gEDA Library v3 Specification

The current version (2.x) of the libgeda shared library has a number of defects:

• Poor separation between public interface and internals, which leads to…
• Frequent changes to the API
• API contains large amounts of code specific to application internals

This page is an attempt to itemise in detail what is required of libgeda, and what the interface to it should look like. This will enable:

• More rigorous testing of library functions
• Creation of language bindings for libgeda
• Easier development of other applications for manipulating schematics

1. Define data structures for representing schematics
2. Provide methods for creating and manipulating schematics
3. Provide methods for reading and storing schematics into files and data streams
4. Provide access to detailed data on errors and exceptions
5. Provide a simple interface for configuring libgeda's behaviour

1. Make no assumptions about applications which will use the library
2. Make no assumptions about the compiler or architecture on which the library is being used
3. Fully reentrant for thread safety
4. Minimise number of dependency libraries
5. Detect and gracefully handle multiple instances accessing the same schematic file

Because libgeda should “make no assumptions about applications which will use the library,” libgeda should use only “world” coordinates throughout (the same coordinate system as used in the files). See the file format specification.

Currently, libgeda relies on an embedded Scheme interpreter (Guile) for configuring settings such as library search paths. This has a number of pros and cons:

Pros:

• All applications which access libgeda can automatically use the same settings
• Arbitrary Scheme code can be used in the configuration process

Cons:

• Large & complex dependency
• Doesn't integrate nicely into applications which use e.g. the GNOME or KDE configuration mechanisms
• Substantially complicates the creation of graphical user interfaces for configuring library settings
• Problems with Guile backward-compatibility in the past

It might, therefore, make more sense to provide a complete API for configuring libgeda, and rely on the application to handle the storage and loading of the configuration.

glib provides a very large number of useful features that would make writing and maintaining libgeda easier. These include:

• Portable definitions of basic types (although intptr.h does this too, and would probably be preferable)
• Doubly- and singly-linked lists, and many other data structures
• Plugin loading, memory allocation, threading, IO abstraction, …

Not having to reinvent the wheel – and letting a much larger project be responsible for optimising and maintaining these features – would make the libgeda code smaller and easier to understand and maintain. libgeda already uses glib to a certain extent, but using it more extensively in future versions of libgeda does not immediately seem to be a bad idea.

gobject is an object-oriented programming system in C. It is used extensively in the GTK+ user interface toolkit. In addition to class-like structures, it provides a signalling system that allows objects to emit, listen for and respond to events. It is designed to be compatible with the object systems of other languages, and this is one of the chief reasons that so many bindings of GTK+ are available (Python, Perl, .NET, Java, C++, etc).

gobject is often considered to be heavyweight, requiring lots of boilerplate code in order to use it effectively. Its use in libgeda has historically been opposed, though its use in future versions may ease creation of bindings for libgeda in more directly object-oriented languages such as Python, Java or C++.

GObject is currently used in libgeda for the GedaList class which wrapps a GList, and provides a notification signal when API is used to modify the list.

• Writing Reentrant and Thread-Safe Code