Added an "approximate_rgb" function.

[exhuma]

Also covered 256 colour codes with test. Not 100% sure if those test are
any good though.

[fwenzel]

This would fix #14, I presume.

[exhuma]

Hmm... just saw the failed tests. Seems it's working on Python 2.7 but not on 2.5 and 2.6. I will need to look into that.

[exhuma]

Tried to install Python 2.5 and 2.6 from source, but I am having trouble making virtualenv work with them... I am giving up for now due to time-constraints :(

If I don't forget it, I will look it up later again.

[erikrose]

I'm fully in favor of a nearest-possible-color fallback in case of limited-color terminals, in case it helps.

[jquast]

I designed a draft proposal in issues #67, please review. It would likely make some use of the given code.

I am going to provide some criticism of this PR, please consider it constructive!

• the comments do not explain the numerical adjustments made.
• the only seemingly useful comment is really just a complaint. https://docs.python.org/devguide/documenting.html#affirmative-tone
• People will study Blessings' internals to re-write analogous libraries in other languages. Blessings should continue to not only be a useful API, but a depth of code that others may learn and implement from. This particular function is very unforgiving in describing the implementation.
• as such, i don't think anybody is qualified to "sign off" on this kind of PR. Speaking for myself, I certainly wouldn't have confidence that it is correct.
• the tests do not provide any assurance that the color space was mapped correctly, to clarify: this could just as easily been created as "I ran it, i got these values, so I reversed it into a test".

So,

The rules of why and how these values are clamped/divided/reduced, then matched along a row/col/face, then some magic about indexes should be declared as a code comment in the implementation: so there is no need to express why such values are matched in the test: it should be desk-checkable by the given description of the implementation.

I think I see three functions to be compoundly tested:

• reducing the 8x8x8-bit colorspace to 6x6x6-bit.
• finding its face/col/row.
• given this face/col/row, finding its range in 0-256.

Currently this is all done in a single function. My proposal in #67 also hopes to make use of variants of these 3 functions to map to a 24-bit colorspace as well, which would require conditionally excluding some of these components.

[exhuma]

I fully agree with your comments. It has been a while since I looked into this code, I don't even remember writing this, and it does not even look like something I had written. Not sure where even it comes from. Right now, I don't get too much time to do some research in this, but I will try to get back to that code as soon as I can and clean things up. I hope I will get to it soon.

[ssbarnea]

Do we have any change of making this a reality?

[jquast]

@ssbarnea this is rolling into next release of blessed https://github.com/jquast/blessed/blob/d7c3d52f8804d92cfb35a8286cd88fd46e633fc9/blessed/terminal.py#L670

[exhuma]

I've stumbled across this issue again. It dropped off my radar again. I will deal with the conflicts and will change the code commend to something objective.

[exhuma]

Done. Should be better now.

[exhuma]

I've also hunted down the original implementation and added a link (for attribution).