why are some frameworks slow?
for example how comes that zend frameworks are slower than code igniter? if you don´t use a lot of classes, just chose the classes of your needs, then it shouldn't get slow?
are you forced to use a lot of classes with zend framework or are they preloading a lot of classes automatically?
It depends how much they have taken care of the speed when writing the code of the framework. CodeIgniter has been written amazingly always keeping the speed in mind. Also big frameworks autoload a lot of nuts and bolts making the framework slower. However, if you are experienced with any of these frameworks, you can speed up the framework to a good deal by filtering out the stuff that you think is not necessary for your current project.
Zend_Loader_Autoloader automatically loads classes when they are requested, so there are no "unneeded classes". There are some tutorials on how to speed up Zend framework powered applications, ie. commenting out "require_once's" and using the PluginLoader include file cache explained here (along with some other methods) http://framework.zend.com/manual/en/performance.html
There are some things you can't beat.
Any framework can be
Simple
Extensible
Fast
Understandable
...
But you can't have all at once. It can't be fast AND extensible. If it is extensible, it has to check if you want to use the internal class or your own. It has to define interfaces for everything. Therefore it has to load a lot of files (for classes and interfaces, abstracts, ...). Note the difference between loading class MySQLDB class (one class for single purpose) and DbLayer_Interface, DbLayer_Abstract, DbLayer_MySQL (complete infrastructure, where you can substitute any part of it).
Also with easy comes with some magic. The more the framework does for you, the more "magic" is going on under the hood. THe magic can be fast, as it's done for one purpose and doesn't have to check if you want to change anything.
That's why ZF is slower then CI and others. Any class can be replaced. For example we extended and injected custom Db_Select class, that handles multilingual column names automatically. You make select like SELECT table.name_en FROM table and the underlaying logic transforms it to SELECT table.name_en, table.name_fr, table.name_de FROM table on the fly... That's the power you receive in trade for speed ;)
Related
What I'm looking for is a way to remove the model from a set of PHP files that make up a website. It's difficult (for me) to explain.
By models I mean models in an MVC sense.
As an example say I have this website:
index.php
about.php
shop.php
checkout.php
All of the above PHP files use the same database. I have separated the views by adding templates using a view.php file that renders the correct template with values passed to it.
I am not looking to use a framework that's already out there. I'm looking at writing my own in some senses, with only the bits I need to use in it.
If anyone would like to explain why this is not necessary, or a better way of doing things, then I'm open to that too.
Thanks in advance.
Writing you own MVC framework will take time, but you will learn a lot in the process. So, if you have the time/resources to do it I definitely encourage you to do so.
In this context here are some small pieces of advise that may help you:
Create your domain model first. I'm assuming that you are going in the OO way, so think about your domain problem and create the abstractions that best represent your problem. Try to keep it decoupled from cross-cutting concerns, like persistence.
Test a lot, test often. Try to test (and run your tests) as you create your domain model. This will be specially valuable when in 6 months you add a new feature and want to make sure that you haven't break anything. If you can separate your domain model from anything external (like the persistence layer or third party web services) the testing it is going to be a lot simpler. Today PHPUnit is pretty much the de-facto standard for unit testing in PHP.
You don't have to write everything from scratch. There are a lot of libraries that can help you to ease the development of an MVC framework, so that you can concentrate on what you really want to develop. For example, you could use Slim to handle the page routing or you could delegate the persistence stuff to Doctrine 2.
It is always nice to analyze how other frameworks solve things. You may want to look at products like Symfony or Kohana or even check how Elgg handles its views system. Also, if you want to check out something radically different you can take a look at Seaside's architecture.
Coming back to your original question, for me the key is to keep things from different layers as decoupled as possible. While I have only used the version 1, Doctrine 2 seems like a good candidate for persistence, since it allows you to create a domain model that is quite independent from the DB. This is a huge step. The second thing is how handle the view system. This is quite developer-taste dependent. For example, I like to model everything with objects, so I like Seaside's approach. On the other hand, Elgg's way of handling views is quite nice and maybe fits better with the way things are handled in PHP. Here is when you may benefit on doing some research before deciding on a route to go.
