This is the starting page for most of our technical documentation. Some extra
nuggets can be found in the
gemrb/docs/ subdir, but they’re old and very
Quick link: GUIScript documentation
Getting the code
Download the git version of GemRB, via the Clone or download button, the git command-line tool or your IDE.
However, if you plan to submit a pull request, log in to GitHub, visit the linked GitHub page and first fork the repository before cloning the code. Once you push your changes to your fork, you can easily create a pull request through the web interface.
Setting up a development environment
GemRB tries to not reinvent the wheel, so it uses several libraries as dependencies. Before attempting to build, be sure to install them (with all their header/devel packages if they exist and for the same architecture, eg. x86-64 or ARM)! The up to date list can be found in the INSTALL file.
The file also contains general compilation instructions, but here’s the gist:
- Windows: Follow the MSVC or the MSYS2/MinGW guide. They include instructions for installing the dependencies.
- macOS: There’s an xcode project file included. You can also try generating it with cmake.
- Open source platforms
mkdir build cd build cmake .. -DCMAKE_BUILD_TYPE=Debug make -j3
Note: GemRB can run from the build dir, so there’s no need to install it by
The organisation currently has these repositories:
- gemrb - the main project
- gemrb.github.io — this very website
- iesh - IE python shell
- gemrb-assets - for demo data source files, misc project art and more
The engine sees by far the most action, but if you’re interested in something different, all the projects are open to contribution.
Check the main README file to see our a header with build bot status, some static code analyzers and more.
We have a separate page dedicated to the Engine overview, discussing the overall architecture and file layout.
Read about our support for non-ASCII characters.
Technical start up details
When you run GemRB, here is what happens:
- The core initialises, loads all the plugins and preloads some data.
- The main game loop is started.
- Main.py in the main GUIScript directory is ran. If unset, GameType is autodetected through Autodetect.py files.
- More core functionality and asset loading occurs.
- The main loop starts.
- Start.py’s OnLoad method in the game’s GUIScript directory is ran:
- LoadGame(None) will set up a new game by loading the default GAM file and linked resources.
- Party members can be created with CreatePlayer().
- EnterGame() to start the GameControl and enter the map.
- MessageWindow.py is run.
- Game is up and drawing.
- QuitGame() and Quit() to terminate.
Check out CONTRIBUTING for tips on where to start, what to work on, tools, useful links and more.
Besides those tools, check our “Cheats” page for a handful of useful hotkeys and instructions for using our inbuilt python console.
Open issues with gists or draft pull requests if you want a structured discussion of your changes before they are ready for uptake. Otherwise just submit the PR.
Save game compatibility
… is something we strive for. The originals are often annoyingly specific in how they want the file and fields to look, so it can be a hard problem to address.
The general approach is to load a save from the original, resave
it in GemRB and then try loading it again. If it works, great, if not, check
the difference between the two files. Reduce it step by step (manually or
through code changes) until the save works.
iesh (or paired with
ielister) can be
tremendously helpful in comparing the two binary files.
Another way is using a debugger, described with an actual example from #659:
- You resave the game in GemRB, and load it with the original running in a debugger like x64dbg. Crash course tutorial on what this is about.
- You hit the crash point and start to look around for anything familiar: dictionary_githzerai_hjacknir is displayed somewhere around there, sounds like PST stuff (Githzerai).
- Load the GAM file into a raw editor like HxD. Cool, the string appears -> looks like it’s about variables, IESDP tells you more. Look at the other one, no match. No wait, hjacknir is there but the whole thing has a space in it’s name.
- You fix the GemRB save either manually or by a tool like NearInfinity.
- You start the modified GAM file with PST: Cool, loading is proceeding way more than before, so that’s it for sure.
- You browse the source code of GAMImporter for gut feelings, look for the section of variable writing, find a suspicious function called strnspccpy, look it up and see in the code: It really cuts away what the Black Isle guys deliberately left in the code, by accident I assume.
- If there are still crashes, go to 1.
The minimal dataset
GemRB comes with a minimal dataset, that can be used to check if the core core functionality still works. It just fires up the engine and as soon as it loads the Start script, it exits. That is the expected behaviour. No window is opened.
