Safe Haskell	None
Language	Haskell2010

Graphics.UI.FLTK.LowLevel.FLTKHS

Contents

Motivation
Goals
Installation (Linux & Mac)
Installation (Windows 64-bit)
Getting Started
Demos
Fluid Support
Stack Traces
API Guide
Slow Compilation Issues
Cabal REPL Issues
Core Types
Widgets
Machinery for static dispatch
Association of widgets and functions

Synopsis

Motivation

This library aims to make it easy for users to build native apps that work portably across platforms.

The original author is also very interested the renewed interested in user interfaces in the programming community, whether the various kinds of functional reactive programming, meta-object protocol UI's, or something like React.js.

The hope is that a low-cost hassle-free way of getting a UI up and running without having to deal with browser, authentication and compilation issues will make it more fun to play around with these great ideas using Haskell.

Why a native toolkit?

Even in this era of web interfaces it is still useful to be able to make native apps. They are usually faster and have fewer security issues.

Why FLTK?

The original author chose FLTK because it was small enough that one person could bind the whole thing in an initial pass. Larger toolkits like QT, although way slicker, would require many man-years of effort.
FLTK is quite featureful
FLTK is mature and maintained, the project is about 20 years old and author has had good experiences with the community
FLTK comes with a simple but quite useful GUI builder, Fluid which is now able to generate Haskell code. See the `Fluid Support` section for more details.

What about HsQML/WxHaskell/Gtk2Hs?

These are all great projects and produce really nice UI's, but they all fail at least one of criterion listed under the Goals section below.

To the author's knowledge, as of the third quarter of 2015, no other package in the Haskell ecosystem meets all those constraints.

Goals

The goals of this library are to provide a low level API to the FLTK that:

provides full coverage of the toolkit allowing the user to write GUI's in pure Haskell.
feels like it has polymorphic dispatch, meaning a single function dispatches to the right implementation based on the type of widget it is given
is not monolithic, meaning new widgets can be incorporated the user's application without needing to recompile this library
is easy to install. This library has a minimum of dependencies and FLTK itself compiles cleanly on most architectures
allows the user to produce statically linked binaries with no external dependencies
includes a lot of complete working demos so that the user can get up and running faster. (see note below)
comes with GUI builder support to alleviate the tedium of laying out widgets by hand

_Note_: As of version 0.4.0.1, the demos have been moved into their own package fltkhs-demos for reasons outlined in the =Demos= section of this document.

Installation (Linux & Mac)

Make sure you have OpenGL installed
Download and install FLTK 1.3.3.
The recommended way of installing FLTK on Linux is from source. The reason is that some package managers seem to put the headers and libraries in a non-standard locations which will cause the Haskell bindings to throw compilation errors. On OSX and Linux it should build and install smoothly using the standard:

> ./configure --enable-shared
> make
> make install

If you are customizing your configure step it is important to at least enable OpenGL and Cairo like so:

> ./configure --enable-shared --enable-gl --enable-cairo

If you didn't install FLTK from source you will need to ensure that the FLTK headers are in the include path and, along with the standard FLTK library, fltk_gl, and fltk_cairo are also in the library path. You will also need the make, autoconf, and autoheader tools to build the Haskell bindings.

Make sure c2hs is installed with `cabal install c2hs`.
Do
```
> cabal install fltkhs
```

or:

> git clone https://github.com/deech/fltkhs
> cabal install

Note: If this step produces a flood of `undefined reference` errors please ensure that you have the right version of FLTK (1.3.3) installed and that the headers are in the expected locations. It seems as though some package managers put the libraries and headers in non-standard places.

Test that the build completed successfully by invoking incredibly unimpressive
```
> fltkhs-buttons
```

Installation (Windows 64-bit)

_UPDATE_: This installation guide only works with MinGHC. Within the last month (January 2016) MinGHC has been deprecated and is no longer supported. If you have GHC 7.10.x, MinGHC will probably work for you but will not work for GHC 8. We are currently working on instructions for building on Windows with Stack.