HTH
As someone who has written his own PHP framework, and with the same sensibility as yours, I can tell you that using a framework is a fine thing to do. That said, start by writing your own - you'll gain greater appreciation for the true structure and utility of a framework.
You'll want to learn about the Singleton object pattern. It is a major differentiator in the kinds of objects you can develop in your framework.
When you have written a few models that your files/controllers (presuming MVC) include, you will begin to see where to abstract a 'base mode' from which others extend (hint: the DB singleton).
When you start pulling in configs and the like, then you'll have your first framework object from which all other bases do their extension.
I have almost completed a PHP project, using MVC, jQuery and Ajax. It is pure PHP project. I don't use any frameworks in the code right know. I would like to change that.
Doing some research, I found, that Yii turns out to be one of the best frameworks out there.
Is it possible to somehow migrate pure PHP project to Yii?
If so, then how to do this? Which steps should I follow in order to reduce the workload and enjoy the benefits the Yii framework presents?
I'm a total Yii newbie, any insights appreciated.
TL;DR : Don't do it. It's a really horrible idea.
The Rant ..
"Framework" is not a magic sauce, that you add to a project, to make it better and shinier.
Doing some research i found Yii turns out to be one of the best frameworks out there.
What a strange research you have done .. I would love to see the materials. Especially, since I would rank it as 3rd worst PHP framework. Only surpassed in it's awfulness by CodeIgniter and CakePHP.
And the reason for it is the extremely bad quality of code, that this framework displays, combined with the bad practices, that it perpetuates.
Why avoid migration?
From your description is obvious, that you are NOT familiar with this framework and have no previous experience with it.
In management of projects there a subject: risk management. And in this case, adding a previously unused framework in final stages of project would constitute a high probability an high impact risk which also, due to the sage of project, is completely unmitigated.
It means that there is a really good chance that something will go wrong. And when it does, it most likely will sink the project. Or at least push back the release data by significant length of time.
In a perfect world frameworks are used to simplify the repetitive tasks of development, at the cost of some performance. Those are the tasks you do at the start of the project. You are not at the start of a project. This means that you will gain no benefits from this "maneuver".
Why not Yii?
As I noted before, there are also reasons not only for avoiding adding framework to an existing project, but also reasons why to avoid Yii in particular.
The inheritance nightmare
All your controller will extend class CController, which extends CBaseController, which extends CComponent
All your "models" will extend ether CActiveRecord or CFormModel, which extends CModel, which extends CComponent.
Both of there chains contain static variables and execute static methods on multitude of different other classes. Combination of these factors will make debugging extremely difficult and tedious.
Global state
There are several forms of global state. One that people in PHP usually know are global variables. But that is not the only form. Every time you have a class that contains a static variable, it also creates a global state, that can (and almost always - will) cause seemingly unrelated instance mysteriously interact.
Use of global state is a core mechanic. You will see static calls all over the codebase, and the Yii's configuration file would not function without global state.
And every time you call Yii::app() you are accessing and/or changing it.
This makes unittesting impossible for Yii applications. And debugging turns into exercise of using grep on your whole project.
Tight coupling
When you create an application in Yii. It becomes bound to it. You cannot execute parts of your application without launching the full framework. Mostly it is due to the static call, that you end up adding to your code.
Every time you add a static call in your own code, that piece of code becomes tied to the name of the class. That essentially is tight coupling.
As you might have noticed (hopefully), there is another way how to achieve the same effect - the use of new operator. That is another way of coupling some code of yours to a specific name of a class.
No interfaces .. none .. whatsoever
No matter how horrible the configuration of a Yii project is, the configuration file was a well intended gesture. The least harmful way to introduce external code and replace existing components in so messed up codebase.
But unfortunately it brings in the focus the problems caused by lack of interfaces and the existing coupling.
One of the component that developers will try to replace is the CUrlManager. Mostly due to way how you can pass additional parameters.
An interface in OOP specifies the contract between two instances. It lets you define the capabilities of an instance, the methods that can be used by others. When it's not there, in a large codebase, you are left guessing, which methods are required and which are not.
In case of Yii components the problem is compounded even further due to static call and deep inheritance. The above mentioned CUrlManager extends CApplicationComponent, which extends CComponent. Also the same file defines CUrlRule and CBaseUrlRule classes.
