[FEATURE] Have custom Regexp timeout configuration for logging

[nina-instrumentl]

Scope check

• This is core LLM communication (not application logic)
• This benefits most users (not just my use case)
• This can't be solved in application code with current RubyLLM
• I read the Contributing Guide

Due diligence

• I searched existing issues
• I checked the documentation

What problem does this solve?

When processing complex files through ruby_llm, the application encounters a timeout error during request logging:

Regexp::TimeoutError: regexp match timeout.

This happens because the logger filter is using default Regexp.timeout for its regular expression: https://github.com/crmne/ruby_llm/blob/main/lib/ruby_llm/connection.rb#L65

This default behavior becomes problematic when ruby_llm is the only part of the application handling large/complex payloads, yet the timeout applies globally.

Proposed solution

Introduce a configurable log_regexp_timeout option in ruby_llm. This would allow consumers to set a custom timeout value for regex operations used within logging filters, separate from the global Regexp.timeout.

Update filters to use custom timeout that is separate from the global Ruby timeout in https://github.com/crmne/ruby_llm/blob/main/lib/ruby_llm/connection.rb#L60

 logger.filter(Regexp.new('[A-Za-z0-9+/=]{100,}', timeout: @config.log_regexp_timeout), 'data":"[BASE64 DATA]"')
 logger.filter(Regexp.new('[-\\d.e,\\s]{100,}', timeout: @config.log_regexp_timeout), '[EMBEDDINGS ARRAY]')

If log_regexp_timeout is not provided, fall back to Regexp.timeout to preserve default behavior.

Why this belongs in RubyLLM

• Regexp.timeout was introduced in Ruby 3.2 to mitigate DDoS attacks. However, it's too rigid when an application processes both simple and complex content.
• ruby_llm is the only component in many applications that deals with large structured data (e.g., base64, embedding vectors), making it a natural place to encapsulate this configuration.
• Since ruby_llm defines its own Faraday logging filters, it's best positioned to support fine-tuned timeout handling without affecting the rest of the application.

[crmne]

That'd be cool to have. Do you want to take a crack at it?

[nina-instrumentl]

@crmne Do I need API keys for all providers in order to pass ruby_llm pre-commit hooks?

Here is a code change (main...nina-instrumentl:ruby_llm:add_custom_log_regexp_timeout), and I tested it last week. BUT committing from the command line fails on the pre-commit hooks.

Running rspec before committing passes, but when I try to commit on the command line the same code, it fails. Examining the changes caused by pre-commit hooks, I see a lot of models removed in aliases.json and models.json.
I haven't looked further, but I am assuming I will need API keys for all of the providers (which I don't currently have). Would you be able to try running the pre-commit on your setup to see if it passes for you?

[crmne]

@nina-instrumentl This is a real problem we have and I hear you! The rake models:update task should not be a bottleneck in contributing. I'll think about a good solution and update you.

[crmne]

@nina-instrumentl I removed the overcommit step that regenerates the model files. I'll just remember to run it periodically myself. 3a904fd

[nina-instrumentl]

Thank you!