Exploring Java OSS with LLMs - A New Way to Approach Open-Source Code Reading

Transcript

1. 0 Exploring Java OSS with LLMs: A New Way to Approach Open-Source Code Reading April 23rd, 2025 Hiroshi Ito Executive Officer / CTO Loglass
2. 1 Exploring Java OSS with LLMs: Self Introduction After joining Goldman Sachs' Technology Division as a new graduate, he involved in the development of the company’s core systems. Later, as a VP/Senior Engineer, he contributed to platform development and spearheaded the company’s open-source software (OSS) strategy. After working at two startups, he joined READYFOR as an Executive Officer and Vice President of Engineering (VPoE), where he managed the engineering organization, led the overhaul of the payment infrastructure, and spearheaded new product development. In October 2022, he joined Loglass Inc. as an engineer in the Development Department. After serving as an Engineering Manager (EM) and VPoE, he was appointed Executive Officer and CTO in November 2024. Loglass Inc. Executive Officer, CTO Hiroshi Ito Hiroshi Ito
3. 2 Exploring Java OSS with LLMs The theme was "Exploring Java OSS with LLMs: A New Way to Approach Open-Source Code Reading." 6 Month Ago, I submitted this proposal
4. 3 Exploring Java OSS with LLMs And Today, Looking Around GIDS sessions.... AI! AI! AI! It's the topic everywhere! All Screenshots are from https://developersummit.com/session/ © 2025-2026 Saltmarch. All rights reserved.
5. 4 Exploring Java OSS with LLMs Which Leads to a Big Question... Will We Keep Reading & Writing Code? Image Credit: ChatGPT/Dall-E, 2025
6. 5 Exploring Java OSS with LLMs • It's a traditional Japanese art form – recreating natural landscapes in a pot. • Characteristics: ◦ Cultivated over many years. ◦ Shaped through careful pruning and tending. ◦ Reflects the creator's aesthetic sense and philosophy. A Bit Sudden, But Do You Know "Bonsai"? In Japan, it's often enjoyed as a hobby in later life.
7. 6 Exploring Java OSS with LLMs • If AI handles most coding tasks in the future... • Would reading & writing code might be.. ◦ A pursuit of deep understanding, cultivated over time? ◦ An act of careful crafting and refinement, like pruning? ◦ A way to express coding aesthetics or philosophy? Or perhaps, simply a refined 'hobby'? Is Code Reading & Writing the Future "Bonsai"? Image Credit: ChatGPT/Dall-E, 2025
8. 7 Exploring Java OSS with LLMs In this session, however, acknowledging it's still too early for the "Bonsai" stage (?), we'll explore how to practically leverage LLMs and AI Agents as useful tools for understanding OSS code today. Well, it might be true, but let’s forget about it Image Credit: ChatGPT/Dall-E, 2025
9. 8 Exploring Java OSS with LLMs Why would you read OSS source code anyways? Image Credit: ChatGPT/Dall-E, 2025 • See real‑world architecture up close – Explore how large projects organise layers and boundaries. • Spot design patterns in action – Notice Strategy, Builder, Reactor and other patterns working together. • Learn how good APIs evolve – Observe how maintainers add features while keeping backward compatibility. • Build your code‑reading stamina – Practise tracing big codebases and finding what matters. • Speed up onboarding for new teammates – Use proven, battle‑tested code as hands‑on learning material.
