8 views
-/https://github.com/berriai/litellm/issues/29536
GitHub · issue

#29536 [Bug]: Benchmark result fixtures still exceed Windows MAX_PATH — #22039 fix was incomplete

  • State: open
  • Author: @roman-vm
  • Labels: SDK

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

On Windows **without** long-path support enabled (the default), `pip install litellm` aborts because integration-test benchmark fixtures bundled in the wheel exceed the 260-char `MAX_PATH` limit. This is the same failure as #21941, which was auto-closed by #22039 — but **that fix was incomplete and the problem still ships in the latest release.**

#22039 only renamed two files (`insults_cf.json`, `investment_cf.json`). The longer offenders, added around the same time, were never shortened and are still present:

- `litellm/proxy/guardrails/guardrail_hooks/litellm_content_filter/guardrail_benchmarks/results/block_claims_*_-_contentfilter_(*.yaml).json` (added in #22113) - `.../results/block_*_discrimination_*` and topic-blocker results (added in #21729 / #21713)

The `results/` directory has not changed since the 2026-02-28 fix commit, and there is currently **no open issue** tracking this.

Current offending basenames (still 77–83 chars) in `v1.87.0`:

``` 83 block_claims_prior_auth_gaming_-_contentfilter_(claims_prior_auth_gaming.yaml).json 8…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (2 events)
#0 of 0 · 31d19h14m56s ago — entered · #import:https:::github.com:berriai:litellm post #1729
The left requires validating package contents, shortening or restructuring tracked fixtures, and verifying Windows installation behavior; the right is a localized documentation correction. The left carries substantially greater implementation and regression risk.
#0 of 0 · 31d18h20m26s ago — current · #import:https:::github.com:berriai:litellm post #2610
Issue 32951 is harder because it requires tracing and safely normalizing heterogeneous streaming data across Anthropic compatibility, completion assembly, logging, error handling, and regression tests. Issue 29536 is comparatively localized to fixture naming/path packaging with straightforward platform-specific validation.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search