Babel sourcemaps rejected as invalid

[dwickern]

When babel sourcemaps are enabled from ember-cli-babel, you still won't have es6 sourcemaps in the ember dist.

The sourcemaps are being rejected by sourcemap-validator. I wrote up the details over at ben-ng/sourcemap-validator#8

It seems the babel generated sourcemaps are technically invalid, but they still work fine in the browser, and other frameworks using babel don't seem to have this issue. Maybe consider getting rid of the sourcemap validation?

/cc @rwjblue @stefanpenner

[rwjblue]

Hehe, we should likely try to fix the babel source maps so they are correct and valid. A little while back @chadhietala created babel/babel#6008 on the subject...

[stefanpenner]

It seems the babel generated sourcemaps are technically invalid, but they still work fine in the browser,

They don't actually work though. And can cause other downstream maps issues.

[gdborton]

Would it be possible to add an option to skip validation? Until babel/transformation tooling is fixed? Using poor sourcemaps are better than opting to use no sourcemaps.

[stefanpenner]

Using poor sourcemaps are better than opting to use no sourcemaps.

Are you sure about that? Having source-maps where the debugger breakpoints don't work, variables are totally un-inspectable.. etc. Also remember, when this occurs downstream source-maps also become broken. So source-maps that are actually valid, become affected.

I personally don't think this is acceptable, and don't look forward to the inevitable bug reports/emo people will have when they force this on, and end up with random other failures.

[gdborton]

@stefanpenner So I've been giving this a bit more thought, and there are reasons other than just babel's source maps for providing this option.

In my case transforms are applied to individual files, then cached, then in a later process transpiled code and maps are used to generate a bundle. For optimal performance it would make the most sense to validator the individual source maps just before caching and mark them with a flag. In this case we wouldn't want to rerun the validator.

[stefanpenner]

Caching the validator result seems like a good idea.

In theory, one could write an efficient source-map decoder that could validate as it encodes.. That is likely much more work, but likely also faster/better across the board.