Refactor duplicated code into shared utilities

[devin-ai-integration]

Summary

Five copy-pasted patterns across src/ are collapsed into shared helpers; net −215 lines with no intended behaviour change (parser output verified byte-identical on real geth/Nethermind/Besu/Reth files, see below).

1. Cached raw-GitHub fetching → new src/github_fetcher.py. SpecFetcher and CodeFetcher each carried their own session/token setup, cache-hit/write dance (4 near-identical copies of "check cache file → GET raw URL → raise_for_status → write cache"), and clear_cache/list_cached_*. Both now subclass CachedGitHubFetcher:

class SpecFetcher(CachedGitHubFetcher):
    DEFAULT_CACHE_DIRNAME = ".spec_cache"

    def fetch_eip(self, eip_number, use_cache=True):
        return self.fetch_raw_file("ethereum", "EIPs", f"EIPS/eip-{eip_number}.md", "master",
                                   cache_key=f"eip-{eip_number}.md", use_cache=use_cache)

Cache-file names, directory defaults, the GitHub Accept header on CodeFetcher, and SpecFetcher.list_cached_specs() are all preserved.

2. Brace-block scanning → _brace_delta / _close_block / _find_block_end in parser.py. The Go, C#/Java and Rust parsers repeated the brace_count += line.count('{') - line.count('}') scan eight times, three of them with the same "opening brace may be on one of the next 3 lines" lookahead. Those call sites are now one line each, e.g. end_line = _find_block_end(lines, i, lookahead=3). The Rust fn case keeps its multi-line-signature scan and only reuses _close_block.

3. Analyzer result construction → BaseAnalyzer._result_from_payload / _error_result. Gemini, OpenAI and Azure each built the same AnalysisResult(status=..., confidence=..., issues=..., summary=...) from the parsed JSON plus the same status="ERROR" fallback. get_analyzer also loses its three identical required-key loops in favour of a _PROVIDERS table.

4. Summary aggregation → new src/summary.py. ReportGenerator._generate_summary and differential.summarize_results were the same severity/confidence/status rollup with different extras; the shared function takes count_verification / count_issue_types flags and both callers keep their existing output shape. One deliberate consequence: an empty results list now summarises as UNCERTAIN rather than COMPLIANT in reports (all([]) == True previously made zero analysed files "compliant"); the differential path already behaved this way.

5. CLI. analyze and diff shared a 12-line except block and a 10-line banner/config-table block → _abort_on_error(e, verbose) and _print_info_panel(title, rows). diff also reimplemented differential.analyze_clients (per-client _run_analysis loop, ClientAnalysis assembly, engine build, LLM synthesis); it now calls that function, which gained an on_client_start callback so the CLI keeps printing per-client progress.

Testing

• pytest: 165 passed, 2 skipped (16 new tests in tests/test_shared_utils.py cover the cache, brace helpers, summary flags and analyzer result helpers); ruff check src tests clean.
• Parser equivalence checked against the pre-refactor parser.py on real client sources (state_transition.go, contracts.go, TxValidator.cs, MainnetTransactionProcessor.java, reth validation.rs, lib.rs) — identical CodeBlock output in every case.

Link to Devin session: https://app.devin.ai/sessions/be6024b1467446b8ad89f21886192779
Requested by: @Fosurero

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring