tidy

docs/docs/wizardmerge.md

@@ -0,0 +1,78 @@
+# WizardMerge: Save Us From Merging Without Any Clues
+
+**Authors:** Qingyu Zhang, Junzhe Li, Jiayi Lin, Jie Ding, Lanteng Lin, Chenxiong Qian
+
+> This markdown version condenses the contents of `wizardmerge.pdf` into a clean, accessible summary. It preserves the
+> paper's structure, key ideas, and reported results while omitting PDF-specific artifacts that appeared in the original
+> automated extraction.
+
+## Abstract
+
+Modern software development relies on efficient, version-oriented collaboration, yet Git's textual three-way merge can
+produce unsatisfactory results that leave developers with little guidance on how to resolve conflicts or detect incorrectly
+applied, conflict-free changes. WizardMerge is an auxiliary tool that augments Git's merge output by retrieving code-block
+dependencies at both the text and LLVM-IR levels and surfacing developer-facing suggestions. In evaluations across 227
+conflicts drawn from five large-scale projects, WizardMerge reduced conflict-handling time by 23.85% and provided
+suggestions for over 70% of code blocks potentially affected by the conflicts, including conflict-unrelated blocks that
+Git mistakenly applied.
+
+## 1. Introduction
+
+Git's default line-oriented three-way merge is fast and general, but it ignores syntax and semantics. Developers therefore
+frequently encounter merge conflicts or incorrect, conflict-free merges that still alter behavior. Prior structured and
+semi-structured merge tools reframe the problem around AST manipulation but still leave developers without guidance when
+conflicts arise. Machine-learning approaches can suggest resolutions but depend on specialized training data, introduce
+length constraints, and may not match developer intent. WizardMerge addresses these gaps by guiding developers toward
+conflict resolution rather than automatically rewriting code, highlighting both conflicting and potentially affected
+non-conflicting regions.
+
+## 2. Background: Git Merging
+
+Git identifies a merge base, aligns modified code blocks from each side, and treats each modified segment as a Differing
+Code Block (DCB). Conflicts occur when both sides touch overlapping regions; non-conflicting DCBs are applied directly but
+may still change behavior in subtle ways. Developers therefore need insight into how DCBs depend on one another and which
+blocks merit closer inspection during reconciliation.
+
+## 3. Design of WizardMerge
+
+WizardMerge combines Git's merge output with LLVM-based static analysis to illuminate dependencies among code blocks. The
+high-level workflow is:
+
+1. **Metadata collection:** Compile each merge input with LLVM to gather intermediate representation (IR) and debug
+   information without adding custom build steps for large projects.
+2. **Dependency graph generation:** Build overall dependency graphs from LLVM IR, aligning Git's DCBs with graph nodes to
+   capture relationships across both text and IR levels.
+3. **Group-wise analysis:** Partition DCBs into relevance groups so that developers can triage related changes together
+   rather than in isolation.
+4. **Priority-oriented classification:** Score and order DCBs based on dependency violations or potential risk, helping
+   developers focus on code most likely to be affected by the merge.
+5. **Resolution suggestions:** Surface actionable hints for resolving conflicts and flag conflict-unrelated blocks that Git
+   applied but still require human attention.
+
+## 4. Evaluation
+
+WizardMerge was evaluated on 227 conflicts from five large-scale projects. Key findings include:
+
+- **Efficiency:** Average conflict-handling time decreased by 23.85% compared to baseline Git workflows.
+- **Coverage:** WizardMerge produced suggestions for more than 70% of code blocks potentially impacted by conflicts.
+- **False-safety detection:** The tool identified conflict-unrelated blocks that Git applied automatically but that still
+  demanded manual review.
+- **Comparison to ML approaches:** Machine-learning-based merge generators struggle with large codebases due to sequence
+  length limits and generalization challenges; WizardMerge avoids these constraints by relying on static analysis rather
+  than learned models.
+
+## 5. Limitations and Threats to Validity
+
+- WizardMerge depends on successful LLVM compilation of both merge inputs; projects that cannot be built or require
+  non-standard toolchains may limit applicability.
+- Static analysis provides conservative approximations and may miss dynamic dependencies, so developer judgment remains
+  essential.
+- The evaluation focuses on a curated set of conflicts; broader studies could further validate effectiveness across diverse
+  languages and project types.
+
+## 6. Conclusion
+
+WizardMerge augments Git's textual merging by revealing dependency-aware relationships among differing code blocks and
+prioritizing developer effort. By coupling Git merge results with LLVM-based analysis, it shortens conflict resolution time
+and highlights risky, conflict-unrelated changes that would otherwise slip through. Future work includes expanding language
+coverage, refining prioritization heuristics, and integrating the tool more deeply into developer workflows.

docs/miscimages/clip_colorschememapping.xml

@@ -0,0 +1,2 @@
+<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
+<a:clrMap xmlns:a="http://schemas.openxmlformats.org/drawingml/2006/main" bg1="lt1" tx1="dk1" bg2="lt2" tx2="dk2" accent1="accent1" accent2="accent2" accent3="accent3" accent4="accent4" accent5="accent5" accent6="accent6" hlink="hlink" folHlink="folHlink"/>