chore: enable schema support for perplexity #14
Conversation
WalkthroughPerplexity adapter: isSchemaSupported() now returns true. Processing of chunks that begin with HTML now delegates to a new protected sanitizeHtmlError(string $html) method which extracts/formats error info (title, h1, optional status); previously raw HTML was returned. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant Client
participant PerplexityAdapter as Perplexity
participant HTMLParser as sanitizeHtmlError()
Client->>PerplexityAdapter: send chunk (response)
alt chunk starts with '<' (HTML)
PerplexityAdapter->>HTMLParser: sanitizeHtmlError(html)
HTMLParser-->>PerplexityAdapter: formatted error string
PerplexityAdapter-->>Client: return readable error
else non-HTML chunk
PerplexityAdapter-->>Client: process/return chunk as before
end
Note over PerplexityAdapter: isSchemaSupported() now returns true
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Poem
Pre-merge checks and finishing touches✅ Passed checks (3 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (1)
src/Agents/Adapters/Perplexity.php (1)
74-77: Approve enabling JSON schema support for Perplexity
Verified that Perplexity’s API supports structured outputs via JSON Schema (response_format.type = "json_schema"withjson_schema.schema). The change toisSchemaSupported()is correct. Consider adding an integration test to validate structured responses (accounting for initial schema prep delay and any<think>prefix).
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (2)
src/Agents/Adapters/Perplexity.php (2)
125-132: Good: sanitize HTML error responses early; consider content-type checkThis improves UX vs returning raw HTML. Optionally, also detect via Content-Type header (text/html) if available on Chunk to avoid false positives on unusual payloads.
Can Chunk expose headers (e.g., getHeaders/getHeader)? If yes, we can branch on Content-Type first and keep the current tag-prefix check as fallback.
166-205: Improve HTML parsing and output consistencyRecommend small tweaks: decode HTML entities, broaden status parsing (allow “401 - …”/“401: …”), and avoid leading newline to keep error formatting consistent.
Confirm that removing the leading newline doesn’t break any tests/consumers expecting it.
Apply this diff:
protected function sanitizeHtmlError(string $html): string { - // Try to extract the error from the title tag - if (preg_match('/<title>(.*?)<\/title>/is', $html, $matches)) { - $errorMessage = trim($matches[1]); - - // Extract status code and message if present (e.g., "401 Authorization Required") - if (preg_match('/^(\d{3})\s+(.+)$/i', $errorMessage, $parts)) { - $statusCode = $parts[1]; - $message = $parts[2]; - - return PHP_EOL.'(http_'.$statusCode.') '.$message; - } - - return PHP_EOL.'(html_error) '.$errorMessage; - } + // Try to extract the error from the title tag + if (preg_match('/<title>(.*?)<\/title>/is', $html, $matches)) { + $errorMessage = html_entity_decode(trim(strip_tags($matches[1])), ENT_QUOTES | ENT_HTML5); + + // Extract status code and message if present (e.g., "401 - Authorization Required" or "401: Authorization Required") + if (preg_match('/^(\d{3})\s*[-:—]?\s*(.+)$/i', $errorMessage, $parts)) { + $statusCode = $parts[1]; + $message = $parts[2]; + + return '(http_'.$statusCode.') '.$message; + } + + return '(html_error) '.$errorMessage; + } // Try to extract from h1 tag - if (preg_match('/<h1>(.*?)<\/h1>/is', $html, $matches)) { - $errorMessage = trim(strip_tags($matches[1])); - - if (preg_match('/^(\d{3})\s+(.+)$/i', $errorMessage, $parts)) { - $statusCode = $parts[1]; - $message = $parts[2]; - - return PHP_EOL.'(http_'.$statusCode.') '.$message; - } - - return PHP_EOL.'(html_error) '.$errorMessage; - } + if (preg_match('/<h1>(.*?)<\/h1>/is', $html, $matches)) { + $errorMessage = html_entity_decode(trim(strip_tags($matches[1])), ENT_QUOTES | ENT_HTML5); + + if (preg_match('/^(\d{3})\s*[-:—]?\s*(.+)$/i', $errorMessage, $parts)) { + $statusCode = $parts[1]; + $message = $parts[2]; + + return '(http_'.$statusCode.') '.$message; + } + + return '(html_error) '.$errorMessage; + } // Fallback for unrecognized HTML errors - return PHP_EOL.'(html_error) Received HTML error response from API'; + return '(html_error) Received HTML error response from API'; }
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (3)
src/Agents/Adapters/Perplexity.php (3)
175-176: Consider usingstrip_tags()on title content for consistency.Line 191 uses
strip_tags()when extracting from the<h1>tag, but line 176 doesn't apply the same treatment to the<title>content. If the title contains nested HTML tags, they would appear in the error message.Apply this diff to maintain consistency:
if (preg_match('/<title>(.*?)<\/title>/is', $html, $matches)) { - $errorMessage = trim($matches[1]); + $errorMessage = trim(strip_tags($matches[1])); // Extract status code and message if present (e.g., "401 Authorization Required")
179-184: Optional: Extract duplicate status code parsing logic.The status code extraction logic on lines 179-184 and 193-198 is identical. Consider extracting it into a helper method or restructuring to parse the status code once after extracting the error message from either title or h1.
Example refactor:
protected function sanitizeHtmlError(string $html): string { $errorMessage = null; // Try to extract the error from the title tag if (preg_match('/<title>(.*?)<\/title>/is', $html, $matches)) { $errorMessage = trim(strip_tags($matches[1])); } // Try to extract from h1 tag if not found if ($errorMessage === null && preg_match('/<h1>(.*?)<\/h1>/is', $html, $matches)) { $errorMessage = trim(strip_tags($matches[1])); } // Parse status code if message was found if ($errorMessage !== null) { if (preg_match('/^(\d{3})\s+(.+)$/i', $errorMessage, $parts)) { return '(http_'.$parts[1].') '.$parts[2]; } return '(html_error) '.$errorMessage; } // Fallback for unrecognized HTML errors return '(html_error) Received HTML error response from API'; }Also applies to: 193-198
74-77: JSON schema support confirmed; adjust parsing and add tests.
- Perplexity supports structured JSON schema across all
getModels()values.- Reasoning models (e.g., sonar-reasoning, sonar-reasoning-pro) emit a
<think>prefix—strip this before JSON parsing.- Add an integration test invoking both standard and reasoning models to verify schema support.
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
src/Agents/Adapters/Perplexity.php(3 hunks)
🔇 Additional comments (1)
src/Agents/Adapters/Perplexity.php (1)
131-131: LGTM! Improved error handling.Delegating to
sanitizeHtmlError()significantly improves the user experience by providing readable error messages instead of raw HTML.
Summary by CodeRabbit
New Features
Bug Fixes