10. 9 Exploring Java OSS with LLMs • Learn how LLMs can accelerate understanding of complex OSS codebases • Discover practical techniques for using AI to navigate OSS projects • Develop an efficient workflow combining traditional and AI-assisted code reading • Apply these techniques to popular Java frameworks and libraries Session Objectives
11. 10 Java OSS Code Reading - The Challenge & The Promise -
12. 11 Exploring Java OSS with LLMs Want to understand the architecture or contribute to that cool Java OSS project! Do You Ever Felt Like This? Image Credit: ChatGPT/Dall-E, 2025
13. 12 Exploring Java OSS with LLMs Want to understand the architecture or contribute to that cool Java OSS project! ...But where to even start? A vast sea of code! Overwhelming! Do You Ever Felt Like This? Image Credit: ChatGPT/Dall-E, 2025
14. 13 Exploring Java OSS with LLMs • Read the manual • Source code deep dive • Find usages: Search for keywords or method calls • Use debugger yourself to go through the code path Result: Time-consuming, prone to frustration, hard to even take the first step. The Old Ways of Tackling the "Code Mountain" Image Credit: ChatGPT/Dall-E, 2025
15. 14 Exploring Java OSS with LLMs Spring Framework: Untangling the bean lifecycle from creation to destruction – how does Spring manage dependencies across hundreds of classes while maintaining circular reference detection? Hibernate ORM: Deciphering lazy loading proxies and collection wrappers – where in the codebase does bytecode enhancement happen and how does it intercept property access? Netty: Tracing asynchronous operations through event loops – following a request from channel registration to completion callback across multiple thread contexts. Apache Kafka: Tracking message flow from producer to consumer – how does offset management work across multiple partitions and consumer group instances? Understanding OSS Internal Code Can Be Hard
16. 15 Exploring Java OSS with LLMs Where should I start reading? How can I grasp the overall picture? Which are the main classes or modules? Where is the code related to this specific feature? Facing vast code, what I need first is a "map"! This "first step" hurdle is extremely high. The Biggest Hurdle: The "First Step"
17. 16 Exploring Java OSS with LLMs A Savior Appears? Enter AI Assistants Image Credit: ChatGPT/Dall-E, 2025
18. 17 Exploring Java OSS with LLMs Summarization: Tells you the purpose/role of the whole project, modules, or classes. Structure Explanation: Explains the rough architecture and relationships between key components. Identifying Key Parts: "Where is feature xxx located?" → Suggests relevant code locations. Illustration: Visualize the key concept with image or modeling notation e.g. mermaid Deep dive with repeated Q&A: Answers "What is the purpose of xx ?", “Why this code is written this way?” etc. How AI Helps the "First Step"
19. 18 Source Code Reading with LLMs - Spring Boot-
20. 19 Exploring Java OSS with LLMs Hello ChatGPT, I’d like to dive into the Spring Boot source code to grasp its high‑level architecture and then perform an in‑depth analysis of some of its core features. Could you please start by providing an overview of Spring Boot and outline an effective step‑by‑step approach for reading its code? Let’s get started!! Image Credit: ChatGPT/Dall-E, 2025
21. 20 Exploring Java OSS with LLMs ChatGPT answers
22. 21 Exploring Java OSS with LLMs Opinionated Spring Spring Boot builds on the core Spring Framework by applying “convention over configuration.” It auto‑configures your application based on the jars on your classpath and sensible defaults, so you can get a production‑ready app up and running with minimal XML or Java configuration. Key Goals 1. Quick Start: Embed a servlet container (Tomcat, Jetty) so you can java -jar your app. 2. Minimal Boilerplate: Provide starter POMs (spring-boot-starter-*) that bring in a curated set of dependencies. 3. Production‑Ready Features: Include health checks, metrics, externalized configuration, and more out of the box. 1. What Is Spring Boot and Why It Exists
23. 22 Exploring Java OSS with LLMs SpringApplication • Entry point (SpringApplication.run(...)): boots the entire framework, sets up the ApplicationContext, triggers auto‑configuration, and then calls any CommandLineRunner or ApplicationRunner beans. Auto‑Configuration • Lives in the spring-boot-autoconfigure module. • Scans for classes annotated with @Configuration and guarded by @ConditionalOn… annotations (e.g. @ConditionalOnClass, @ConditionalOnMissingBean). • Decouples “if X is on classpath, then configure Y” logic into many small auto‑config classes. 2. Core Architectural Building Blocks
24. 23 Exploring Java OSS with LLMs Starter POMs • Modules named spring-boot-starter-* (web, data-jpa, security, etc.). • Bring in the right spring‑boot dependencies without you having to pick versions. Actuator • In spring-boot-actuator. • Exposes operational endpoints (health, metrics, env, loggers) over HTTP or JMX to monitor and manage your app. Developer Tools • spring-boot-devtools for hot reloading, automatic restart, and remote debugging support. CLI • spring-boot-cli lets you write Groovy-based “scripts” that run on Spring Boot. 2. Core Architectural Building Blocks
25. 24 Exploring Java OSS with LLMs 3. High‑Level Module Structure spring-boot/ ├─ spring-boot-project/ │ ├─ spring-boot/ │ ├─ spring-boot-autoconfigure/ │ ├─ spring-boot-actuator/ │ ├─ spring-boot-test/ │ └─ spring-boot-cli/ ├─ docs/ ├─ build-system/ (Gradle scripts) └─ samples/ Tip: Start with spring-boot-autoconfigure to see how each piece (e.g. DataSource, Jackson, Security) wires itself up when relevant classes are present.
