230 lines
7.5 KiB
TypeScript
230 lines
7.5 KiB
TypeScript
import { nanoid } from 'nanoid';
|
|
import {
|
|
buildDocumentBundle,
|
|
documentArtifactToParsedPdfContent,
|
|
getDocumentExtractorProviders,
|
|
getMediaExtractorProviders,
|
|
MAX_DOCUMENT_BUNDLE_FILES,
|
|
MAX_DOCUMENT_BUNDLE_TOTAL_SIZE_BYTES,
|
|
} from '@/lib/document';
|
|
import { normalizeDocumentMimeType, SUPPORTED_MEDIA_MIME_TYPES } from '@/lib/document/mime';
|
|
import type { MediaArtifact } from '@/lib/document/types';
|
|
import {
|
|
getServerPDFProviders,
|
|
isServerConfiguredProvider,
|
|
resolveManagedAliDocMindCredentials,
|
|
resolvePDFApiKey,
|
|
resolvePDFBaseUrl,
|
|
} from '@/lib/server/provider-config';
|
|
|
|
export const MAX_COURSE_MATERIAL_FILE_SIZE_BYTES = 50 * 1024 * 1024;
|
|
|
|
export class CourseMaterialError extends Error {
|
|
constructor(
|
|
readonly code: string,
|
|
readonly status: number,
|
|
message: string,
|
|
) {
|
|
super(message);
|
|
this.name = 'CourseMaterialError';
|
|
}
|
|
}
|
|
|
|
function mediaArtifactText(artifact: MediaArtifact): string {
|
|
const sections: string[] = [];
|
|
const synopsis =
|
|
artifact.providerRaw &&
|
|
typeof artifact.providerRaw === 'object' &&
|
|
'synopsis' in artifact.providerRaw
|
|
? String((artifact.providerRaw as { synopsis?: unknown }).synopsis ?? '').trim()
|
|
: '';
|
|
if (synopsis) sections.push(`## Synopsis\n\n${synopsis}`);
|
|
if (artifact.transcript?.length) {
|
|
const transcript = artifact.transcript
|
|
.filter((segment) => segment.text.trim())
|
|
.map((segment) => `[${Math.floor(segment.startMs / 1000)}s] ${segment.text.trim()}`)
|
|
.join('\n');
|
|
if (transcript) sections.push(`## Transcript\n\n${transcript}`);
|
|
}
|
|
if (artifact.keyframes?.length) {
|
|
const keyframes = artifact.keyframes
|
|
.map((keyframe) => keyframe.description || keyframe.ocrText || '')
|
|
.filter(Boolean)
|
|
.join('\n');
|
|
if (keyframes) sections.push(`## Keyframes\n\n${keyframes}`);
|
|
}
|
|
return sections.join('\n\n');
|
|
}
|
|
|
|
function managedDocumentProvider(mimeType: string) {
|
|
const configured = new Set(Object.keys(getServerPDFProviders()));
|
|
const candidates = getDocumentExtractorProviders().filter(
|
|
(provider) =>
|
|
provider.supportedMimeTypes.includes(mimeType) &&
|
|
(provider.id === 'plain-text' || provider.id === 'unpdf' || configured.has(provider.id)),
|
|
);
|
|
// Prefer an operator-managed high-fidelity extractor. Plain text and unpdf
|
|
// are safe in-process fallbacks and are selected only when no managed
|
|
// provider for the MIME exists.
|
|
return (
|
|
candidates.find((provider) => isServerConfiguredProvider('pdf', provider.id)) ?? candidates[0]
|
|
);
|
|
}
|
|
|
|
function managedMediaProvider(mimeType: string) {
|
|
return getMediaExtractorProviders().find(
|
|
(provider) =>
|
|
provider.supportedMimeTypes.includes(mimeType) &&
|
|
isServerConfiguredProvider('pdf', provider.id),
|
|
);
|
|
}
|
|
|
|
function managedExtractorConfig(providerId: string) {
|
|
const ali = providerId === 'alidocmind' ? resolveManagedAliDocMindCredentials() : undefined;
|
|
return {
|
|
providerId,
|
|
apiKey: resolvePDFApiKey(providerId),
|
|
baseUrl: ali?.baseUrl ?? resolvePDFBaseUrl(providerId),
|
|
accessKeyId: ali?.accessKeyId,
|
|
accessKeySecret: ali?.accessKeySecret,
|
|
allowEnvFallback: isServerConfiguredProvider('pdf', providerId),
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Extract files with server-managed providers and fold them into the exact
|
|
* pdfContent shape consumed by the original classroom/framework runners.
|
|
* Caller-supplied provider credentials and storage keys never enter this API.
