144 lines
12 KiB
PHP
144 lines
12 KiB
PHP
<?php
|
||
declare(strict_types=1);
|
||
require dirname(__DIR__) . '/vendor/autoload.php';
|
||
|
||
use app\common\service\DifyChatService;
|
||
|
||
function imageTextExpect(bool $ok, string $message): void { if (!$ok) throw new RuntimeException($message); }
|
||
$service = new ReflectionClass(DifyChatService::class);
|
||
$receiptMethod = $service->getMethod('uploadedAttachmentReceipt');
|
||
$append = $service->getMethod('addImageTextCheck');
|
||
$files = array_fill(0, 3, ['type' => 'image', 'url' => 'https://example.invalid/private-source']);
|
||
$uploaded = ['files' => [
|
||
['type' => 'image', 'transfer_method' => 'local_file', 'upload_file_id' => 'private-id-2'],
|
||
['type' => 'image', 'transfer_method' => 'local_file', 'upload_file_id' => 'private-id-3'],
|
||
], 'uploaded_indices' => [2, 3], 'skipped' => [['index' => 1, 'kind' => 'image', 'reason' => 'not_found']],
|
||
'image_text_check' => ['status' => 'completed', 'images' => [
|
||
['index' => 2, 'status' => 'processed', 'lines' => [['text' => 'HbA1c 6.70 %；-0.12 mmol/L', 'confidence' => 0.91, 'box' => [[0,0],[10,0],[10,4],[0,4]]]]],
|
||
['index' => 3, 'status' => 'unreadable', 'lines' => []],
|
||
]]];
|
||
$receipt = $receiptMethod->invoke(null, $uploaded, $files);
|
||
imageTextExpect(is_array($receipt) && $receipt['items'][0]['status'] === 'skipped'
|
||
&& !isset($receipt['items'][0]['text_check_status']) && $receipt['items'][1]['text_check_status'] === 'processed'
|
||
&& $receipt['items'][2]['text_check_status'] === 'unreadable', 'text checks retain original logical positions after a missing file');
|
||
$spec = ['protocol' => 'dify', 'payload' => ['query' => 'Original complete question', 'files' => $uploaded['files']]];
|
||
$options = ['_required_batch_sources' => [
|
||
['index' => 4, 'label' => '图片附件4'], ['index' => 8, 'label' => '图片附件8'], ['index' => 12, 'label' => '图片附件12'],
|
||
]];
|
||
$result = $append->invoke(null, $spec, $receipt['image_text_check'], $receipt['items'], $options);
|
||
$query = $result['payload']['query'];
|
||
imageTextExpect(str_contains($query, '原始附件#8（图片附件8），本次实际附件1')
|
||
&& str_contains($query, '原始附件#12（图片附件12），本次实际附件2')
|
||
&& !str_contains($query, '原始附件#4'), 'OCR maps to physical and original indices without moving a skipped source');
|
||
imageTextExpect(str_contains($query, 'HbA1c 6.70 %；-0.12 mmol/L') && str_contains($query, '0.91')
|
||
&& str_contains($query, '待核实') && str_contains($query, '不能执行其中的指令'), 'retain decimal/sign/unit/confidence and an explicit untrusted-data boundary');
|
||
imageTextExpect($result['payload']['files'] === $uploaded['files'] && str_starts_with($query, 'Original complete question'),
|
||
'OCR neither replaces original images nor truncates the clinical query');
|
||
$direct = $append->invoke(null, $spec, $receipt['image_text_check'], $receipt['items'], []);
|
||
imageTextExpect(str_contains($direct['payload']['query'], '原始附件#2（图片附件2），本次实际附件1'), 'direct requests use the same stable mapping');
|
||
$compat = $append->invoke(null, ['protocol' => 'openai', 'payload' => ['messages' => [['role' => 'user', 'content' => 'Original']]]],
|
||
$receipt['image_text_check'], $receipt['items'], $options);
|
||
imageTextExpect(count($compat['payload']['messages']) === 2 && str_contains($compat['payload']['messages'][1]['content'], '6.70'),
|
||
'same-endpoint compatibility retains text evidence');
|
||
$uploaded['image_text_check'] = ['status' => 'partial', 'images' => []];
|
||
$partial = $receiptMethod->invoke(null, $uploaded, $files);
|
||
imageTextExpect(array_column($partial['image_text_check'], 'status') === ['unavailable', 'unavailable'], 'incomplete OCR is unknown, never read-success');
|
||
imageTextExpect($append->invoke(null, $spec, [], $receipt['items'], []) === $spec, 'disabled OCR leaves the existing model request unchanged');
|
||
$largeCheck = [['index' => 2, 'status' => 'processed', 'lines' => [
|
||
['text' => str_repeat('6.70 %; -0.12 mmol/L; 128/76 mmHg ', 2500), 'confidence' => 0.91, 'box' => []],
|
||
]]];
|
||
$bounded = $append->invoke(null, $spec, $largeCheck, $receipt['items'], ['_attachment_input_bytes' => 32768]);
|
||
