Regarding "Dynamic Source Maps"

Hello, I'm Nick Fitzgerald. This is my weblog. You can also check out my shared items from Reader, moments on Twitter, and code on GitHub. Feel free to contact me about whatever.

Regarding "Dynamic Source Maps"

January 22nd, 2013

Peter van der Zee recently wrote about source maps; in particular, he detailed the ways that he has found them lacking and proposes changes to the source map spec that he believes would improve them to support his use cases.

What Peter wants is to generate and consume source maps at runtime in the browser, and develop without an extra compilation and refresh step. When you are editing and compiling sources inside of the browser, rather than sources on a file server, you can't point to a remote URL where the original source (the source before the compilation step) is located. This is problematic because source URLs are exactly what source maps require, as defined by the spec.

Peter then outlines proposed changes to the source map spec to support his need for "dynamic" source maps. First, adding a new comment pragma //@ sourceMapping=<source map JSON blob>. Second, to include the full original source(s) inside the source map rather than including URLs pointing to the location of the original source(s).

Peter is a very intelligent person, and I have been following his blog for more than two years (you should follow, too), but I believe that he jumped the gun by leaping from his use cases to proposing changes to the source map spec. The one doesn't lead to the other because his use case is already supported by the current specification as-is.

How? Data URLs. Encode the original source(s) as data URLs in the source map's sources list, and encode the whole source map as a data URL in the //@ sourceMappingURL comment pragma appended to the generated source. By doing this, both the whole source map and complete original source(s) are attached directly to the generated source. This accomplishes the same goals that Peter's solution does: you can dynamically generate source maps inside your browser environment (presumably with the mozilla/source-map library) and all of the resources can exist inside your development environment instead of on a remote server.

The benefits to using data URLs are:

No changes to the source map spec. No waiting on debates, or for the spec authors to come to an agreement. Furthermore, existing tooling around source maps will continue to work, including integration with browsers' built in debuggers.
No need for yet another comment pragma. We already have //@ sourceMappingURL, //@ sourceURL, and //@line. Some are only supported by some tools, and some browsers; only one is part of a standard; two either are or should be deprecated. This is confusing enough already, and once we add another comment pragma it will be there forever.
It is optional to include the full text of the original source(s) in the source map. Developers who are using a more "traditional" development cycle will not be negatively affected by having a larger source map to download.
Encoding and decoding data URLs is simple. Browser environments already have window.btoa and window.atob to encode and decode base 64 respectively, which makes integration with whatever tooling you are using that much easier.