Skip to content

Document input size guidance for HTTP-exposed ISCC generation #25

Description

@titusz

Problem

None of the content-code IEPs (IEP-0003 through IEP-0009) specify input size limits. For Data-Code and Instance-Code this is fine (streaming hashers handle arbitrarily large inputs), but several functions process full inputs in memory:

  • gen_text_code_v0 (IEP-0003): Full text n-gram generation in memory
  • gen_image_code_v0 (IEP-0004): Image pixel array in memory
  • gen_audio_code_v0 (IEP-0005): Chromaprint feature array in memory
  • gen_video_code_v0 (IEP-0006): Frame signature array in memory

When ISCC generation is exposed via HTTP APIs, unbounded inputs create denial-of-service potential.

Proposal

Add implementer guidance (as a section in existing IEPs or as a standalone security considerations document):

  • When exposing ISCC generation via HTTP, implementers SHOULD enforce request-level size limits.
  • Text content >10 MB provides diminishing returns for similarity hash quality.
  • Audio chromaprint arrays >1M entries and video frame arrays >10K frames exceed practical utility.

This keeps the core algorithms unconstrained (no spec changes needed) while giving API implementers the security guidance they need.

Checklist

  • Security considerations section or guidance document created
  • Recommended size limits documented for web-exposed deployments

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions