Dustin Sallings

I've apparently completely forgotten about this. I think it would be fine since it seems to be useful for people.

Is there a test case that works reasonably here? I updated the commit to build properly, but my test data + no transform = doesn't seem to do anything.

I like this idea, but it changes the API of Get and decreases the coverage while adding new, uncovered functions. Can you include some tests? I'm at 100% test coverage...

Possibly, but in an additive way. Can you rework the set and come up with the smallest set of patches that are distinct? e.g., add the code and the tests...

Did you push the squashed commits? This is what I pulled: 4ca989e Removed t.Fail() calls that come after t.Errorf(). 6ac1350 fixes based on pull-request comments. a3dfb09 100% coverage. 16b8d33 optimized...

I don't think it needs to be one. Each commit needs to make sense on its own, though. I think your mutation operations with their tests can be in one...

Then you'd also need specialized code for Indian commas which starts getting weird at lakh and then gets even weirder at maha-sankh just because why not (unless you're using the...

The problem seems to be `next/textproto`'s `MIMEHeader` type. Not sure why it was a map, but it's not surprising [nobody can figure out how to handle headers](https://stackoverflow.com/questions/4580548/how-does-one-parse-http-headers-with-libcurl).

You mean like "%.2f%%" ? I'm not sure what we could do there that wouldn't be worse than what's built-in.

Perhaps it would be good to understand why go only rolls up to the nearest hour. I suppose if we're approximating, we can throw away precision and be fairly loose...