user-agents-v2

Comprehensive open-source collection of broadly-compatible regular expression patterns to identify and analyze podcast player user agents.

Quick start

Given a HTTP User-Agent found in your podcast episode server logs, to find a deterministic entity match:

• Remove any newlines (never occurs except from bad actors)
• Iterate the following json files from the src directory in this order: bots, apps, libraries, browsers
• Iterate the pattern file entries array in order, returning the first entry where pattern matches the User-Agent
• This will always result in either 0 or 1 entry
• If found, the containing file can be used as the type of the entry (e.g. bot if found in the bots file)

(Optional) If type is not bot, to additionally break down by device:

• Iterate the devices pattern file entries array in order, returning the first entry where pattern matches the User-Agent
• This will always result in either 0 or 1 entry
• If found, the device will also have a category for high-level category such as mobile, smart_speaker, or computer

(Optional) If type is browser and you also have the HTTP Referer header in your logs, to additionally break down by known web apps:

• Remove any newlines (never occurs except from bad actors)
• Iterate the referrers pattern file entries array in order, returning the first entry where pattern matches the Referer
• This will always result in either 0 or 1 entry
• If found, the referrer entity may also have a category of app (for web-based apps) or host (for podcast hosting company players)

Approach

This collection is an evolution of the original OPAWG User agent list, refactored in some ways and overlaid with ideas from the excellent Buzzsprout Podcast Agents Ruby gem.

Some of the goals of this collection:

• Data files instead of code: This is not an NPM package or dependent on any specific programming environment or platform, the patterns files are in standard JSON format, with a shared JSON schema (automatic type-checking and code-assist in IDEs). There is no code outside of the .github/workflows folder, which runs the automated tests on every change.
• Deterministic: Given a User-Agent, everyone should end up with the same unique result (0 or 1 entry), regardless of programming language or environment.
• Fast and compatible: Keep to basic regular expression patterns, avoid features such as lookaheads that are expensive or unavailable in certain environments.
• Comprehensive: Goal is to match the vast majority of current podcast user-agents in the wild.
• Multi-dimensional: While basic entity matching is deterministic, optional breakdowns by device and device category are separated out into a separate file - making the patterns simpler, and focusing on attributes useful for standard podcast stats reporting.
• Web-aware: Optionally identify known web-based apps and other players using the Referer header, given that web-based apps cannot set a custom User-Agent.
• Testable: Examples are included attached to the entries themselves, automated checks are run against the patterns after every push and pull request, to ensure quality going forward.
• Easy to contribute: Help make this collection better by adding examples to an existing file, or adding new entries. Sanity checks will run automatically on any pending pull requests.

Evolution

These patterns were initially created with a one-time automated transform of the original OPAWG User agent list, with the following transformations:

• Converted top-level array to multiple top-level files by type, each with a top-level object - easier to deserialize in some environments than a top-level array.
• Removed unnecessary forward-slash escaping \/ in patterns.
• Merged duplicate entries into a single entry, then sorted alphabetically.
• Removed lookheads, re-ordered certain entries if necessary to emulate.
• Combined multiple expressions for a single entry into a single regex pattern (separated by |), simpler and faster than compiling multiple patterns per entry.
• Fixed some of the incorrect patterns.
• Dropped support for OPAWG "device" and "os" attributes. Instead, introduced a new devices entries file ported from Buzzsprout's Ruby code. Simplified patterns that no longer needed app+device+os specificity.
• Added a JSON schema, fixed validation errors found - like incorrect info urls.
• Imported several new entries and examples from Buzzsprout's test data file.
• Ran against a large set of found data, and added yet more entries and examples.
• Fixed issues found when running against the new automated checks. In addition to basic JSON-level data checks, the automated tests ensure each example matches against its parent entry when running through the deterministic matching algorithm mentioned above.