Note: Currently this package only works on 64-bit Windows.

This install guide has been tested on Windows 7 64-bit with 8GB of RAM.

The general steps involved are:

Install MinGHC and upgrade cabal-install
Install CMake
Download & compile FLTK 1.3.3
Install this package

Install MinGHC

MinGHC installers are available here. This package has been tested with GHC 7.8.4 64-bit and may work with newer versions.

Once MinGHC is downloaded and installed per the instructions in the link above, install the newest version of cabal-install. This is an important step because the "Setup.hs" that ships with this package uses some of the newer features of the Cabal 1.2.x API.

> cabal update
> cabal install cabal-install

Install CMake

Simply click through the CMake with the default options. Once this process is complete the cmake executable should be available at the CMD prompt:

> cmake --version
cmake version 2.8.11.2

Your version will likely be `cmake version 3.2.x` but so long as it is above `cmake version 2.6.x` the install should proceed smoothly.

Download & compile FLTK-1.3.3

Please make sure to only download version FLTK 1.3.3.

Once this is downloaded, since the tar and gunzip utilities ship with MinGHC, you should be able to unpack it at the command prompt:

> tar -zxvf fltk-1.3.3-source.tar.gz

Then descend into the source directory:

> cd fltkh-1.3.3-source

Make a build directory to hold the generated CMake files:

> mkdir build

Descend into the build directory:

> cd build

And run cmake telling it to work with the make utility that ships with MSYS:

> cmake .. -G "MSYS Makefiles"

Please note the `..` after cmake in the command. This is important because it tells CMake to build using the CMake specification found in the parent directory.

Once these steps have been completed you should be able to compile the source:

> make

Each compile step may spit out a scary looking warning about a pointer to int cast, but this can be safely ignored. It seems to be an issue with the MinGW g++/gcc compiler.

Also, the final demo windowFocus may fail to compile and complain about a symbol being redefined. This is again an issue with MinGW and can be safely ignored.

Now it is necessary to define the FLTK_HOME environment variable so this package knows where to look for header and library files:

> set FLTK_HOME=<path-to-the-FLTK-root-directory>

Install this package

At this point it is highly likely that running `cabal --version` at the command line still invokes the old version even though cabal was updated. This is because the location of the newly installed cabal has not been added to the PATH variable. You can do so with this command:

> set PATH=C:\Users\<your-username>\AppData\Roaming\cabal\bin;%PATH%

Now you can finally install this package:

> > cabal install fltkhs

Please note that compared to the Linux and Mac installs the compilation process takes a very long time. The reason for this is the MinGW gcc/g++ compilers are much slower.

Once this has completed, you can test it by invoking the incredibly unimpressive:

> fltkhs-buttons

Getting Started

It is assumed that by this point the user has successfully installed this library and demos in fltkhs-demos (see above).

Quick Start

The quickest way to get started is to the download the FLTKHS project skeleton. It is a simple `hello-world` program that shows the basics of widget creation and callbacks. The project's Cabal file comes pre-configured with the options necessary to produce a statically-linked executable so a simple 'cabal configure && cabal build' should result in an 'fltkhs-hello-world' executable in the dist directory.

Other demos that ship with this package show more complicated usage of the API.

Since the API is a low-level binding, code using it takes on the imperative style of the underlying toolkit. Fortunately it should look pretty familiar to those who have used object-oriented GUI toolkits before.

Demos

_Update_: As of version 0.4.0.1 all the demos have been moved to their own package fltkhs-demos. This speeds up installation of this package significantly and also allows the examples to be run interactively with `cabal repl`. This was an issue pre-0.4.0.1 because the REPL needed to link against the C library containing the C++ bindings and was having trouble finding it because it was just created by the Cabal build step but not yet installed. This seems like a better solution than hacking hs to make the library visible.

There are currently two sets of FLTKHS demos, the ones that are part of the fltkhs-demos package and ones that show off Fluid support.

