summaryrefslogtreecommitdiff
path: root/ext/js/language/dictionary-importer.js
diff options
context:
space:
mode:
authorDarius Jahandarie <djahandarie@gmail.com>2023-11-09 13:30:31 +0000
committerGitHub <noreply@github.com>2023-11-09 13:30:31 +0000
commit5c5a167b4792af379cdacf633513cebf20728cd2 (patch)
tree5b6be3620a557d0b9177047003f6d742d9d2a32d /ext/js/language/dictionary-importer.js
parentb64f51c3b13a46af4dd7f1e43048ac19c781ca7b (diff)
parent0f4d36938fd0d844f548aa5a7f7e7842df8dfb41 (diff)
Merge pull request #307 from themoeway/modernize
Modernize codebase
Diffstat (limited to 'ext/js/language/dictionary-importer.js')
-rw-r--r--ext/js/language/dictionary-importer.js163
1 files changed, 66 insertions, 97 deletions
diff --git a/ext/js/language/dictionary-importer.js b/ext/js/language/dictionary-importer.js
index 718d9f1c..791d1a77 100644
--- a/ext/js/language/dictionary-importer.js
+++ b/ext/js/language/dictionary-importer.js
@@ -16,13 +16,11 @@
* along with this program. If not, see <https://www.gnu.org/licenses/>.
*/
-/* global
- * JSZip
- * JsonSchema
- * MediaUtil
- */
-
-class DictionaryImporter {
+import * as ajvSchemas from '../../lib/validate-schemas.js';
+import {BlobWriter, TextWriter, Uint8ArrayReader, ZipReader, configure} from '../../lib/zip.js';
+import {stringReverse} from '../core.js';
+import {MediaUtil} from '../media/media-util.js';
+export class DictionaryImporter {
constructor(mediaLoader, onProgress) {
this._mediaLoader = mediaLoader;
this._onProgress = typeof onProgress === 'function' ? onProgress : () => {};
@@ -39,20 +37,36 @@ class DictionaryImporter {
this._progressReset();
- // Read archive
- const archive = await JSZip.loadAsync(archiveContent);
+ configure({
+ workerScripts: {
+ deflate: ['../../lib/z-worker.js'],
+ inflate: ['../../lib/z-worker.js']
+ }
+ });
+ // Read archive
+ const zipFileReader = new Uint8ArrayReader(new Uint8Array(archiveContent));
+ const zipReader = new ZipReader(zipFileReader);
+ const zipEntries = await zipReader.getEntries();
+ const zipEntriesObject = {};
+ for (const entry of zipEntries) {
+ zipEntriesObject[entry.filename] = entry;
+ }
// Read and validate index
const indexFileName = 'index.json';
- const indexFile = archive.file(indexFileName);
+ const indexFile = zipEntriesObject[indexFileName];
if (!indexFile) {
throw new Error('No dictionary index found in archive');
}
- const index = JSON.parse(await indexFile.async('string'));
+ const indexContent = await indexFile.getData(
+ new TextWriter()
+ );
+ const index = JSON.parse(indexContent);
- const indexSchema = await this._getSchema('/data/schemas/dictionary-index-schema.json');
- this._validateJsonSchema(index, indexSchema, indexFileName);
+ if (!ajvSchemas.dictionaryIndex(index)) {
+ throw this._formatAjvSchemaError(ajvSchemas.dictionaryIndex, indexFileName);
+ }
const dictionaryTitle = index.title;
const version = index.format || index.version;
@@ -75,15 +89,14 @@ class DictionaryImporter {
// Load schemas
this._progressNextStep(0);
- const dataBankSchemaPaths = this._getDataBankSchemaPaths(version);
- const dataBankSchemas = await Promise.all(dataBankSchemaPaths.map((path) => this._getSchema(path)));
+ const dataBankSchemas = this._getDataBankSchemas(version);
// Files
- const termFiles = this._getArchiveFiles(archive, 'term_bank_?.json');
- const termMetaFiles = this._getArchiveFiles(archive, 'term_meta_bank_?.json');
- const kanjiFiles = this._getArchiveFiles(archive, 'kanji_bank_?.json');
- const kanjiMetaFiles = this._getArchiveFiles(archive, 'kanji_meta_bank_?.json');
- const tagFiles = this._getArchiveFiles(archive, 'tag_bank_?.json');
+ const termFiles = this._getArchiveFiles(zipEntriesObject, 'term_bank_?.json');
+ const termMetaFiles = this._getArchiveFiles(zipEntriesObject, 'term_meta_bank_?.json');
+ const kanjiFiles = this._getArchiveFiles(zipEntriesObject, 'kanji_bank_?.json');
+ const kanjiMetaFiles = this._getArchiveFiles(zipEntriesObject, 'kanji_meta_bank_?.json');
+ const tagFiles = this._getArchiveFiles(zipEntriesObject, 'tag_bank_?.json');
// Load data
this._progressNextStep(termFiles.length + termMetaFiles.length + kanjiFiles.length + kanjiMetaFiles.length + tagFiles.length);
@@ -124,7 +137,7 @@ class DictionaryImporter {
// Async requirements
this._progressNextStep(requirements.length);
- const {media} = await this._resolveAsyncRequirements(requirements, archive);
+ const {media} = await this._resolveAsyncRequirements(requirements, zipEntriesObject);
// Add dictionary descriptor
this._progressNextStep(termList.length + termMetaList.length + kanjiList.length + kanjiMetaList.length + tagList.length + media.length);
@@ -214,68 +227,27 @@ class DictionaryImporter {
return summary;
}
- async _getSchema(fileName) {
- const schema = await this._fetchJsonAsset(fileName);
- return new JsonSchema(schema);
- }
-
- _validateJsonSchema(value, schema, fileName) {
- try {
- schema.validate(value);
- } catch (e) {
- throw this._formatSchemaError(e, fileName);
- }
- }
-
- _formatSchemaError(e, fileName) {
- const valuePathString = this._getSchemaErrorPathString(e.valueStack, 'dictionary');
- const schemaPathString = this._getSchemaErrorPathString(e.schemaStack, 'schema');
-
- const e2 = new Error(`Dictionary has invalid data in '${fileName}' for value '${valuePathString}', validated against '${schemaPathString}': ${e.message}`);
- e2.data = e;
+ _formatAjvSchemaError(schema, fileName) {
+ const e2 = new Error(`Dictionary has invalid data in '${fileName}'`);
+ e2.data = schema.errors;
return e2;
}
- _getSchemaErrorPathString(infoList, base='') {
- let result = base;
- for (const {path} of infoList) {
- const pathArray = Array.isArray(path) ? path : [path];
- for (const pathPart of pathArray) {
- if (pathPart === null) {
- result = base;
- } else {
- switch (typeof pathPart) {
- case 'string':
- if (result.length > 0) {
- result += '.';
- }
- result += pathPart;
- break;
- case 'number':
- result += `[${pathPart}]`;
- break;
- }
- }
- }
- }
- return result;
- }
-
- _getDataBankSchemaPaths(version) {
+ _getDataBankSchemas(version) {
const termBank = (
version === 1 ?