When you are writing a replacement, you have to write some code, plug it in the configuration and then test it by running your applications. That way you know which method (or parameter) next you need to add.
Basically, it's the "save-an-see-what-blows-up" method of development.
That's not MVC!
Yii does not implement MVC or any of MVC-inspired design patterns. What it calls "MVC" could be described as ActiveRecord-Template-Logic pattern.
Instead of having proper model layer (yes, it should be a layer), the creator(s) of Yii opted for collection of active record and form wrappers. This forces the application logic to be forced in the "controllers".
On the other hand you have glorified templates, instead of proper view instances for containing presentation logic. It is somewhat mitigated by use of widgets, but those instead suffer from SRP violations, because widgets are forced to contain bits of presentation logic and perform partial rendering. The rest of presentation logic ends up again in the controllers.
And to make it all worse, the "controllers" also have to deal with authorization. This usually will mean, that whenever you change the access scheme, you will have to go through every single instance of CController to check whether it needs to be changed too.
It's not MVC. It's a mess with names taken from MVC design pattern and slapped on some components.
All the small things ..
The framework also comes with few minor issue, that do not deserve a separate section:
Defining more then one class per file:
This will get annoying quite fast, because there will be classes that are shoehorned at the class files with completely unrelated filenames. This will mean, that debugging will quite often require use of search.
Slapped on "modules":
By the looks of it, the modules where added to the framework after the fact. Which is why, when you need to set default module, you will have to set it in the configuration files parameter, that is called 'defaultController'.
I actually recently converted a MVC pattern website I had built from the ground up into Yii. It did take some time to set it all up but in all honesty it was well worth it. I was able to throw away a lot of code because there were already Yii extensions doing what I needed. I would also suggest that you keep your database because you can create the controllers and Models using Gii which will save you a ton of time.
I don't know of any quick solutions to this. It depends upon how the code was written. You have the database and your views so it is not really a complete new project when you take into yii. Yii will generate the database models for you. You already have the views from the existing project. Write the controller and actions and modify the views if necessary.
try these links as they refer to the same problem.
How do you convert an old oop project into Yii
tips on migrating an existing site to Yii
Drupal to Yii Migration
Since you already have a code in mvc, things will be much easier for you to migrate. However, while migrating to Yii, since it can generate controller and model very easily using gii, you can take the advantage of it.
So, first generate controller and model using gii, then you can replace your existing code (by replace I mean, substitute your code to the specific function in the controller and model) to the built in controller and model so that the functionality of your site still works. You can modify your view accordingly. But that won't be much of a work.
You can simply register your script for ajax, jquery and css. Those will work as well.
And yes, Yii is the best framework out there so take as much benefit as you can.
Thanks,
Ujjwal
In this project you converted php to yii framework. Its really easy for you if you do following step.
Since you already have a code in mvc, things will be much easier for you to migrate. However, while migrating to Yii, since it can generate controller and model very easily using gii, you can take the advantage of it.
second, If your database is accurate then 50% work complete.when you create CRUD operation using gii then automatically model-view-controller create.if you create mvc in php then it benifit for you.
third,You can simply include your script for ajax, jquery and css. Those will work as well you create a folder in themes(CSS,JS,AZAX,BOOTSTRAP).
four-Protected->view->layout, where you can change your theme..thats all
you also help www.yiiframework.com/doc-2.0/guide-intro-yii.html
if you think my answer is help you then rating me...thank you.
I'm currently developing a framework which uses an object of a Core class (this class has huge functionality & makes the framework working). The framework follows MVC architecture & has loosely coupled Model, Control, View classes. Theses classes need a reference to the Core class heavily. What I've done so far is: creating single object of the Core class & referencing to it by PHP keyword global in Model, Control, View classes.
I don't like using this approach mainly because:
This way is not true object oriented way in my sense
The IDE (netbeans) can't provide documentation to the object of the Core class - a pain for developers who will be using this framework.
I'm really worried about performance issues - have no idea whether global is slower or whatever.
I've searched & did not find any information regarding performance issue. I've also searched stackoverflow & found Does using global create any overhead? & The advantage / disadvantage between global variables and function parameters in PHP? etc links but they don't contain much information. Right now my main concern is performance, so please help.
