little distorted. On the other hand despite a big change in
resolution size of the object is similar in real-world values.
Windows Presentation Foundation declared 96PPI screen's pixel being the
device-independent pixel because such displays were pretty common on
desktops. Almost all coordinates map perfectly on this screen. Notice the
approximation of the right side of the violet rectangle.
![Lower DPI](
* Fact that the screen has higher density doesn't mean that coordinates mapping
perfectly on a lower density screen will map well to a higher density
one. Take this HDPI screen. Almost all coordinates are floats while on the
MDPI display they had all integer values.
* Higher resolution makes rectangles look better (borderline is thinner and
distortions are less visible to the eye). Here is XXXHDPI:
* Some printers have a really high DPI, here is imaginary 2560 DPI printer.
Funnily enough its accuracy exceeds our screen density so the red border which
is meant to show the "ideal" rectangle is a little off (it's fine if we scale
the image though).
![HQ Printer](
* Until now we've seen some screens with square pixels (or dots). Let's take a
look at something with a really low density - a character terminal. To make
the illustration better we assume an absurd terminal which has 5x8 DP per
character (too small to be seen by a human eye). Notice that the real size of
the rectangles is still similar.
![Character terminal](
It is time to deal with `graphics` orientation (Y-axis grows towards the
top). An imaginary plotter with 80DPI resolution will be used to illustrate two
solutions (the first one is wrong!). Knowing the plotter screen height is
important to know where we start the drawing.
* Graft reverts Y axis and sends the image to the plotter. Do you see what is
wrong with this picture? We have defined both rectangles in default
orientation, so our drawing should look similar disregarding the medium we
print on. We do preserve the real size but we don't preserve the image
orientation – cyan rectangle should be higher on the plot.
![Plotter (bad transformation)](
* Correct transformation involves reverting Y axis and translating objects by
the screen height. See the correct transformation (on 80DPI and on MDPI
![80DPI and MDPI plotters](
[^7]: the Ideal border is composed of lines which are 1-dimensional objects
which doesn't occupy any space. Red border in drawings marks "ideal" object
boundaries. Points are labeled in device units (with possible fractions).
## Sheets written with a special device in mind
There is still an unanswered question - how can we program applications with a
specific device limitations in mind? As we have discussed earlier default sheet
unit should be dip and default sheet orientation is the same as a paper
Writing an application for a terminal is different than writing an application
for a web browser. The number of characters which fit on the screen is limited,
drawing shapes is not practical etc. To ensure that the application is rendered
correctly we need a special kind of sheet which will operate on units being
characters. Take a look at the following example.
![Sheets with different units.](
The first sheet base unit is a character of a certain physical size. We print
some information on it in various colors. Nothing prevents us from drawing a
circle[^9] but that would miss the point. We use a character as a unit because
we don't want rounding and approximation. Background and foreground colors are
The second sheet base unit is dip. It has two circles, solid grey background and
a few strings (one is written in italic).
Ideally, we want to be able to render both sheets on the same physical
screen. The graft and the medium will take care of the sheet approximation. The
effect should look something like this.
![Different kinds of screens.](
The first screen is a terminal. Circles from the second sheet are approximated
with characters and look ugly but the overall effect resembles the original
application. Proportions are preserved (to some degree). We see also the
terminal-specialized sheet which looks exactly as we have defined it.
The second screen is mDPI display. The second sheet looks very much like
something we have defined. What is more interesting though is our first sheet –
it looks exactly the same as on the terminal.
[^8]: Providing means to change defaults requires additional thought and is a
material for a separate chapter. McCLIM doesn't allow it yet.
[^9]: As we have noted sheets are ideal planar spaces where line thickness is 0
and there is nothing preventing us from using fractions as coordinates.
<!-- Writing applications for terminal require a special kind of focus: space is -->
<!-- very limited and off-by-one errors yield a huge difference in visual appearance in -->
<!-- the application. Rounding from dip is a big no for some use cases. On the other -->
<!-- the hand we don't want to provide a separate implementation of each pane and gadget -->
<!-- (which are special kinds of sheets) in case our application declares non-default -->
<!-- units. These two concerns yield the following conclusion: in the application frame -->
<!-- different sheets may have different units and orientation (and effectively different -->
<!-- grafts engrafted to their mediums). -->
<!-- It is important to remember, that this behavior -->
<!-- must be transparent from the perspective of other sheets – we need to find a clever -->
<!-- way to convert the sheet-specific unit to the default unit dip. If a sheet is grafted -->
<!-- then this should be easy enough, but what about ungrafted sheet? What is the size -->
<!-- of a character? Also, spec clearly says, that sheet hierarchy may be grafted only -->
<!-- to one graft. Hmm, so maybe we can't really reuse the components.. -->
<!-- sheet-region is in sheet coordinate system! -->
<!-- -->
<!-- move-and-resize-sheet, sheet-transformation -->
# Port and graft protocols
Now we know *what* we want. Time to think about *how* to achieve it. Let me
remind you what kind of objects we are dealing with:
* Port is a logical connection to a display server. For instance, it may contain
a foreign handler which is passed to the external system API. It is
responsible for the communication – configuring, writing to and reading[^10]
from a device, we are connected to.
* Graft is a logical screen representation on which we draw. It is responsible
for all transformations necessary to achieve the desired effect on the
physical screen. The same port may have many associated grafts for
applications with different units and orientations.
* Medium is a representation of the sheet on a physical device. Sheet is the
Form which is a region and may be drawn – it doesn't concern itself with
physical limitations.
In the next post I will show how to implement the port and the graft (and a bit
of the medium) for the charming backend. I will cover only bare minimum for
mediums important to verify that graft works as expected. Complete medium
implementation is the material for a separate post.
[^10]: Sometimes we deal with devices which we can't take input from – for
instance a printer, a PDF render or a display without other peripherals.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment