Speaker
Description
We report here our experience of using AI to make change/improvement to open source software like XRootD. We choose three real world tasks to see if AI can have a comprehensive understanding of both XRootD software stack, other software used by Xrootd (e.g. Fuse, Scitoken), and our intention. Theses tasks are: creating a separate version of xrdadler32 that does not depend on any XRootD libraries; upgrading XrdFfs (xrootdfs) from using Fuse 2.6 to Fuse 3.x with limited functionality changes; and asking AI to help us understand how Sci-tokens are passed to the XRootD clients. In each of these tasks, AI worked incredibly well as a coding assistants - it was quick to figure out underline relations within the XRootD stack, and make code change with minimum bugs. But AI does make mistakes that require human corrections. We will point out and analyze those mistakes.