UPSTREAM PR #17679: server: fixing naming conflict res_error in server-models.cpp

[loci-dev]

Mirrored from ggml-org/llama.cpp#17679

Fix naming conflict with res_error from resolv.h (used by httplib).
This is related to PR #17243
The issue persists in the latest commit, so this PR addresses it.

[loci-review]

Explore the complete analysis inside the Version Insights

Performance Analysis Summary: PR #400

Overview

This PR renames the static function res_error to res_err in tools/server/server-models.cpp to resolve a symbol conflict with resolv.h. The change affects 1 file with 8 call site updates. Performance analysis shows zero measurable impact across all binaries and functions.

Performance Metrics

• Response Time Change: 0 ns across all functions
• Throughput Change: 0 ns across all functions
• Power Consumption: All binaries show changes within ±0.0002%, equivalent to sub-nanosecond execution time differences

Impact on Inference Performance

No impact on tokens per second. The modified code is in the server routing layer (server-models.cpp), which handles HTTP request validation and model management. The core inference functions (llama_decode, llama_encode, llama_tokenize) remain unchanged. No tokenization or inference paths are affected by this rename.

Binary-Level Analysis

Power consumption analysis across 16 binaries shows negligible changes:

• libllama.so: +0.3 nJ (+0.0002%)
• llama-cvector-generator: +0.25 nJ (+0.0001%)
• llama-tts: -0.45 nJ (-0.0002%)
• All other binaries: 0.0% change

These sub-nanosecond variations are within measurement noise and represent no functional performance difference.

Conclusion

This is a symbol conflict fix with zero performance impact. The function rename is purely cosmetic at the binary level, with identical compiled output. No inference paths, tokenization logic, or performance-critical areas are affected.