149 lines
4.6 KiB
TypeScript
149 lines
4.6 KiB
TypeScript
|
import { Extractor } from '../utils/extractor';
|
||
|
import { CliOptionsInterface } from './cli-options.interface';
|
||
|
import { TranslationCollection } from '../utils/translation.collection';
|
||
|
import { ParserInterface } from '../parsers/parser.interface';
|
||
|
import { PipeParser } from '../parsers/pipe.parser';
|
||
|
import { DirectiveParser } from '../parsers/directive.parser';
|
||
|
import { ServiceParser } from '../parsers/service.parser';
|
||
|
import { CompilerInterface } from '../compilers/compiler.interface';
|
||
|
import { JsonCompiler } from '../compilers/json.compiler';
|
||
|
import { NamespacedJsonCompiler } from '../compilers/namespaced-json.compiler';
|
||
|
import { PoCompiler } from '../compilers/po.compiler';
|
||
|
|
||
|
import * as fs from 'fs';
|
||
|
import * as path from 'path';
|
||
|
import * as mkdirp from 'mkdirp';
|
||
|
import * as chalk from 'chalk';
|
||
|
import * as yargs from 'yargs';
|
||
|
|
||
|
const options: CliOptionsInterface = yargs
|
||
|
.usage('Extract strings from files for translation.\nUsage: $0 [options]')
|
||
|
.help('help')
|
||
|
.option('dir', {
|
||
|
alias: 'd',
|
||
|
describe: 'Paths you would like to extract strings from. Multiple paths can be specified',
|
||
|
default: process.env.PWD,
|
||
|
type: 'array'
|
||
|
})
|
||
|
.option('output', {
|
||
|
alias: 'o',
|
||
|
describe: 'Path you would like to save extracted strings to. Multiple paths can be specified',
|
||
|
default: process.env.PWD,
|
||
|
type: 'array'
|
||
|
})
|
||
|
.option('format', {
|
||
|
alias: 'f',
|
||
|
describe: 'Output format',
|
||
|
default: 'json',
|
||
|
type: 'string',
|
||
|
choices: ['json', 'namespaced-json', 'pot']
|
||
|
})
|
||
|
.option('replace', {
|
||
|
alias: 'r',
|
||
|
describe: 'Replace the contents of output file if it exists (Merges by default)',
|
||
|
default: false,
|
||
|
type: 'boolean'
|
||
|
})
|
||
|
.option('sort', {
|
||
|
alias: 's',
|
||
|
describe: 'Sort translations in the output file in alphabetical order',
|
||
|
default: false,
|
||
|
type: 'boolean'
|
||
|
})
|
||
|
.option('clean', {
|
||
|
alias: 'c',
|
||
|
describe: 'Remove obsolete strings when merging',
|
||
|
default: false,
|
||
|
type: 'boolean'
|
||
|
})
|
||
|
.argv;
|
||
|
|
||
|
const patterns: string[] = [
|
||
|
'/**/*.html',
|
||
|
'/**/*.ts'
|
||
|
];
|
||
|
const parsers: ParserInterface[] = [
|
||
|
new ServiceParser(),
|
||
|
new PipeParser(),
|
||
|
new DirectiveParser()
|
||
|
];
|
||
|
|
||
|
let compiler: CompilerInterface;
|
||
|
let ext: string;
|
||
|
switch (options.format) {
|
||
|
case 'pot':
|
||
|
compiler = new PoCompiler();
|
||
|
ext = 'pot';
|
||
|
break;
|
||
|
case 'json':
|
||
|
compiler = new JsonCompiler();
|
||
|
ext = 'json';
|
||
|
break;
|
||
|
case 'namespaced-json':
|
||
|
compiler = new NamespacedJsonCompiler();
|
||
|
ext = 'json';
|
||
|
break;
|
||
|
}
|
||
|
|
||
|
const extractor: Extractor = new Extractor(parsers, patterns);
|
||
|
|
||
|
let extractedStrings: TranslationCollection = new TranslationCollection();
|
||
|
|
||
|
// Extract strings from paths
|
||
|
console.log(chalk.bold('Extracting strings from...'));
|
||
|
options.dir.forEach(dir => {
|
||
|
const normalizedDir: string = path.resolve(dir);
|
||
|
if (!fs.existsSync(dir) || !fs.statSync(dir).isDirectory()) {
|
||
|
console.log(`The path you supplied was not found: '${dir}'`);
|
||
|
process.exit(1);
|
||
|
}
|
||
|
|
||
|
console.log(chalk.gray('- %s'), normalizedDir);
|
||
|
extractedStrings = extractedStrings.union(extractor.process(normalizedDir));
|
||
|
});
|
||
|
console.log(chalk.green('Extracted %d strings\n'), extractedStrings.count());
|
||
|
|
||
|
// Save extracted strings to output paths
|
||
|
options.output.forEach(output => {
|
||
|
const normalizedOutput: string = path.resolve(output);
|
||
|
|
||
|
let outputDir: string = normalizedOutput;
|
||
|
let outputFilename: string = `template.${ext}`;
|
||
|
if (!fs.existsSync(normalizedOutput) || !fs.statSync(normalizedOutput).isDirectory()) {
|
||
|
outputDir = path.dirname(normalizedOutput);
|
||
|
outputFilename = path.basename(normalizedOutput);
|
||
|
}
|
||
|
const outputPath: string = path.join(outputDir, outputFilename);
|
||
|
|
||
|
console.log(chalk.bold('Saving to: %s'), outputPath);
|
||
|
if (!fs.existsSync(outputDir)) {
|
||
|
console.log(chalk.dim('- Created output dir: %s'), outputDir);
|
||
|
mkdirp.sync(outputDir);
|
||
|
}
|
||
|
|
||
|
let processedStrings: TranslationCollection = extractedStrings;
|
||
|
|
||
|
if (fs.existsSync(outputPath) && !options.replace) {
|
||
|
const existingStrings: TranslationCollection = compiler.parse(fs.readFileSync(outputPath, 'utf-8'));
|
||
|
if (existingStrings.count() > 0) {
|
||
|
processedStrings = processedStrings.union(existingStrings);
|
||
|
console.log(chalk.dim('- Merged with %d existing strings'), existingStrings.count());
|
||
|
}
|
||
|
|
||
|
if (options.clean) {
|
||
|
const collectionCount = processedStrings.count();
|
||
|
processedStrings = processedStrings.intersect(processedStrings);
|
||
|
const removeCount = collectionCount - processedStrings.count();
|
||
|
console.log(chalk.dim('- Removed %d obsolete strings'), removeCount);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
if (options.sort) {
|
||
|
processedStrings = processedStrings.sort();
|
||
|
console.log(chalk.dim('- Sorted strings'));
|
||
|
}
|
||
|
|
||
|
fs.writeFileSync(outputPath, compiler.compile(processedStrings));
|
||
|
console.log(chalk.green('OK!\n'));
|
||
|
});
|