26. 25 Exploring Java OSS with LLMs OK, so let’s start with what ChatGPT suggests :) spring-boot-autoconfigure
27. 26 Exploring Java OSS with LLMs Starting the Journey
28. 27 Exploring Java OSS with LLMs The Big Picture: Spring Boot Auto-Configuration
29. 28 Exploring Java OSS with LLMs The Big Picture: Spring Boot Auto-Configuration
30. 29 Exploring Java OSS with LLMs The Big Picture: Spring Boot Auto-Configuration
31. 30 Exploring Java OSS with LLMs The Big Picture: Spring Boot Auto-Configuration
32. 31 Exploring Java OSS with LLMs The Big Picture: Spring Boot Auto-Configuration
33. 32 Exploring Java OSS with LLMs The Big Picture: Spring Boot Auto-Configuration
34. 33 Exploring Java OSS with LLMs The Big Picture: Spring Boot Auto-Configuration
35. 34 Exploring Java OSS with LLMs The Big Picture: Spring Boot Auto-Configuration
36. 35 Exploring Java OSS with LLMs Going Deeper into Source Code
37. 36 Exploring Java OSS with LLMs Going Deeper into Source Code
38. 37 Exploring Java OSS with LLMs Going Deeper into Source Code
39. 38 Exploring Java OSS with LLMs Going Deeper into Source Code
40. 39 Exploring Java OSS with LLMs Going Deeper into Source Code
41. 40 Exploring Java OSS with LLMs Going Deeper into Source Code
42. 41 Exploring Java OSS with LLMs Going Deeper into Source Code
43. 42 Exploring Java OSS with LLMs Going Deeper into Source Code
44. 43 Exploring Java OSS with LLMs Going Deeper into Source Code
45. 44 Exploring Java OSS with LLMs Going Deeper into Source Code
46. 45 Exploring Java OSS with LLMs At this point, I still didn't have a clear mental model of how the different parts of the source code connected. So I asked ChatGPT to create a sequence diagram to help visualize the flow.
47. 46 Exploring Java OSS with LLMs Visual Aid - Sequence Diagram
48. 47 Exploring Java OSS with LLMs Visual Aid - Sequence Diagram
49. 48 Exploring Java OSS with LLMs Hands-on with Debugging
50. 49 Exploring Java OSS with LLMs Hands-on with Debugging
51. 50 Exploring Java OSS with LLMs Hands-on with Debugging
52. 51 Exploring Java OSS with LLMs Hands-on with Debugging
53. 52 Exploring Java OSS with LLMs Hands-on with Debugging
54. 53 Exploring Java OSS with LLMs Another Visual Aid - Debug Flow Illustrated
55. 54 Exploring Java OSS with LLMs But wait, I still get lost
56. 55 Exploring Java OSS with LLMs Summarizing Participant Role
57. 56 Exploring Java OSS with LLMs Summarizing Participant Role
58. 57 Exploring Java OSS with LLMs Summarizing Participant Role
59. 58 Exploring Java OSS with LLMs Summarizing Participant Role
60. 59 Exploring Java OSS with LLMs Summarizing Participant Role
61. 60 Exploring Java OSS with LLMs Summarizing Participant Role
62. 61 Exploring Java OSS with LLMs Summarizing Participant Role
63. 62 Exploring Java OSS with LLMs Summarizing Participant Role
64. 63 Exploring Java OSS with LLMs Summarizing Participant Role
65. 64 Exploring Java OSS with LLMs Summarizing Participant Role
66. 65 Exploring Java OSS with LLMs Summarizing Participant Role
67. 66 Exploring Java OSS with LLMs o3 is creative enough to illustrate the actors Image Credit: ChatGPT/Dall-E, 2025
68. 67 Exploring Java OSS with LLMs Now I can explore code with debugging following ChatGPT’s instructions
69. 68 Exploring Java OSS with LLMs Going back and forth between code and diagram
70. 69 Exploring Java OSS with LLMs Participant role helps to add contexts in the flow
71. 70 Exploring Java OSS with LLMs At this point, you can gradually build your mental model with clear image of what each actor is responsible for