Please see the README in the fltkhs-demos package for more information.

Fluid Support

This package also comes with a utility (fltkhs-fluidtohs) that takes a user interface generated using the Fluid GUI builder that ships with FLTK and generates Haskell code.

Now the user can drag and drop widgets into place instead of having to calculate coordinates and sizes by hand. Additionally arbitrary Haskell code can be inserted into Fluid interfaces allowing user to do most of callback wiring directly from Fluid.

The quickest way to get started is to download the Fluid/Haskell project template. The "Setup.hs" that comes with the skeleton is configured to use the 'fltkhs-fluidtohs' utility to automatically convert any Fluid in src directory into a Haskell module of the same name during the preprocess step. This means using Fluid in a FLTKHS project is as simple as creating a Fluid interface and running 'cabal build' or 'cabal install'.

Additionally the fltkhs-fluid-examples comes with a number of demos that show off how Fluid integrates with FLTKS.

Stack Traces

In a traditional callback-heavy API such as FLTKHS null pointers happen which is why FLTKHS supports partial stack traces. All FLTK functions throw an error along with a stack trace when given a null Ref.

For pre-7.10 GHCs stack traces will only be shown if the 'xc' flag is used when compiling FLTKHS.

If compiled with GHC > 7.10 a partial stack trace is transparently available to the user. The recently minted 'CallStack' implicit parameter is used to get a trace of the function that made the offending call along with a file name and line number. So, for example, in the following code:

buttonCb :: Ref Button -> IO ()
buttonCb b' = do
  FL.deleteWidget b'
  l' <- getLabel b'
  ...

main :: IO ()
main = do
 window <- windowNew ...
 begin window
 b' <- buttonNew ...
 setCallback b' buttonCb
 ...

a button is placed inside a window in the main method but the first time it is clicked the callback will delete it and then try the extract the label from the null Ref. The resulting stack trace will look something like:

Ref does not exist. ?loc, called at src/Graphics/UI/FLTK/LowLevel/Fl_Types.chs:395:58 in fltkh_Cx8029B5VOwKjdT0OwMERC:Graphics.UI.FLTK.LowLevel.Fl_Types
  toRefPtr, called at src/Graphics/UI/FLTK/LowLevel/Fl_Types.chs:403:22 in fltkh_Cx8029B5VOwKjdT0OwMERC:Graphics.UI.FLTK.LowLevel.Fl_Types
  withRef, called at src/Graphics/UI/FLTK/LowLevel/Hierarchy.hs:1652:166 in fltkh_Cx8029B5VOwKjdT0OwMERC:Graphics.UI.FLTK.LowLevel.Hierarchy
  getLabel, called at src/Main.hs:11:10 in main:Main

It says that the null pointer was originally detected in the library function toRefPtr function which was called by the library function withRef, which was called by getLabel on line 11 of 'src/Main.hs'. Notice, however, that the trace stops there. It does not tell you getLabel was invoked from buttonCb. For a more detailed trace the CallStack implicit parameter needs to be passed to each function in the chain like:

buttonCb :: (?loc :: CallStack) => Ref Button ...
 ...
main :: IO ()
 ...

API Guide

Guide to the Haddock Docs

Convenient access to the underlying C++ is achieved using typeclasses and type-level programming to emulate classes and subclasses. This approach makes Haddock very unhappy and the generated documentation is frequently unhelpful. For instance, newcomers to this library are urged not to look at Graphics.UI.FLTK.LowLevel.Dispatch or Graphics.UI.FLTK.LowLevel.Hierarchy. The purpose of this guide is to point users in a more useful direction.

The documentation provided with this API is not yet self-contained and is meant to be used in tandem with the C++ documentation. The rest of this document is about how the Haskell functions and datatypes map to the C++ ones.

Widget Construction

Each widget has its own module all of which are listed below under the Widgets heading . Most modules include a function named `<widgetName>New` that returns a reference to that widget. Although the user does not have to deal with raw pointers directly it might help to understand that this reference is a pointer to a void pointer to a C++ object.

