InfoQ: Lock-free Algorithms [1h10m]

dgryski · 2012-05-25T17:18:17.798020+00:00

Most of the time I have little respect for computer science, but this is an area in which I always welcome advances.

dickeytk · 2012-05-25T16:17:37.059699+00:00

I am a developer, the network admins makes sure the servers I connect to are running.

yannisrizos · 2012-05-25T16:12:12.084282+00:00

Silverlight has to be tested in the browser unfortunately. It isn't supported by normal test runners.

And actually it was 14000 tests according to my notes. It wasn't a small application, we had three teams just on th UI.

uriel · 2012-05-25T07:53:21.563233+00:00

That is going to end badly, with people constantly arguing about when to use panic vs an error code. Like Java's checked vs unchecked exceptions, except with a much, much higher cost if you change your mind.

uriel · 2012-05-25T07:51:18.148822+00:00

It is just a game of semantics.

In C# we have "contextual keywords". Just like a "predeclared identifiers", they were added in such a way as to not break older code. But since they are called "keywords" they are automatically more "complex".

yannisrizos · 2012-05-25T04:22:44.358142+00:00

It all goes back to Java and a misunderstanding about what an "interface" is. Java developers, and later .NET developers, saw the interface keyword and mistakenly believed that's what people meant when they said "program to the interface, not the implementation".

They didn't understand that what people really meant was "use the damn API I gave you and stop using pointers to muck about with my data structure's internal memory".

unclemat · 2012-05-25T04:16:58.475807+00:00

Non-private code only needs to be tested via the public code. If your public API never passes a 0 to div(x, y) then you can usually assume that nothing will.

unclemat · 2012-05-25T04:15:21.148641+00:00

Sure, but how do you make that guarrantee?

For .NET that means marking those other classes as "internal" and moving the whole lot of them to a separate library. And I'm not necessarily against that.

yannisrizos · 2012-05-25T04:14:00.720859+00:00

Oh no, not at all. In a XAML-based application it is not uncommon to see all of these events fire when the FirstName property changes:

PropertyChanged: IsChanged
PropertyChanged: FirstName
PropertyChanged: FullName
PropertyChanged: HasErrors
ErrorsChanged: FullName

In WPF or Silverlight the built-in controls will be listening to these events and responding accordingly. So if an event isn't fired, your UI may become out of sync with the underlying data.

I did a write up about writing "comprehensive unit tests" recently that used FirstName as an example.

http://www.infoq.com/articles/Comprehensive-Unit-Test

yannisrizos · 2012-05-25T04:08:03.981316+00:00

DI works just as nicely with your separation architecture.

Sure, I love what they are doing with MEF 2 and ASP.NET MVC. Dependency Injection frameworks definitely have a place, but the assumption that you need DI for unit testing is very troubling to me.

So your not using model binding then?

It's psuedo-code, I didn't want the example to be platform specific.

yannisrizos · 2012-05-25T02:22:35.747583+00:00

Silverlight and .NET have different interfaces for even basic stuff like validation.

unclemat · 2012-05-24T23:46:01.205305+00:00

This reminds me of the CSLA framework. In their examples their classes require a specific private method that is only accessed via reflection. Took me a couple hours to figure that one out, and it was just a trivial demo app.

unclemat · 2012-05-24T23:43:46.527246+00:00

I have to disagree. I would rather have a large class with a small surface area than lots of little classes with a very large combined surface area.

unclemat · 2012-05-24T23:42:42.588455+00:00

However moving something to be private is not going to fix any real design problems.

No, but it will make fixing the design issues much, much easier. For me it is just step one of the process.

grauenwolf · 2012-05-24T21:44:54.513202+00:00

Performance. If the benchmarks are to be believed, you get a significant improvement without giving up your existing PHP-based assets.

dickeytk · 2012-05-24T21:40:18.611283+00:00

That's nonsense. With the right covering indexes and sufficient RAM, most of your queries can be against the cache.

dickeytk · 2012-05-24T20:11:45.983039+00:00

You aren't making any sense. If they know that disk I/O is going to be a limiting factor, why would they design the database to be I/O bound? They should be designing it to use less I/O, even if it means a larger CPU cost.

dickeytk · 2012-05-24T13:09:30.337373-07:00

Link please

dickeytk · 1 day

I/O is important therefore CPUs are not?

Yeah, I'm not buying it. And replication is always going to have a host of latency issues that I would rather not deal with.

dickeytk · 1 day

That is a rather bold accusation. Can you offer more details?

dickeytk · 1 day

Not really, I'm also trying to drive my Karma back down to 1.

sidcool1234 · 1 day

Pregenerate all of the HTML and serve it from CDNs.

dickeytk · 1 day

It's a database. I really couldn't care less what the OS is because nothing else should be on the same machine.

dickeytk · 1 day

You misunderstood. I am calling MySQL completely rotten and saying it makes PostgreSQL look bad to compare it to the former.

dickeytk · 1 day

No, but by some accounts they butchered it.

Six-Year Club	Best Link 2011-11-30
reddit mold	Verified Email

grauenwolf

TROPHY CASE

you'll need to login or register to do that

create a new account

login