wppaste
WordPress

AbstractOpenAiCompatibleTextGenerationModel::parseResponseChoiceToCandidate( WordPress\AiClient\Providers\OpenAiCompatibleImplementation\ChoiceData $choiceData, int $index ): WordPress\AiClient\Results\DTO\Candidate

Since
0.1.0
Source
wp-includes/php-ai-client/src/Providers/OpenAiCompatibleImplementation/AbstractOpenAiCompatibleTextGenerationModel.php:463
Parses a single choice from the API response into a Candidate object.

Compatibility

WordPress
since 0.1.0
PHP
7.4–8.6-dev
  • 6.7.7
  • 6.8.8
  • 6.9.7
  • 7.0.4
  • 7.1.0

Present in 2 of the 5 tracked releases, added in 0.1.0, and compiles on PHP 7.4 through 8.6-dev.

Parameters

$choiceDataWordPress\AiClient\Providers\OpenAiCompatibleImplementation\ChoiceData
The choice data from the API response.
$indexint
The index of the choice in the choices array.

Return value

WordPress\AiClient\Results\DTO\Candidate
The parsed candidate.

Performance profile

How much work a call to AbstractOpenAiCompatibleTextGenerationModel::parseResponseChoiceToCandidate() does, and what it touches: the algorithmic scaling, the Zend instruction count per call across PHP versions, the hooks it hands control to, and the core code that calls it. Measured from the compiled opcodes, not a stopwatch, so every number is identical on any machine running the same PHP version, and every function in core is ranked by cost.

Cost class
Trivial

Touches nothing outside its own arguments.

Scaling
Constant

No loop in the body: the same number of instructions runs whatever you pass in.

Instructions
16–51

Executed per call on PHP 8.5, depending on the branch taken. The body compiles to 98.

Plugin surface
None

Nothing here hands control to plugin code.

Called by
0

Nothing in core calls this; the cost is only what you spend yourself.

What one call costs · 4 distinct outcomes

One number would be a lie: the work depends on which branch runs. These are every distinct cost AbstractOpenAiCompatibleTextGenerationModel::parseResponseChoiceToCandidate() can have, taken from its control-flow graph on PHP 8.5.

WhenInstructionsCalls it makes
always16–19->providerMetadata(), ->getName(), ::ResponseException()
isset($choiceData)24–29array_is_list(), ->providerMetadata(), ->getName(), ::ResponseException()
isset($choiceData) && !array_is_list()36–44array_is_list(), ->parseResponseChoiceMessage(), ::FinishReasonEnum()
isset($choiceData) && !array_is_list()42–51array_is_list(), ->parseResponseChoiceMessage(), ->providerMetadata(), ->getName(), ::ResponseException()

Across PHP versions

PHPCompiledExecutedBranchesNotes
8.6-dev9816–5110
8.59816–5110
8.49816–51101 fewer instruction than PHP 8.3
8.39916–5210
8.29916–52101 more instruction than PHP 8.1
8.19816–51101 fewer instruction than PHP 7.4
7.49916–5210

An instruction is not a fixed amount of time, so a matching count is not necessarily the same speed; what it rules out is a difference in the work itself.

Uses · 10

  • array_is_list()Polyfill for `array_is_list()` function added in PHP 8.1.
  • \WordPress\AiClient\Providers\Http\Exception\ResponseException::fromMissingData()
  • \WordPress\AiClient\Providers\OpenAiCompatibleImplementation\AbstractOpenAiCompatibleTextGenerationModel::providerMetadata()
  • \WordPress\AiClient\Providers\OpenAiCompatibleImplementation\AbstractOpenAiCompatibleTextGenerationModel::parseResponseChoiceMessage()
  • \WordPress\AiClient\Results\Enums\FinishReasonEnum::stop()
  • \WordPress\AiClient\Results\Enums\FinishReasonEnum::length()
  • \WordPress\AiClient\Results\Enums\FinishReasonEnum::contentFilter()
  • \WordPress\AiClient\Results\Enums\FinishReasonEnum::toolCalls()
  • \WordPress\AiClient\Providers\Http\Exception\ResponseException::fromInvalidData()
  • \WordPress\AiClient\Results\DTO\Candidate::__construct()

Source code

    protected function parseResponseChoiceToCandidate(array $choiceData, int $index): Candidate    {        if (!isset($choiceData['message']) || !is_array($choiceData['message']) || array_is_list($choiceData['message'])) {            throw ResponseException::fromMissingData($this->providerMetadata()->getName(), "choices[{$index}].message");        }        if (!isset($choiceData['finish_reason']) || !is_string($choiceData['finish_reason'])) {            throw ResponseException::fromMissingData($this->providerMetadata()->getName(), "choices[{$index}].finish_reason");        }        $messageData = $choiceData['message'];        $message = $this->parseResponseChoiceMessage($messageData, $index);        switch ($choiceData['finish_reason']) {            case 'stop':                $finishReason = FinishReasonEnum::stop();                break;            case 'length':                $finishReason = FinishReasonEnum::length();                break;            case 'content_filter':                $finishReason = FinishReasonEnum::contentFilter();                break;            case 'tool_calls':                $finishReason = FinishReasonEnum::toolCalls();                break;            default:                throw ResponseException::fromInvalidData($this->providerMetadata()->getName(), "choices[{$index}].finish_reason", sprintf('Invalid finish reason "%s".', $choiceData['finish_reason']));        }        return new Candidate($message, $finishReason);    }

Changelog

Introduced in 0.1.0. Unchanged from 7.0.4 through 7.1.0.

  1. 7.0.4
  2. 7.1.0

Signature, return type and hooks compared across 2 parsed releases.

About this page

Parsed data
Generated from the wordpress-develop 7.1.0 tag, from src/wp-includes/php-ai-client/src/Providers/OpenAiCompatibleImplementation/AbstractOpenAiCompatibleTextGenerationModel.php, and regenerated for each WordPress release so it tracks the code rather than a snapshot of it.
Corrections
Something wrong on this page? Report it and it gets fixed in the next regeneration.