For instance, windowNew creates a Ref Window which is a pointer to a C++ object of type Fl_Window, the FLTK class that knows how to draw, display and handle window events.

This value of type Ref Window is then passed along to various functions which, transparent to the user, extract the pointer and pass it to the appropriate Fl_Window instance method.

Widget Methods

The Haskell functions that bind to the instance methods of an FLTK class are listed under the Functions heading in that widget's module. It's worth remembering that these type signatures associated with the functions listed under the Functions heading are not the real ones but are artifically generated because they are much more helpful to users. For instance, the actual type of activate exposes all the type level arithmetic required so it can be used by subclasses of Widget but is pretty unhelpful as a reference compared to the artificial type under Functions heading of Graphics.UI.FLTK.LowLevel.Widget.

Unfortunately to see this more helpful type signature the poor reader has to navigate to the corresponding widget's module, find the Functions header and scroll down to the desired function. Haddock, unfortunately, does not support anchors that link to a named point in the page. The author is very open to ideas on how to make this easier.

Carrying on the previous example from the Widget Creation section, the methods on a Ref Window widget are documented in Graphics.UI.FLTK.LowLevel.Window under Functions. Each function takes the Ref Window reference as its first argument followed by whatever else it needs and delegates it appropriately.

As this is a low-level binding the Haskell functions are kept as close as possible in name and argument list to the underlying C++. This allows users familiar with the FLTK API to use this library with less learning overhead and it lets newcomers to FLTK take advantage of the already extensive C++ documentation.

Functions are named to make it as easy as possible to find the corresponding C++ function, however there are some naming conventions to keep in mind:

setters and getters are prefixed with set and get respectively. In C++ both have the same name; the setter takes an argument while the getter does not. Since Haskell does not support overloading this convention is used.
in many cases C++ uses overloading to provide default values to arguments. Since Haskell does not support overloading, these arguments are Maybe types eg. the hotspot function in Graphics.UI.FLTK.LowLevel.Window. In other cases where the common use case leaves the default arguments unspecified the binding provides two functions, a longer less-convenient-to-type one that takes the default argument and a short one that does not eg. drawBox and drawBoxWithBoxtype, also in Graphics.UI.FLTK.LowLevel.Window.
error codes are Either types
function arguments which are pointers to be filled are not exposed to the API user. For instance, a common C++ idiom is to return a string by taking a pointer to some initialized but empty chunk of memory and filling it up. The corresponding Haskell function just returns a String.

It is hoped that until the documentation becomes more self-sustaining the user can use these heuristics (and the type signatures) along with the official FLTK documentation to "guess" what the binding functions do.

Widget Hierarchy

Every widget module in the API has a Hierarchy heading that shows all its parents.

The design of the API makes all the parent functions transparently available to that widget. This is also the reason why the actual type signatures are so complicated requiring the manual generation of artificial type signatures.

For instance, the Functions section under Graphics.UI.FLTK.LowLevel.Window shows that a Ref Window can be passed to getModal to check if the window is modal, but it can also be passed to children in Graphics.UI.FLTK.LowLevel.Group which counts up the number of widgets inside the Window and getX in Graphics.UI.FLTK.LowLevel.Widget which returns the X coordinate of the Window's top-left hand corner.

The hierarchy corresponds almost exactly to the underlying C++ class hierarchy so, again, the user is able to take advantage of the C++ documentation to use the binding API.

Overriding C++ methods

The binding API provides a very limited form of inheritance by allowing the user to provide Haskell implementations of certain key methods on a small set of classes. Which methods and which classes were determined by original author of this library based on a casual look at FLTK code in the wild and what was required for porting the demoes. If there is a pressing need for more methods or classes, please contact the maintainer.

When providing custom methods the object constructor is no longer `<widgetName>New` but `<widgetName>Custom`, which, in addition the parameters taken by `<widgetName>New` also takes records of Haskell functions which are then passed to the C++ side.

