GEGL

Development

This document describes some handy things to know when developing the gegl internals. The GEGL glossary might help make sense of the terms used.

Getting sources, and building

Links and information about various ways of getting a build environment for GEGL.

GEGL inheritance tree

Generated class inheritance graph generated from runtime introspection.

Setting up

Ubuntu 8.10

Setup instructions for Ubuntu 8.10 Intrepid Ibex

To install the mandatory dependencies:

$ sudo apt-get install libtool automake glib libglib2.0-dev libpng12-dev libgtk2.0-dev git

Some of the other dependencies:

$ sudo apt-get install asciidoc enscript libjpeg62 libopenraw graphviz-dev

For running gegl the GEGL_PATH, which is used for dynamically loading the operations, has to be set:

$ export GEGL_PATH=~/Dev/gegl-dev/operations/

BABL

When using a development version of babl, gegl has to know from where to get it. This is done by setting the BABL_LIBS environment variable before (or during as shown below) running gegl’s autogen.sh:

$ ./autogen.sh BABL_LIBS=~/Dev/babl/babl/.libs/libbabl-0.0.so.0.23.0 CFLAGS="-O0"

Then when running the program, the babl library automatically loads extensions that are either located in the BABL_PATH directory or in the default installation directory, in mentioned order of preference.

$ export BABL_PATH=~/Dev/babl/extensions

Netbeans 6.5

There are some key points to consider when setting up GEGL in an IDE (tested on Netbeans):

  • have to configure the IDE to use the autogen.sh as configure script

  • normally have to use gegl/bin/.libs/gegl as the executable, not gegl/bin/gegl which is a script.

  • in some (?) case has to use bin/.libs/lt-gegl as the executable, which is some kind of relinked gegl binary

Debugging

By default gegl and babl uses the flag -g for debug instrumentation, but however it doesn’t use the -O0 flag for turning off optimisations. This leads to unexpected jumps in the code when stepping in a debugger. You have to feed this flag to autogen:

$ ./autogen.sh CFLAGS="-O0"
$ make

Debug output

GEGL has built in mechanisms for logging debug information.

GEGL_NOTE (CACHE, "foo %s", bar);
GEGL_TIMESTAMP(PROCESSOR);
GEGL_MARK()

Where CACHE and PROCESSOR is used the following logging domains are available:

PROCESS, CACHE, BUFFER_LOAD, BUFFER_SAVE, TILE_BACKEND and PROCESSOR

Actual printing of these can be enabled by setting the GEGL_DEBUG environment variable like:

GEGL_DEBUG=processor,cache

or even

GEGL_DEBUG=all

There are also a few functions that are useful as you debug from within a debugger such as GDB. In GDB for example, you call a function interactively in the prompt, while a breakpoint is hit for example, by typing

print function_name(args)

Some useful functions are:

  • gegl_dot_node_to_png_default() Writes a PNG to /tmp/node.png with the dependency graph for the passed node

  • gegl_node_dump_depends_on() Dumps to stdout the nodes that the passed node depends on. With this you can work yourself backwards in a dependency graph.

  • gegl_node_get_debug_name() Prints a debug string representation of a node.

Graphviz export

The gegl library has a utility that permits to export the DAG into a graphviz format. Graphviz is a library that converts graph descriptions in textual format into an image. See graphviz website

It is done using:

#include "../gegl/gegl-dot.h"
/* for printing the dot output, note that gegl_node is a GeglNode pointer */
gchar *dot_output = gegl_to_dot( gegl_node );
printf( "%s\n", dot_output );
g_free( dot_output );

For creating the graph image:

$ gegl --file gaussian-blur.xml --output out.png | dot -Tpng > gb.png

This is the gaussian-blur.xml file:

<?xml version='1.0' encoding='UTF-8'?>
<gegl>
       <node operation='gegl:gaussian-blur'>
               <params>
                       <param name='std-dev-x'>0.999</param>
                       <param name='std-dev-y'>0.999</param>
               </params>
       </node>
       <node operation='gegl:load'>
               <params>
                       <param name='path'>in.png</param>
               </params>
       </node>
</gegl>

You can also just call the function gegl_dot_node_to_png() directly from within gdb to show the graphviz graph of a node and its dependencies.

Tests

There are regression tests in the subfolder tests. These are run with make check

Operation reference renders

For the operation documentation available at The GEGL website GEGL generates a 200x200 PNG image, based on a set of available input images and default parameters - or optionally with a custom representative GEGL graph stored in meta-data in the operations.

GEGL tries to tune its settings to be as deterministic as possible when rendering these with the CPU, making the md5sums of the raster content (not the PNG files) possible to use for regression testing. To force a re-run of these tests remove the operations sub-folder in docs/ and run make from docs/ again.

XML Composition tests

The tests under tests/compositions are high-level system tests for GEGL and its operations. Together with our Jenkins server that runs all our tests each night, the composition tests make a powerful framework for detecting regressions.

Adding an XML composition test

To add a composition test for a operation called gegl:new-operation, do the following:

  1. Create a GEGL XML file tests/compositions/new-operation.xml (will typically look approximately like tests/compositions/pixelise.xml)

  2. Produce a reference image: cd tests/compositions; gegl -o /tmp/new-operation.png new-operation.xml (make sure your operation is installed so gegl finds it)

  3. Manually inspect the reference image /tmp/new-operation.png and move it to tests/compositions/reference if it looks like you expect

  4. Add run-new-operation.xml.sh to the TESTS variable in tests/compositions/Makefile.am

  5. Run make check in tests/compositions to verify that your test works (note that you must have configured GEGL with autogen.sh in order for your change to the TESTS variable to be taken into account)

And you’re done. Do not manually create run-new-operation.xml.sh, it will be created automatically for you during build time. It will run gegl with tests/compositions/new-operation.xml and compare the result with tests/compositions/reference/new-operation.png. If the result differs, the test will fail, and mails will be sent to the GEGL maintainers. As stated above, this test will run each night, so if someone breaks your contributed GEGL operation, it will be discovered at most 24 hours later, making it easy to fix, either by reverting the bogus commit or by adjusting it.

An example of a commit that adds a composition test for a GEGL operation is Add composition test for gegl:gamma.

Documentation

This document describes how to document GEGL using it’s build system.

There are three utilities used:

  1. asciidoc - used for generating html from text files

  2. enscript - used for converting source files (.c/.h) to html

  3. a home-made ruby script - used for generating api documentation (not yet documented here)

All documentation resources are placed in /doc and the generation is controlled by the file Makefile.am

asciidoc

Add .txt files to the docs dir, and type $ make sync-txt # this will cause the text file to be added to the html pages to generate.

enscript

TODO This example will show how a new c/h file is converted into html using enscript

Inheritance tree

Here is an automatically generated inheritance tree of the gobjects used by gegl: GEGL inheritance tree Note that the operations are also gobjects, but they are not included in the inheritance tree.