72. 71 Exploring Java OSS with LLMs So now I asked ChatGPT to create mini-boot
73. 72 Exploring Java OSS with LLMs So now I asked ChatGPT to create mini-boot
74. 73 Exploring Java OSS with LLMs Code Demo
75. 74 Exploring Java OSS with LLMs mini-boot illustrated
76. 75 Exploring Java OSS with LLMs mini-boot vs Spring Boot mapping
77. 76 Exploring Java OSS with LLMs Debugging Help with LLM
78. 77 Exploring Java OSS with LLMs While debugging, sometimes you get lost when the debugger stops many times at the same breakpoint
79. 78 Exploring Java OSS with LLMs ChatGPT helps clarifying what I don’t get
80. 79 Exploring Java OSS with LLMs ChatGPT helps clarifying what I don’t get
81. 80 Exploring Java OSS with LLMs ChatGPT helps clarifying what I don’t get
82. 81 Exploring Java OSS with LLMs Now I got confused why Condition doesn’t stop
83. 82 Exploring Java OSS with LLMs Summarization: Tells you the purpose/role of the whole project, modules, or classes. Structure Explanation: Explains the rough architecture and relationships between key components. Identifying Key Parts: "Where is feature xxx located?" → Suggests relevant code locations. Illustration: Visualize the key concept with image or modeling notation e.g. mermaid Deep dive with repeated Q&A: Answers "What is the purpose of xx ?", “Why this code is written this way?” etc. Quick Recap: Here are the overall steps
84. 83 Source Code Reading with LLMs - Eclipse Collections -
85. 84 Exploring Java OSS with LLMs What is Eclipse Collections?
86. 85 Exploring Java OSS with LLMs I asked ChatGPT to illustrate the data architecture and algorithm of FastList in Eclipse Collections
87. 86 Exploring Java OSS with LLMs This is an interesting image of hulcination
88. 87 Exploring Java OSS with LLMs Another example: comparison between FastList and ArrayList
89. 88 Exploring Java OSS with LLMs Interestingly it is still difficult for ChatGPT o3 to illustrate logical sequence in an image
90. 89 Exploring Java OSS with LLMs But if I asked to create a visualization in a text, it perfectly illustrates the logical sequence!
91. 90 Exploring Java OSS with LLMs But if I asked to create a visualization in a text, it perfectly illustrates the logical sequence!
92. 91 Exploring Java OSS with LLMs But if I asked to create a visualization in a text, it perfectly illustrates the logical sequence!
93. 92 Exploring Java OSS with LLMs But if I asked to create a visualization in a text, it perfectly illustrates the logical sequence!
94. 93 Exploring Java OSS with LLMs Let ChatGPT speak about why FastList is fast
95. 94 Exploring Java OSS with LLMs Let ChatGPT speak about why FastList is fast
96. 95 Exploring Java OSS with LLMs Let ChatGPT speak about why FastList is fast
97. 96 Exploring Java OSS with LLMs Let ChatGPT speak about why FastList is fast
98. 97 Exploring Java OSS with LLMs Let ChatGPT speak about why FastList is fast
99. 98 Exploring Java OSS with LLMs Let ChatGPT speak about why FastList is fast
100. 99 Final Remark
101. 100 Exploring Java OSS with LLMs • LLMs provide that crucial "map" to help us take the first step into complex codebases • They can explain architecture, summarize code, and identify key components • Visual aids like sequence diagrams and role descriptions add another dimension of understanding • Combining AI assistance with traditional debugging creates a powerful learning workflow ... and it’s fun!!! Enjoy OSS code reading!! Conclusion