mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-11 03:38:38 +00:00
c7e5f71251
Branch on status before reading upstream error bodies and stream at most the error cap for document fetch, Reducto upload, Azure poll, and primary provider responses; add a chunked no-Content-Length oversized-error regression. Reuse a bounded cache of security-validating document clients keyed by scheme, host, port, and the full validated address set so redirects across ports cannot reuse stale pins, replacing per-call client construction while preserving the anti-rebinding contract. Import the optional native RustOcrInputError directly behind ImportError instead of getattr, keeping the test override. Pin the url workspace dependency to =2.5.8 and expand the IPv6 denylist to NAT64 64:ff9b::/96 and 64:ff9b:1::/48 with embedded private or metadata IPv4, discard 100::/64, Teredo 2001::/32, benchmarking 2001:2::/48, ORCHID and ORCHIDv2, and 6to4 2002::/16 with embedded private IPv4, keeping globally routable targets allowed. Remove PR-added comments and docstrings across Rust, Python, and tests. Co-Authored-By: Ishaan Jaffer <155045088+ishaan-berri@users.noreply.github.com> |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| test_ocr_file_input.py | ||
| test_rust_bridge.py | ||