Having a single entry point to a website. Bad? Goo

问题:

This question stems from watching Rasmus Lerdorf's talk from Drupalcon. This question and his talk have nothing specifically to do with Drupal, by the way... it was just given at their con. My own question also has nothing specific to do with PHP. It is the single entry point in general that I am curious about.

These days it seems that most frameworks offer a single entry point for whatever you build with them. In his talk Rasmus mentions that he thinks this is bad. It seems to me that he would be correct in this thinking. If everyone hitting the site is coming in through the same entry point wouldn't things bog down after traffic reached a certain point? Wouldn't it be more efficient to allow people direct access to specific points in a site without having their request go through the same point? But perhaps the actual impact is not very bad? Maybe modern architecture can handle it? Maybe you have to be truly gigantic in scale before it becomes even worth considering? I'm curious as to what people on this site think about this issue.

回答1:

In short, Rasmus or the interpretation is wrong.

This shows a clear lack of understanding how computers work. The more something gets used, the more likely it's closer to the CPU, and therefore faster. Mind you, a single point of entry != single point of failure. But that's all beside the point, when people say single point of entry, we're talking about the app, it is a single point of entry for your logic.

Not to mention it's architecturally brain-dead not to have a central point of entry, or reduce the number of entries points in general. As soon as you want to do one thing across your app at every entry point, guess how many places need to change? Having dealt with an app that each page stood on it's own, it sucked having to change, and I assure you, we needed it.

回答2:

The important thing is that you use a web framework that supports scalability through methods like load-balancing and declarative code.

No, a single-entry point does not in itself make a bottleneck. The front page of Google gets a lot of hits, but they have lots of servers.

So the answer is: It doesn't matter.

回答3:

Like anything in software development, it depends. Rasmus's objection to the front-controller style frameworks is the performance hit you take from having to load so much code up-front on each request. This is 100% true. Even if you're using a smart-resource loading module/object/etc of some kind, using a framework is a performance trade-off. You take the performance hit, but in return you get back

1. Encouraged seperation of "business logic" (whatever that is) and Template/Layout Logic

2. Instant and (more importantly) unified access to the objects you'll use for database queries, service called, your data model, etc.

To a guy like Rasmus, this isn't worth the performance hit. He's a C/C++ programmer. For him, if you want to separate out business logic in a highly performant way, you write a C/C++ Extension to PHP.

So if you have an environment and team where you can easily write C/C++ extensions to PHP and your time-to-market vs. performance ratio is acceptable, then yes, throw away your front-controller framework.

If that's not your environment, consider the productivity increases a front-controller framework can bring to your (likely) simple CRUD Application.

回答4:

I think one of the biggest advantages you have over having only a single point of entry is security. All of the input going in is less likely to corrupt the system if it is checked and validated in a single place.

回答5:

I think it's a big misunderstanding discussing this from the point of "one file" vs. "several files".

One tends to think that because the entry point is in a single file, then all we have to focus on is the code in that one file - that's wrong.

All of the popular frameworks has tons of files with entry manipulation code, interpretation code, and validation code for requests. The code is not located in one place, rather is spread around in a jungle of require/include statement calling different classes depending on whats being requested and how.

In both cases the request is really handled by different files.

Why then should I have a single entry point with some kind of _detect_uri() function that needs to call several other functions like strpos(), substr(), strncmp() to clean up, validate, and split up the request string when I can just use several entry points eliminating that code all together?

Take a look at CodeIgniters _detect_uri() function in URI.php. Not to pick on CodeIgniter, it's just an example. The other frameworks does it likewise.

You can achieve the goals of a MVC pattern with a single entry point as well as with several entry points.

回答6:

This is what I thought at first, but it doesn't seem to be an impact. After all, your entry point is (usually) only doing a couple of things: setting some environment constants, including the bootstrap loader, optionally catching any exceptions thrown and dispatching the front controller. I think the reason that this is not inefficient is because this file does not change depending on the controller, action or even user.

I do feel this is odd however. I'm building a small MVC framework myself at the moment but it's slightly reverse to most frameworks I've used. I put controller logic in the accessed file. For example, index.php would contain the IndexController and it's actions. This approach is working well for me at least.

回答7:

As most of the php mvc frameworks use some sort of url rewriting, or at least parse anything after index.php at their own, a single entry point is needed.

Besides that, i like to provide entry points per context, say web(/soap)/console/...

回答8:

Just to add, the thing people usually think is that, since there is one php page, it is a single page serving all requests. Sort of like queuing.

The important thing to note is that, each request creates an instance of the script and thus, the load is the same as if two different pages were being accessed at the same time. So, still the same load. Virtually.

However, some frameworks might have a hell whole lot of stuff you don't need going on in the entry script. Sort of like a catchall to satisfy all possible scenarios, and that might cause load.

Personally, I prefer multi pages, the same way I mix oop and procedural. I like php the old school way.

回答9:

There are definitely disadvantages to using a front-controller file architecture. As Saem explains, accessing the same file is generally advantageous to performance, but as Alan Storm and Rasmus Lerdorf explain, trying to make code handle numerous situations (such as a front-controller tries to handle the request to every "page" on the site) leads to a bloated and complicated collection of code. Using a bloated and complicated collection of code for every page load can be considered bad practice.

Saem is arguing that a front-controller can save you from having to edit code in multiple locations, but I think that in some cases, that is better solved by separating out redundancies.

Having a directory structure that is actually used by the web server can make things simpler for the web server and more obvious for the developers. It is simpler for a web server to serve files by translating the url into the location as they would traditionally represent than to rewrite them into a new URL. A non-front-controller file architecture can be more obvious to the developer because instead of starting with a controller that is bloated to handle everything, they start with a file/controller that is more specific to what they need to develop.