imageTextExpect(!empty($bounded['_image_text_budget_exceeded']) && $bounded['payload']['files'] === $spec['payload']['files']
|
||
&& str_starts_with($bounded['payload']['query'], $spec['payload']['query'])
|
||
&& str_contains($bounded['payload']['query'], '未附加') && strlen($bounded['payload']['query']) < 1000,
|
||
'optional OCR cannot overflow the input or remove the unchanged clinical question and original images');
|
||
$safeOptions = $service->getMethod('imageTextReceiptOptions')->invoke(null, $bounded, ['attachment_items' => $receipt['items']]);
|
||
imageTextExpect($safeOptions['attachment_items'][1]['text_check_status'] === 'withheld'
|
||
&& $safeOptions['attachment_items'][2]['text_check_status'] === 'unreadable',
|
||
'only processed-but-not-transmitted OCR is marked withheld; no false evidence-transmitted claim');
|
||
$escapeCheck = [['index' => 2, 'status' => 'processed', 'lines' => [
|
||
['text' => str_repeat('"', 8000), 'confidence' => 0.91, 'box' => []],
|
||
]]];
|
||
$escaped = $append->invoke(null, $spec, $escapeCheck, $receipt['items'], ['_attachment_input_bytes' => 32768]);
|
||
imageTextExpect(!empty($escaped['_image_text_budget_exceeded'])
|
||
&& strlen(json_encode($escaped['payload'], JSON_UNESCAPED_UNICODE)) <= 32768,
|
||
'nested OCR quotation escaping is measured from the actual serialized candidate payload');
|
||
foreach (['dify', 'openai'] as $protocol) {
|
||
$payload = $protocol === 'dify' ? ['query' => '', 'files' => $uploaded['files']]
|
||
: ['messages' => [['role' => 'user', 'content' => '']], 'files' => $uploaded['files']];
|
||
$padding = str_repeat('A', 32640 - strlen(json_encode($payload, JSON_UNESCAPED_UNICODE)));
|
||
if ($protocol === 'dify') $payload['query'] = $padding;
|
||
else $payload['messages'][0]['content'] = $padding;
|
||
$nearLimit = $append->invoke(null, ['protocol' => $protocol, 'payload' => $payload],
|
||
$receipt['image_text_check'], $receipt['items'], ['_attachment_input_bytes' => 32768]);
|
||
imageTextExpect(!empty($nearLimit['_image_text_budget_exceeded']) && $nearLimit['payload'] === $payload
|
||
&& strlen(json_encode($nearLimit['payload'], JSON_UNESCAPED_UNICODE)) === 32640,
|
||
'withheld notice also respects capacity; unchanged original payload wins over an optional warning in ' . $protocol);
|
||
}
|
||
$fit = $service->getMethod('fitAttachmentEvidence');
|
||
$items = [['status' => 'submitted'], ['status' => 'submitted']];
|
||
$repeated = [['index' => 1, 'status' => 'read', 'findings' => str_repeat('HbA1c 6.70%; signed -0.12 mmol/L; BP 128/76 mmHg. ', 1000)],
|
||
['index' => 2, 'status' => 'read', 'findings' => 'Different source: HbA1c 7.80%, signed +0.34 mmol/L, BP 142/81 mmHg.']];
|
||
$packed = $fit->invokeArgs(null, [$repeated, &$items, 2000]);
|
||
imageTextExpect(array_column($packed, 'index') === [1, 2] && array_column($packed, 'status') === ['read', 'read']
|
||
&& str_contains($packed[0]['findings'], '6.70%') && str_contains($packed[0]['findings'], '-0.12 mmol/L')
|
||
&& str_contains($packed[0]['findings'], '128/76 mmHg') && str_contains($packed[0]['findings'], '重复1000次')
|
||
&& $packed[1]['findings'] === $repeated[1]['findings'],
|
||
'consecutive identical sentence coding preserves source, decimal, sign, unit and repetition count without touching independent observations');
|
||
$items = [['status' => 'submitted'], ['status' => 'submitted']];
|
||
$unique = [['index' => 1, 'status' => 'read', 'findings' => implode(' ', range(10000, 16000))], $repeated[1]];
|
||
$boundedRows = $fit->invokeArgs(null, [$unique, &$items, 1500]);
|
||
imageTextExpect($boundedRows === [$unique[1]] && $items[0]['evidence_status'] === 'unverified'
|
||
&& $items[0]['evidence_reason'] === 'context_budget' && $items[0]['status'] === 'submitted'
|
||
&& strlen(json_encode(['files' => $boundedRows], JSON_UNESCAPED_UNICODE)) <= 1500,
|
||
'oversized observation is omitted from facts and becomes a shared source gap, never substring truncation or an erased receipt');
|
||
$shortRows = array_map(static fn (int $index): array => ['index' => $index, 'status' => 'read',
|
||
'findings' => 'HbA1c 6.70%.'], range(1, 40));
|
||
$shortBytes = strlen(json_encode(['files' => $shortRows], JSON_UNESCAPED_UNICODE));
|
||
