Segfaults parsing Solidity, Rust, and TypeScript files

[chrispyspearbit]

Trailmark 0.4.0 segfaults/SIGBUS while parsing realistic Solidity, Rust, and TypeScript files. This appears related to #61, but it is not limited to Go.

I can reproduce on:

• macOS ARM64 via uvx --from trailmark python
• Linux via python:3.12-slim Docker

Environment from Linux repro:

trailmark 0.4.0
python:3.12-slim
tree-sitter 0.26.0
tree-sitter-language-pack 1.12.1

Minimal public-file repro:

docker run --rm -it python:3.12-slim sh -lc '
  apt-get update >/dev/null &&
  apt-get install -y git >/dev/null &&
  pip install -q trailmark &&
  git clone --depth 1 https://github.com/uniswap/v3-core.git /tmp/v3-core &&
  cd /tmp/v3-core &&
  python -X faulthandler - <<PY
from trailmark.parse import parse_file

f = "contracts/libraries/Oracle.sol"
print("parse", f, flush=True)
g = parse_file(f, language="solidity")
print(len(g.nodes), len(g.edges))
PY
'

Observed:

parse contracts/libraries/Oracle.sol
Fatal Python error: Segmentation fault

Current thread ...:
  File "/usr/local/lib/python3.12/site-packages/trailmark/parsers/_common.py", line 33 in node_text
  File "/usr/local/lib/python3.12/site-packages/trailmark/parsers/solidity/parser.py", line 364 in _extract_docstring
  File "/usr/local/lib/python3.12/site-packages/trailmark/parsers/solidity/parser.py", line 130 in _extract_contract
  File "/usr/local/lib/python3.12/site-packages/trailmark/parsers/solidity/parser.py", line 106 in _visit_top_level
  File "/usr/local/lib/python3.12/site-packages/trailmark/parsers/solidity/parser.py", line 95 in _visit_module
  File "/usr/local/lib/python3.12/site-packages/trailmark/parsers/solidity/parser.py", line 73 in parse_file
  File "/usr/local/lib/python3.12/site-packages/trailmark/parse.py", line 105 in parse_file
Segmentation fault

I also reproduced crashes on these public files:

Solidity:
- https://github.com/ValantisLabs/valantis-stex/blob/27b5c748db73d5b5e7d7aeaee902b70c2b74a829/src/STEXAMM.sol
- https://github.com/chronicleprotocol/scribe/blob/a5e0da3a498cb5883502a8fe05e16096254ec3ab/src/Scribe.sol
- https://github.com/uniswap/v3-core/blob/d0831dc6b8a318df3872b6d68f6de135c9f3ec29/contracts/UniswapV3Pool.sol

TypeScript:
- https://github.com/uniswap/v3-core/blob/d0831dc6b8a318df3872b6d68f6de135c9f3ec29/test/Oracle.spec.ts

Rust:
- https://github.com/base/base/blob/6ee3da6325e7416812065c58358274a086f9f223/actions/harness/src/batcher/actor.rs

Notes:

• Tiny synthetic Solidity/Rust/TypeScript files parse successfully and produce calls edges.
• The full files above crash reliably.
• Running one file per subprocess avoids losing the whole batch, but the crashing files still cannot be extracted.
• On the Solidity repros, faulthandler usually points to node_text or nearby AST child iteration inside the Solidity parser.

Expected:

Trailmark should either parse these files or return a Python-level parse/extraction error. It should not terminate the process via SIGSEGV/SIGBUS.

[chrispyspearbit]

I narrowed this down further. The crash appears to be triggered by the resolved tree-sitter version, not by Trailmark's Python parser code directly.

Tested in python:3.12-slim Docker against trailmark==0.4.0 and the Solidity repro file ValantisLabs__valantis-stex/src/STEXAMM.sol:

• tree-sitter==0.25.2 + tree-sitter-language-pack==1.12.1: passes
• tree-sitter==0.25.2 + tree-sitter-language-pack==1.8.1: passes
• tree-sitter==0.26.0 + tree-sitter-language-pack==1.8.1: segfaults
• tree-sitter==0.26.0 + tree-sitter-language-pack==1.12.1: segfaults

I also re-tested clean trailmark==0.4.0 with this pinned install:

trailmark==0.4.0
tree-sitter==0.25.2
tree-sitter-language-pack==1.12.1

That successfully parsed all six previously crashing repro files:

• ValantisLabs__valantis-stex/src/STEXAMM.sol
• chronicleprotocol__scribe/src/Scribe.sol
• uniswap__v3-core/contracts/UniswapV3Pool.sol
• uniswap__v3-core/contracts/libraries/Oracle.sol
• uniswap__v3-core/test/Oracle.spec.ts
• base__base/actions/harness/src/batcher/actor.rs

So the immediate workaround is to pin tree-sitter<0.26, specifically tree-sitter==0.25.2 worked in my tests. The current Trailmark dependency range tree-sitter>=0.25,<1.0 lets pip resolve to 0.26.0, which reproduces the segfault.

[tob-scott-a]

If I understand correctly, #64 should have fixed this. I'll cut a release as soon as I'm certain the issues are resolved.