Project Update – 2024-08-10 (Inluding news about the future release of version 2.1)

[pwgit-create]

Project Update – 2024-08-10

Here is an update on the current plans for this project, which includes information on upcoming features and release dates. There is also a possibility for readers to request features that they wish to see in the upcoming v.2.1 release.

Preliminary date of Release for version 2.1

November 2024

The features included in version 2.1

Change for Continue an application

As of now, you can continue an application, and the new application will be available for you as well as the old version. It works fine, but there is a problem when the application has iterated a few times and the code that it contains is more substantial in size (Especially notable when you generate games).

What happens then is that the AI model often will provide you the correct code for the change that you want to see, but it references existing code with comments which means that you yourself need to cut and paste the existing code into the result. That is not optimal and takes away the purpose of Appwish Ollama as being a help to those who has never coded before. Since the new app and continue an app function is devoted (primarily) to non-coders (as opposed to code base generation) it’s not a problem that should be persistent in an application like this.

To solve this problem, the questions to the AI-Models will be modified (in the continue an app algorithm) to ask the model to provide reference numbers to the line start and line end of the existing code (in its comment) so that the code it refers to could be inserted automatically to the output file.

The fix is not difficult to implement, but it requires a lot of testing before it can reach production quality. The preliminary release date for version 2.1 is in December because of this. Code-Base generation could be enhanced with the same principle. There are no issues with the CodeStral model for code base generation in the current version. However, it is a drawback in other models that I have evaluated, and the change will enhance the creation of code bases for those models.

You can of course reduce the risk of having existing code comments in the output from the AI-Model by experimenting with the num_ctx factor, but it´s kind of advanced and requires good knowledge of the model that you are using along with good hardware. For Example, I have got very good result by using CodeStral:22b Q8 with a num_ctx value of 10240.

Improving the Release Script

As of now, the official install script for the framework and service to the amazing Ollama is run in the background in the install script for AppWish. This is due to solve the problem with systems that already have Ollama installed and running and thus have the port to the local Ollama server blocked. That will make the AppWish install script fail and you must run it again. To solve that I decided to install / Update the Ollama service in the background. However, I´ve noticed that by doing so, it could sometimes install an additional Ollama Service that is sort of temporary but still takes up disk space and unnecessary downloads a model again even though you did no need to.

To solve that, I will make the script run in the foreground again, and if the Ollama Service is running (with the 11434 port to the local Ollama server in use) I will kill that process. By doing so, there will never be a problem with updating the Ollama Service by using the AppWish install script. If you have trouble installing AppWish Ollama on the current version (2.0.5) just run the script again and it will work.

It's important to note that without running the AppWish Install script, AppWish Ollama won't work. If you fail to run the AppWish install script for every new release, you may encounter issues with newer models as the Ollama Service continually updates and improves its capabilities. You have the option to download and update Ollama yourself at ollama.com, if you prefer. I use the same script in the AppWish install script. It's important to remember that updating the paths to your AppWish folder is not possible if you don't use the AppWish install script. That means that updating AppWish Ollama from v2.04 to v.205 while having those releases at different locations in your file system requires a change in the aliases. The aliases in the bashrc file will only point to v.205 if you run the AppWish install script, but if not, they will point to v.204 .

External libraries

A problem may arise if you create an application that requires the Selenium library or JSON features (such as the police event apps).

To complete a compilation, you must have those jars (available at the mvn repository) in your classpath.

In the current version of Appwish, you don't have that possibility and must have knowledge of Java to add external libraries to your classpath.
Adding a folder called External Libs to the appwish release folder will resolve that issue. When this function is implemented, you can add the jars that your application requires to this folder. You won't have to do anything to add them to your class path.

The addition of all the jars in that folder to the class path created an opportunity for exploiting it. If a hacker gains control of your computer, place malicious jars there. Although the risk is low, I will create a configuration parameter and set it to External_libs=off to ensure safety.

Your suggestions

Do you have a feature that you would like to see in AppWish, please write a post below , and I will do my best to implement it 😊

2playerAiGame.mp4

[pwgit-create]

External libraries ✅

This is the procedure for using the external libraries function in the upcoming version 2.1.
Since the development of this function went quickly, I adjusted the release date for the next release to November.
Once version 2.1 is released, all you have to do is download the jar files from the mvn repository and place them in the external library folder.
Please note that the default value for using external libraries is set to off for security reasons (avoid supply-chains attack). To enable it, use the alias ollama-config and set the below property to on.

USE_EXTERNAL_LIBRARIES=on

It's worth noting that despite AI models being instructed to solely use libraries from Java 19, they may experience failure in certain instances. If you have app requests that are very specific, like JSON or web page automation apps, that could occur. It is possible that AI models that are open to the public can only achieve that with JSON or Selenium libraries.
If an import doesn't begin with 'java' (such as import com.google.gson.Gson), it won't compile unless external libraries are used. In those cases, adding that library to the external lib folder will work. However, you will have to wait to November until it’s released unless you want to play around with the code in the branch yourself 😉

Here are the external libraries that pwgit-create uses:
https://mvnrepository.com/artifact/org.json/json/20240303
https://mvnrepository.com/artifact/org.seleniumhq.selenium/selenium-java
https://mvnrepository.com/artifact/com.google.gson/gson