|
|
*/
|
|
export async function extractManagedCourseMaterials(
|
|
files: readonly File[],
|
|
): Promise<{ text: string; images: string[] }> {
|
|
if (files.length === 0) return { text: '', images: [] };
|
|
if (files.length > MAX_DOCUMENT_BUNDLE_FILES) {
|
|
throw new CourseMaterialError(
|
|
'too_many_materials',
|
|
413,
|
|
`At most ${MAX_DOCUMENT_BUNDLE_FILES} course materials are allowed`,
|
|
);
|
|
}
|
|
const totalBytes = files.reduce((total, file) => total + file.size, 0);
|
|
if (totalBytes > MAX_DOCUMENT_BUNDLE_TOTAL_SIZE_BYTES) {
|
|
throw new CourseMaterialError(
|
|
'materials_too_large',
|
|
413,
|
|
'Course materials exceed the 150 MiB aggregate limit',
|
|
);
|
|
}
|
|
|
|
const parts = await Promise.all(
|
|
files.map(async (file, index) => {
|
|
if (file.size <= 0 || file.size > MAX_COURSE_MATERIAL_FILE_SIZE_BYTES) {
|
|
throw new CourseMaterialError(
|
|
'material_too_large',
|
|
413,
|
|
`Material "${file.name}" must be between 1 byte and 50 MiB`,
|
|
);
|
|
}
|
|
const mimeType = normalizeDocumentMimeType({ mimeType: file.type, fileName: file.name });
|
|
if (!mimeType) {
|
|
throw new CourseMaterialError(
|
|
'unsupported_material',
|
|
415,
|
|
`Unsupported course material type for "${file.name}"`,
|
|
);
|
|
}
|
|
const buffer = Buffer.from(await file.arrayBuffer());
|
|
const source = {
|
|
id: `material_${nanoid(8)}`,
|
|
name: file.name || `material-${index + 1}`,
|
|
size: file.size,
|
|
mimeType,
|
|
order: index + 1,
|
|
};
|
|
|
|
if (SUPPORTED_MEDIA_MIME_TYPES.includes(mimeType)) {
|
|
const provider = managedMediaProvider(mimeType);
|
|
if (!provider) {
|
|
throw new CourseMaterialError(
|
|
'extractor_not_configured',
|
|
422,
|
|
`No server-managed media extractor supports "${file.name}"`,
|
|
);
|
|
}
|
|
const artifact = await provider.extract({
|
|
buffer,
|
|
fileName: file.name,
|
|
fileSize: file.size,
|
|
mimeType,
|
|
config: managedExtractorConfig(provider.id),
|
|
});
|
|
const text = mediaArtifactText(artifact);
|
|
if (!text.trim()) {
|
|
throw new CourseMaterialError(
|
|
'material_parse_failed',
|
|
422,
|
|
`No usable transcript or synopsis was extracted from "${file.name}"`,
|
|
);
|
|
}
|
|
return { source, text, rawTextLength: text.length, images: [] };
|
|
}
|
|
|
|
const provider = managedDocumentProvider(mimeType);
|
|
if (!provider) {
|
|
throw new CourseMaterialError(
|
|
'extractor_not_configured',
|
|
422,
|
|
`No server-managed document extractor supports "${file.name}"`,
|
|
);
|
|
}
|
|
const artifact = await provider.extract({
|
|
buffer,
|
|
fileName: file.name,
|
|
fileSize: file.size,
|
|
mimeType,
|
|
config: managedExtractorConfig(provider.id),
|
|
});
|
|
const parsed = documentArtifactToParsedPdfContent(artifact);
|
|
if (!parsed.text.trim() && parsed.images.length === 0) {
|
|
throw new CourseMaterialError(
|
|
'material_parse_failed',
|
|
422,
|
|
`No usable content was extracted from "${file.name}"`,
|
|
);
|
|
}
|
|
const fallbackImages = parsed.images.map((src, imageIndex) => ({
|
|
id: `img_${imageIndex + 1}`,
|
|
src,
|
|
pageNumber: 0,
|
|
description: undefined,
|
|
width: undefined,
|
|
height: undefined,
|
|
}));
|
|
const images = (parsed.metadata?.pdfImages ?? fallbackImages).map((image) => ({
|
|
id: image.id,
|
|
src: image.src,
|
|
pageNumber: image.pageNumber,
|
|
description: image.description,
|
|
width: image.width,
|
|
height: image.height,
|
|
}));
|
|
return {
|
|
source,
|
|
text: parsed.text,
|
|
rawTextLength: parsed.text.length,
|
|
pageCount: parsed.metadata?.pageCount,
|
|
images,
|
|
};
|
|
}),
|
|
);
|
|
|
|
const bundle = buildDocumentBundle(parts);
|
|
return {
|
|
text: bundle.text,
|
|
images: bundle.images
|
|
.filter((image) => image.visionPriority > 0)
|
|
.sort((a, b) => b.visionPriority - a.visionPriority)
|
|
.map((image) => image.src),
|
|
};
|
|
}
|