Much like a callback, the Haskell functions are passed as function pointers to the C++ side and called whenever the event loop deems appropriate. Unlike callbacks, they can be set only on object instantiation.

An example of this is Graphics.UI.FLTK.LowLevel.Widget which, since it is a base class for most widgets and doesn't have much functionality of its own, only allows custom construction using widgetCustom. This constructor takes a CustomWidgetFuncs datatype which is a record of functions which tells a Graphics.UI.FLTK.LowLevel.Widget how to handle events and draw, resize and display itself.

Again Graphics.UI.FLTK.LowLevel.Window can be used a motivating example. Its custom constructor windowCustom, in fact, takes two records. A CustomWidgetFuncs which allows the user to override methods in its Graphics.UI.FLTK.LowLevel.Widget parent class and also a CustomWindowFuncs record which allows the user to override flush, a method on the Window class which tells the window how to force a redraw. For example, the demo src/Examples/doublebuffer.hs (which corresponds to the executable 'ftlkhs-doublebuffer') tells both windows how to draw themselves in a Haskell function that uses low-level FLTK drawing routines by overriding the draw function of their Graphics.UI.FLTK.LowLevel.Widget parent.

Every widget that supports customizing also provides a default function record that can be passed to the constructor. For example, Graphics.UI.FLTK.LowLevel.Widget provides defaultCustomWidgetFuncs and Graphics.UI.FLTK.LowLevel.Window has defaultCustomWindowFuncs. In the demo mentioned above the singleWindowCustom function is given defaultCustomWidgetFuncs but with an overridden drawCustom.

Another case where customization comes up a lot is when using Graphics.UI.FLTK.LowLevel.Table which is a low-level table widget that needs to be told, for example, how to draw its cells. The demo src/Examples/table-simple.hs (corresponding to the executable 'fltkhs-table-simple') shows this in action.

Hopefully the demos just mentioned and others included with this library show that even though customizing is limited it is possible to do a lot.

Additionally for most customizable functions a widget provides a corresponding function that calls the default C++ implementation as the function name suffixed with "Super". For instance, the user can provide a custom implementation of handle on Graphics.UI.FLTK.LowLevel.Window via the constructor and every call to handle invokes that implementation, but the user can also call handleSuper to get at the default C++ implementation. This comes in handy when the custom function is just setting up variables or logging and wants underlying implementation to take over at some point.

Warning: since calls to the default implementation of the function, handleSuper, for example, are available to the widget they are automatically available to all subclasses of the widget. The author acknowledges this is a confusing aspect of the library's design but hopes the benefit of being able to get at the default implementation out-weighs the trap the user might fall into. If this becomes a pervasive problem the author is open to removing this functionality.

Slow Compilation Issues

As described above, the API emulates multiple dispatch using type-level programming, closed type families and typeclasses. While this is makes for a nice API it has also slowed down compilation of executables much more than expected.

To clarify the time taken to compile the library itself has not changed, but applications that use the library to create executables are taking a lot longer to compile. To further emphasize, there does not appear to be any runtime performance issues. This is only a compile time problem.

To preserve the user's and the author's sanity a flag fastCompile has been introduced to the fltkhs-demos and to the fltkhs-fluid-examples. This flag which tells the compiler to skip the specialising step when compiling executables, dramatically decreases compile time but also bloats the resulting executable size and probably makes runtime performance much slower. In this package and fltkhs-fluid-examples it is enabled by default, since the executables are just demos that are not meant to show off performance. To disable this flag, tell Cabal to ignore it during the configure step:

cabal configure -f-fastCompile

In the fltkhs and the fltkhs-fluid project skeletons this flag is disabled by default to provide the best runtime performance. To enable the flag for a smoother development workflow, tell Cabal to enable it during the configure step:

cabal configure -f fastCompile

Cabal REPL Issues

The recommended way to running the REPL in a fltkhs application is `cabal repl`. It works in GHC 7.8.x, but unfortunately will not work in GHC 7.10.x to due to a regression