Skip to content

SOURCEGRAPH_SEARCH.md

Latest commit

 

History

History
44 lines (31 loc) · 1.38 KB

SOURCEGRAPH_SEARCH.md

File metadata and controls

44 lines (31 loc) · 1.38 KB
Raw

Sourcegraph Search Behavior

This document describes how query strings are formatted before being sent to Sourcegraph. The goal is to match C/C++ source code literals as reliably as possible.

Why formatting is required

Strings in results.json are loaded by Python with real control characters: JSON \\n becomes an actual newline. If sent raw, Sourcegraph splits the query into multiple terms and the exact string literal is lost.

Query workflow

  1. Trim trailing whitespace

    • rstrip() is applied to reduce accidental mismatches.

  2. Convert control chars to C-style escapes

    • \r -> \\r
    • \n -> \\n
    • \t -> \\t

  3. Quote and escape with json.dumps

    • Produces a safe content:"..." filter with properly escaped quotes and backslashes.

  4. Force keyword mode and limit languages

    • Ensures literal matching and avoids regex interpretation.

  5. CRLF fallback

    • If the cooked string contains \n but not \r, a second query is added with \n replaced by \r\n using OR.

Example query

type:file patterntype:keyword count:5 (lang:c OR lang:c++)
content:"HTTP/1.0 400 Bad Request\\nServer: CS241Serv v0.1\\nContent-Type: text/html\\n\\n"

Implementation location

  • ghidra_string_sniper_ext/data/python/sourcegraph_query.py
    • stringify_for_c_source()
    • to_sourcegraph_content_filter()
    • build_sg_query()