News for hackers. Stuff that matters.

A comment on Google: 90% of our engineers use the software you wrote (Homebrew), but.... Parent comment

Re:Google: 90% of our engineers use the software you wrote (... by vezzy-fnord on Wednesday June 10, 2015 @05:35PM (#9695645)
In fairness, package managers can get surprisingly complicated from a CS aspect. openSUSE wrote an elaborate reusable SAT solver library just for the dependency resolution component. Then let's not ignore that a package manager at its core must deal with the finicky nature of contemporary dynamic linking, the OS environment and all the baggage that comes with heterogenous library contexts and namespaces, maintaining transaction consistency, preferably ensuring atomic operations and so on.

That said, I don't think Homebrew is one of these. Always seemed like a quick fix for an OS X deficiency to me, and its competitors MacPorts and Fink being mediocre themselves certainly helped with its popularity. Not to rain on the achievements of the developers, but still.

[ Parent | on HN ]
Re:Google: 90% of our engineers use the software you wrote (... by Alupis on Wednesday June 10, 2015 @05:38PM (#9695673)
To further your point, Package Managers have been done... and have been getting done for a long time.

Blazing the trail in machine learning, ai, etc... that's new territory where there are no knowledge bases or places you can go to source information about how to do it.

That's a different ballgame, and requires a scientist, not an engineer/developer.

[ Parent | on HN ]
Re:Google: 90% of our engineers use the software you wrote (... by vezzy-fnord on Wednesday June 10, 2015 @05:48PM (#9695755)
Package managers have been done, but by and large not as well as they could be. It's unsurprising that Nix originates from an academic background. It's actually complicated work, and not really a solved problem yet.

Moreover, I'd heavily dispute that there are "no knowledge bases" for machine learning and AI. There very much is, given that the fields have a history as long as most contemporary computing. "Machine learning" in essence reflects the transition from AI being symbolic to statistical, so there's plenty of reference material on the latter and the various intersectionary disciplines like computational linguistics, data mining, computer vision and so forth. Hell, OpenCV - there's some good source information, for instance.

[ Parent | on HN ]