search

SergioBenitez / Figment

A hierarchical configuration library so con-free, it's unreal.
Apache License 2.0
604 stars 35 forks source link

Flexible inputs, second attempt #100

Closed nmathewson closed 5 months ago

nmathewson commented 5 months ago

This is a second attempt to implement the idea of #97.

Instead of changing the default behavior of impl Deserialize for *, it instead provides new extract_lossy and extract_inner_lossy methods that implement the lossy behavior across the entire figment object.

(As an alternative, as discussed on #97, the user might annotate all their booleans and numbers with deserialize_with. But it can be error-prone to do so, especially if the configuration structure is large and complex.)

ijackson commented 5 months ago

Hi. I'm one of @nmathewson's colleagues. I mostly just wanted to say thanks to @SergioBenitez for engaging with this, to try to help meet our needs. @nmathewson, I'd be happy to help here. Eg, would you like me to read the code in this MR?

I did have one question. Why are we calling these methods "lossy"? I hope they aren't really lossy. But I just came in here so maybe I am confused...

SergioBenitez commented 5 months ago

I really like this idea, and I'm committed to reviewing/merging this as soon as I free up a bit (in the next few days)!

@ijackson Thank you! I chose "lossy" because we lose the true type information in the conversion process. In this comment I show how this is truly a loss and not just a conversion.

nmathewson commented 5 months ago

@ijackson Thank you! I chose "lossy" because we lose the true type information in the conversion process. In https://github.com/SergioBenitez/Figment/pull/97#issuecomment-2030957484 I show how this is truly a loss and not just a conversion.

I do agree that when we consider these conversions as applied to individual strings or numbers, they are lossy: When you convert "yes" to true, you can't recover the original "yes" if it turns out that what you really wanted was a string.

But I wonder whether "lossy" is the right word for the overall feature when applied to an entire configuration in extract_lossy. When we convert "yes" to true in the context of a boolean option, we aren't losing any information, any more than you we are when we treat 1.0 and 1.00 as the same float.

(Edited to add: That said, having this option is more important to me than having the perfect name for this option. :) )

SergioBenitez commented 5 months ago

Published in 0.10.18! Thanks for working with me on this. Fantastic work!