Comment by PaulHoule | Hacker News Reader

Comment by PaulHoule | Hacker News Reader

Architecture, in many ways, has trumped algorithms.

Back when I was a 10 year old kid, if you wanted to sort something on a microcomputer you had to write your own sort in BASIC or assembler, or if you were really lucky, FORTH.

Today if you want to sort something you can use UNIX sort or the sort built into languages like Java/PHP/C#/Ruby or you can use ORDER BY in SQL.

The increasing trend is that complex software systems encapsulate algorithms. Declarative languages like SQL are the ultimate -- a good relational database has a collection of internal memory, external memory and possibly parallel algorithms for which it will find the best choice for your particular query and data.

I'm skeptical of the "baby steps" approach that people have advocated where you do some busywork to figure out how to re-implement the addition operator and such. That's how computer science works, but CS education has a terrible batting average at producing effective developers. I can't stand hanging out on proggit these days because so many people are stuck on sorting algorithms and obsolete content from Knuth's ACP.

Personally I look at the algorithms literature the way a Somali pirate looks at a shipping lane. I look specifically for advanced algorithms that solve my problem and that I can't find good implementations of. It can take some time to wrap my head around, say, binary space partitioning or suffix trees, but at least I'm motivated.

I'm always thinking about the choice I have between implementing something new and reuse.

>CS education has a terrible batting average at producing effective developers.

That's likely because "producing effective developers" is not a focus of CS education. It's to make you understand fundamentals of computation, such as complexity and algorithms, as well as give a deeper understanding of computers both on a technical as well as a theoretical level.

These are all skills that can aid in becoming a great developer, but they are only supplementary. I'd say that a better course of education to take if one plans to become a developer is some form of software engineering, but not computer science.

As a Dev and CS major I can tell you that schools have gotten to be pretty good about teaching more than just the difference between merge sort and quick sort. My school offers things like Databases, OS, and even a web dev class this semester. Schools do so much to get kids to become good developers along side good programmers.

The problem is that the kids aren't trying to become good developers. How many college students in a class have a github (or some online code repo) that is publicly view-able? How many have a stackoverflow account and are at least somewhat active on it. What percentage of kids in a class will code a program outside of class just because they want to code something cool.

You seem to be implying that a publicly-viewable github repo or active stackoverflow account are necessary conditions for coding outside of class.

He gave those examples as common qualities of CS students who go above and beyond school work.

That's fair. I suppose I just wanted to point out that they are by no means required. There are plenty of people who go above and beyond and yet eschew the more "social" aspects of coding.

All these encapsulated algorithms are really useful, until you need something slightly different and hit all kinds of leaky abstractions. Declarative languages are especially bad at this, since you have to sacrifice a chicken and wait for the right phase of the moon to get the automatic optimizer to produce the code that you actually indent to run.

This is the original article that coined "leaky abstractions". http://www.joelonsoftware.com/articles/LeakyAbstractions.htm...

It's a tradeoff.

I think reuse wins in the long term because there's a competitive market -- a thousand flowers bloom and the better things float to the top.

SQL, for instance, has benefitted from implementations that vary from Microsoft Access all the way to Oracle. Because of fierce competition, vendors (even open source projects) have had to raise their game.

The same is true for Java frameworks -- if you looked at frameworks like Struts back in 2002 one could come to the conclusion that frameworks were part of the problem, not the solution. Today you can find frameworks that fit your tastes, whatever they are.

Reply on news.ycombinator.com