Badly performing regexp when aligning trailing comments after long line

[papandreou]

Hi! I've run into a case where escodegen never seems to complete the formatting of a program. It happens with a combination of a very long line (a minified JavaScript bundle) that has two trailing comments.

This reproduces it:

console.log(require('./escodegen').generate({
    type: 'Program',
    body: [
        {
            type: 'ExpressionStatement',
            expression: {
                type: 'Literal',
                value: 'a'.repeat(1000000)
            },
            trailingComments: [
                {
                    type: 'Line',
                    value: '# sourceMappingURL=foo.map'
                },
                {
                    value: '# sourceMappingURL=bar.map',
                    type: 'Line'
                }
            ]
        }
    ]
}, {
    format: { newline: '' },
    comment: true
}));

It seems like escodegen tries to align the two comments by putting the second one on a new line with 1000004 leading spaces, and then fragment.replace(/\s+$/, '') in this line takes a very long time due to exponential performance characteristics.

I discovered it because of a bug in assetgraph that caused those two sourceMappingURL comments to be there under very specific circumstances. I've fixed the bug now, but still thought I'd let you know :)

An easy fix might be to disable the indentation of the subsequent trailing line comments when the first line is outrageously long?