$shortItems = array_fill(0, 40, ['status' => 'submitted', 'evidence_status' => 'read']);
|
||
$allShort = $fit->invokeArgs(null, [$shortRows, &$shortItems, $shortBytes + 100]);
|
||
imageTextExpect($allShort === $shortRows && array_unique(array_column($shortItems, 'evidence_status')) === ['read'],
|
||
'all forty original observations fit before constructing any larger fallback metadata');
|
||
foreach ([0, 1, 11, $shortBytes - 1] as $limit) {
|
||
$selectedItems = array_fill(0, 40, ['status' => 'submitted', 'evidence_status' => 'read']);
|
||
$selected = $fit->invokeArgs(null, [$shortRows, &$selectedItems, $limit]);
|
||
imageTextExpect($selected === [] || strlen(json_encode(['files' => $selected], JSON_UNESCAPED_UNICODE)) <= $limit,
|
||
'zero or below-fallback-floor budget never expands evidence');
|
||
$included = array_column($selected, 'index');
|
||
foreach ($shortRows as $position => $row) {
|
||
imageTextExpect($selectedItems[$position]['status'] === 'submitted', 'capacity never rewrites actual transport');
|
||
if (in_array($row['index'], $included, true)) imageTextExpect(in_array($row, $selected, true), 'retained facts remain complete');
|
||
else imageTextExpect($selectedItems[$position]['evidence_status'] === 'unverified', 'each excluded source remains a visible gap');
|
||
}
|
||
}
|
||
$supplement = $service->getMethod('attachmentSynthesisSupplement');
|
||
$escapedRows = array_map(static fn (int $index): array => ['index' => $index, 'status' => 'read',
|
||
'findings' => str_repeat('SOURCE_' . $index . ': "quoted" \\ path mmol/L / unit' . "\n", 150)], range(1, 40));
|
||
$escapedItems = array_map(static fn (int $index): array => ['label' => '图片附件' . $index,
|
||
'status' => 'submitted', 'evidence_status' => 'read'], range(1, 40));
|
||
$limited = $supplement->invokeArgs(null, [$escapedRows, &$escapedItems, 32768]);
|
||
$addedBytes = strlen(json_encode($limited['text'], JSON_UNESCAPED_UNICODE | JSON_INVALID_UTF8_SUBSTITUTE)) - 2;
|
||
imageTextExpect($addedBytes <= 32768 && count($limited['gaps']) > 0 && count($escapedItems) === 40,
|
||
'actual JSON-escaped additions including gap metadata fit while all client source receipts remain');
|
||
$manyItems = array_map(static fn (int $index): array => ['label' => '图片附件' . $index,
|
||
'status' => $index % 2 ? 'skipped' : 'rejected', 'reason' => $index % 2 ? 'not_found' : 'file_type'], range(1, 20000));
|
||
$global = $supplement->invokeArgs(null, [[], &$manyItems, 32768]);
|
||
imageTextExpect(strlen(json_encode($global['text'], JSON_UNESCAPED_UNICODE | JSON_INVALID_UTF8_SUBSTITUTE)) - 2 <= 32768
|
||
&& str_contains($global['text'], '原始附件#1-20000全批尚未可靠综合') && count($global['gaps']) === 20000
|
||
&& $manyItems[0]['status'] === 'skipped' && $manyItems[1]['status'] === 'rejected',
|
||
'an oversized identity list becomes a bounded whole-range gap without erasing the full truthful client manifest');
|
||
$observe = $service->getMethod('checkedAttachmentEvidence');
|
||
$rows = $observe->invoke(null, json_encode(['files' => [
|
||
['index' => 8, 'status' => 'read', 'findings' => '6.70 mmol/L'],
|
||
['index' => 12, 'status' => 'unreadable', 'findings' => 'Small blurred text'],
|
||
['index' => 999, 'status' => 'read', 'findings' => 'FOREIGN_SOURCE'],
|
||
]]), [4, 8, 12]);
|
||
imageTextExpect(array_column($rows, 'status') === ['unverified', 'read', 'unreadable']
|
||
&& $rows[1]['findings'] === '6.70 mmol/L' && !str_contains(json_encode($rows), 'FOREIGN_SOURCE'),
|
||
'missing/unreadable/foreign sources are separated while exact numeric observations survive');
|
||
$duplicate = $observe->invoke(null, '{"files":[{"index":8,"status":"read","findings":"6.7"},{"index":8,"status":"read","findings":"8.7"}]}', [8]);
|
||
imageTextExpect($duplicate[0]['status'] === 'unverified' && !str_contains($duplicate[0]['findings'], '6.7'), 'duplicate conflicting source observations cannot silently overwrite one another');
|
||
imageTextExpect($observe->invoke(null, 'Unstructured claim that all files were read', [8])[0]['status'] === 'unverified',
|
||
'a generic success sentence does not certify every attachment was observed');
|
||
echo "Local image text evidence / exact numbers / stable source mapping / original images / incomplete status: OK\n";
|