Google Git
Sign in
fuchsia/fuchsia/main/./tools/symbolizer
tree: 19e96d653569f68090be9a250e49123599942f5f
  1. test_cases/
  2. analytics.cc
  3. analytics.h
  4. analytics_test.cc
  5. BUILD.gn
  6. command_line_options.cc
  7. command_line_options.h
  8. command_line_options_test.cc
  9. e2e_test.cc
  10. error.h
  11. log_parser.cc
  12. log_parser.h
  13. log_parser_test.cc
  14. main.cc
  15. OWNERS
  16. README.md
  17. sanitizer_symbolizer_markup_test.cpp
  18. symbolizer.h
  19. symbolizer_impl.cc
  20. symbolizer_impl.h
  21. symbolizer_impl_test.cc
tools/symbolizer/README.md

Symbolizer

This is a C++ implementation of a Fuchsia log symbolizer. A symbolizer takes text logs as input, containing unsymbolicated stack traces indicated via a special Symbolizer markup format, and symbolizes them provided debug symbols.

E2E tests

E2E tests are disabled by default because they depend on the presence of //prebuilt/test_data/symbolizer/symbols, which is not downloaded by default. This could be done by jiri init -fetch-optional=symbolizer-test-data && jiri fetch-packages.

Once there are symbols, the E2E tests could be built by adding //tools/symbolizer:e2e_tests to args.gn and executed by fx test symbolizer_e2e_tests.

Log Parser Architecture

LogParser::ProcessNextLine processes incoming log streams using a lazy, zero-allocation pull parser architecture:

  • Lexing (LogTokenIterator): Scans input lines into zero-allocation std::string_view tokens (kText and kMarkup) on demand via .Next(). When an opening {{{ delimiter is followed by another {{{ prior to a closing }}}, the earlier {{{ fragment is classified as kText. This recovers cleanly from stray delimiters (for example, random {{{ text {{{bt:0:0x1000}}}).
  • Evaluation & Buffering: Consumes tokens from LogTokenIterator in a single streaming pass without allocating intermediate token vectors. It maintains unprinted preceding text and dropped non-printing tag text (pending_prefix) before invoking active markup callbacks. Surrounding syslog headers (for example, [klog] INFO: ) are discarded for lines that contain only valid non-printing metadata tags (module or reset) and no active output tags.

Output and Tag Handling Strategy

The parser balances cleaning up noisy metadata logs with preserving raw output and surrounding log context:

  • Valid Printing Tags (e.g., {{{bt:...}}}): The symbolizer transforms these tags into human-readable backtrace frames. Preceding text is prepended to the first emitted frame, and trailing suffix text is appended to the last frame. Output is buffered in a FIFO queue to maintain original stream ordering.

  • Valid Silent Tags (e.g., {{{module:...}}}, {{{mmap:...}}}, {{{reset}}}): These tags configure symbolizer state without producing output lines.

    • When a line contains only valid silent tags, both the tag text and any surrounding syslog prefixes (such as [klog] INFO: ) are suppressed, producing zero output.
    • When a line mixes silent tags with printing or invalid tags, the silent tag text is omitted, but the surrounding prefix text is carried forward and emitted alongside the other tags.

  • Invalid or Unrecognized Markup: If a markup tag is syntactically malformed or unrecognized, the raw tag text ({{{...}}}) is preserved verbatim. The parser transitions the line disposition to emit surrounding text, ensuring unparsed log content is printed rather than silently lost.