If you installed GemRBb, try the included fhs.cfg:
gemrb -c /usr/share/gemrb/minimal/fhs.cfg
But, you can just run the sample config from the build directory as-is:
gemrb/gemrb -c ../gemrb/GemRB.cfg.noinstall.sample
This is what our build bots use as a trivial sanity check.
Exploring the games
As far as game resources go, the main complementary documentation is on the IESDP - The Infinity Engine Structures Description Project. It represents the accumulated knowledge of IE internals and in a sense, GemRB could be considered a reference implementation. Sometimes it is also helpful to look at the sources of other tools like NearInfinity.
Our iesh is not polished, but is exceptionally useful when trying to figure out if anything uses fields or bits previously thought unused or not working. Or if you need to do any processing. Basically for anything that is much faster when automated, especially since most IE tools have poor search capabilities. It can be used interactively, but also comes with several scripted examples.
Direct reverse engineering is also possible and is the reason for much of the understanding of the way the originals worked. There are several projects using DLL injection to do minor fixes and extensions (ToBex, EEex, something for iwd2) — they sometimes contain useful info.
While most games preserved a debug console that can be helpful, Planescape: Torment’s was too buggy and was disabled for release. See the linked page for a mod to remedy that.
This is something particularly interesting to porters to platforms with
comparatively weak hardware. Some tricks to lower memory usage, I/O and/or
CPU use: change the settings
GCDebug=1536. If OpenGL(ES) is available, building with that enabled
will of course help avoid software rendering.
Cocoa Application Wrapper
Apple platforms (Mac OS X & iOS) are built with a special wrapper. This wrapper enables the application to interface with OS specific features easily without having to alter the GemRB core. The meat of the wrapper is simply an NSApplication Delegate object that can be dynamically extended using categories loaded from the various GemRB plugins. The application delegate has an Objective-C++ interface to the GemRB core so it is possible to map system specific features to trigger action inside the core or get core status.
Currently the wrapper doesn’t do much, it does allow GemRB to be built as a proper application bundle instead of a command line executable and it overrides application quit (on Mac) to allow GemRB to handle a quit event in its own way (currently GemRB simply quits, but it should have a popup warning about losing progress etc).
The iOS version of the wrapper runs a configuration utility prior to running gemrb (Download sample configuration). It also attaches an accessory view to the keyboard for toggling modifier keys.
Apple specific features that could be implemented with the wrapper:
- Using NSUserDefaults to dynamically set engine options at runtime
- GUI interface for editing the config (iOS prefpane or simple preference interface for Mac)
- a quick launch GUI to prompt for game type on launch.
- setup special environments (ex. Python for iOS)
Git is a tool to track changes in files and especially useful for collaboration. You have a file and someone changes a line; you can go back, a year later, and check who added this line, and if they wrote why, mentioned test cases or any other useful info.
It also enables bisection, a quick search to find with what change a problem started occuring. Something was working 2 weeks ago, but isn’t now? You feed git this info, and it will automatically present you with the least amount of necessary checks (changes to verify) to find the exact commit that broke it. Check out Tower’s explanation for a visualization of the process.
git bisect is the subcommand used to nail down the commit that broke the
thing you’re testing.
To start a bisection:
git bisect start
- Assuming the current revision has the problem, enter
git bisect bad
Now we need to find a state when things were working fine. Since you probably
have to guess, let’s try 50 changes before now with
git checkout HEAD~50.
Rebuild as usual, eg. on linux:
cd build make -j2
If it’s bad, run git checkout again and repeat until you find a good version.
Once you have it:
git bisect good.
Now the main cycle starts. Git will suggest a new commit to test, estimate the number of steps and repeat the bisection until it finds the commit that introduced the problem.
- Rebuild as usual
- It works?
git bisect good
- It doesn’t?
git bisect bad
- You cannot test it because of some other error, for example it doesn’t
build at all?
git bisect skip
A couple of runs later, you will be presented with the “first bad commit”, which you should report (or use yourself).
To end the bisect, enter
git bisect reset. This will bring you back to
where you were before in git history.
Using rr for hard to reproduce bugs
In case you want to use rr to get help tracking down a hard to reproduce bug, make sure to build without OpenGL, since it doesn’t support it.