- '/data/schemas/dictionary-term-bank-v1-schema.json' :
- '/data/schemas/dictionary-term-bank-v3-schema.json'
+ 'dictionaryTermBankV1' :
+ 'dictionaryTermBankV3'
);
- const termMetaBank = '/data/schemas/dictionary-term-meta-bank-v3-schema.json';
+ const termMetaBank = 'dictionaryTermMetaBankV3';
const kanjiBank = (
version === 1 ?
- '/data/schemas/dictionary-kanji-bank-v1-schema.json' :
- '/data/schemas/dictionary-kanji-bank-v3-schema.json'
+ 'dictionaryKanjiBankV1' :
+ 'dictionaryKanjiBankV3'
);
- const kanjiMetaBank = '/data/schemas/dictionary-kanji-meta-bank-v3-schema.json';
- const tagBank = '/data/schemas/dictionary-tag-bank-v3-schema.json';
+ const kanjiMetaBank = 'dictionaryKanjiMetaBankV3';
+ const tagBank = 'dictionaryTagBankV3';
return [termBank, termMetaBank, kanjiBank, kanjiMetaBank, tagBank];
}
@@ -335,9 +307,9 @@ class DictionaryImporter {
return target;
}
- async _resolveAsyncRequirements(requirements, archive) {
+ async _resolveAsyncRequirements(requirements, zipEntriesObject) {
const media = new Map();
- const context = {archive, media};
+ const context = {zipEntriesObject, media};
for (const requirement of requirements) {
await this._resolveAsyncRequirement(context, requirement);
@@ -427,13 +399,16 @@ class DictionaryImporter {
}
// Find file in archive
- const file = context.archive.file(path);
+ const file = context.zipEntriesObject[path];
if (file === null) {
throw createError('Could not find image');
}
// Load file content
- let content = await file.async('arraybuffer');
+ let content = await (await file.getData(
+ new BlobWriter()
+ )).arrayBuffer();
+
const mediaType = MediaUtil.getImageMediaTypeFromFileName(path);
if (mediaType === null) {
throw createError('Could not determine media type for image');
@@ -525,42 +500,36 @@ class DictionaryImporter {
}
}
- _getArchiveFiles(archive, fileNameFormat) {
+ _getArchiveFiles(zipEntriesObject, fileNameFormat) {
const indexPosition = fileNameFormat.indexOf('?');
const prefix = fileNameFormat.substring(0, indexPosition);
const suffix = fileNameFormat.substring(indexPosition + 1);
const results = [];
- for (let i = 1; true; ++i) {
- const fileName = `${prefix}${i}${suffix}`;
- const file = archive.file(fileName);
- if (!file) { break; }
- results.push(file);
+ for (const f of Object.keys(zipEntriesObject)) {
+ if (f.startsWith(prefix) && f.endsWith(suffix)) {
+ results.push(zipEntriesObject[f]);
+ }
}
return results;
}
- async _readFileSequence(files, convertEntry, schema, dictionaryTitle) {
+ async _readFileSequence(files, convertEntry, schemaName, dictionaryTitle) {
const progressData = this._progressData;
- let count = 0;
let startIndex = 0;
- if (typeof this._onProgress === 'function') {
- schema.progressInterval = 1000;
- schema.progress = (s) => {
- const index = s.getValueStackLength() > 1 ? s.getValueStackItem(1).path : 0;
- progressData.index = startIndex + (index / count);
- this._progress();
- };
- }
const results = [];
- for (const file of files) {
- const entries = JSON.parse(await file.async('string'));
+ for (const fileName of Object.keys(files)) {
+ const content = await files[fileName].getData(
+ new TextWriter()
+ );
+ const entries = JSON.parse(content);
- count = Array.isArray(entries) ? Math.max(entries.length, 1) : 1;
startIndex = progressData.index;
this._progress();
- this._validateJsonSchema(entries, schema, file.name);
+ if (!ajvSchemas[schemaName](entries)) {
+ throw this._formatAjvSchemaError(ajvSchemas[schemaName], fileName);
+ }
progressData.index = startIndex + 1;
this._progress();