I must agree with NevilleK, that you Core` class sounds line an God Object antipattern.
And for anyone dumb enough to suggest use of singletons/registries i would suggest to do a little research on the subject. They create the same global state as your classical global variables.
Global state is not so much matter of performance ( though in php it has some minor impact ), but it created untestable and tightly coupled code.
You really should look into the Dependency Injection. That might show you another way , which does not require to have such a Core class in your code.
Some additional videos for you:
Global State and Singletons
Don't Look For Things!
Advanced OO Patterns
Cake is a Lie
Clean Code: Arguments
I've solved a similar problem in Agile Toolkit by creating a different pattern for adding object and using it system-wide. This passes a property to a newly created objects called "api" which always references Application class.
Application class is not really a God class but it delegates all sorts of functionality to system controllers, pages etc. This screencast explains how the very basic objects are structured, it might be something you are also looking for:
http://www.youtube.com/watch?v=bUNEHqYVOYs
Firstly, whilst you are concerned with performance, you may want to read http://en.wikipedia.org/wiki/God_object first - your "core" class sounds like a "God object", which is a fairly well-established anti pattern.
In terms of performance - the best way to find out is to test it. If you're writing a framework, I'm assuming you're writing unit tests to validate it's behaviour; it's not hard to extend that unit testing to include simple performance metrics. You might also invest in test scripts using JMeter or similar to exercise a "reference implementation" of a few pages built using the framework. You'll get far better information on your specific situation from doing this than by trying to optimize the design based on Stack Overflow's collective knowledge of the general way things work.
In general, I'd say that having a global class shouldn't impact performance too much, as long as it isn't doing much work. Simply loading the class into memory, parsing it, etc. does have a performance impact - but it's not likely to be measurably slower than any other routes you might take.
If, however, your "core" class does lots of initialization logic when it's accessed by the page, it's obviously going to impact performance.
I've been using the CodeIgniter framework, however after learning Java and awesome features like Interfaces, Abstract classes, packages, and that PHP 5 also supports most of these features, I'm ready to graduate and build a real OO framework in PHP which uses all of these features (including namespaces), so I can build much more elegant designs.
What I'm thinking of is that rather than everything in the system sharing a single $this-> object like its done in CodeIgniter, e.g:
$this->load->model('box');
$this->box->something();
You would do the following to load the Box model and to call its something() method.
$box = new Framework\Models\Box();
$box->something();
Or the following
abstract class BaseController
{
public function getModel($name)
{
$model = new Framework\Models\$model(); //Is this valid code?
return $model;
}
}
class MyController extends BaseController
{
public function index()
{
$box = $this->getModel('box');
$box->something();
}
}
Are there any suggestions/pointers for building this, such as which minimum system classes I would need for the framework, what namespaces I should have, or any other features?
One thing I've noticed is that a framework is usually built for a specific purpose. Generic frameworks (like CodeIgniter) are good for smaller sites and getting things up and running quickly. However once you have specific things, which fall outside of the generic framework building your own does become a reality.
The only thing I would suggest is to be consistent. That is relentlessly consistent. If you decide to name things in camelCase then don't deviate from that. Or if you decide to use the NounVerb convention of naming methods i.e. spaceJump then don't switch to jumpSpace even if it 'sounds' better at the time.
Choose how you'll accept parameters and be consistent on that. Will you accept only parameters or associative arrays of parameters? Choose and stick with it.
I would also not over-engineer before you've written anything. The consistency thing will take you pretty far before you'll need to refactor... but I wouldn't be afraid to do that as well. (A unit test or two should ease those fears).
Yup, there's usually no right or wrong way to do things as long as you're consistent. Did I mention...
Be Consistent!
A lot of experience is needed to build a technical framework (no offense) and there are already 1'000's of CMS/basic objects (session management, ORM, etc.) framworks/libraries.
Instead of wasting precious time (no offense again), you better have to:
Evaluate and select the framework that better suits your needs (some have a really good OO architecture).
Learn will using this good framework.
Build your own OO business/application framework. (This is where you can learn, differentiate and create value).
You can build your own but why not make use of the vast libraries included with frameworks available and if required extend their functionality.
Why don't you take a look at the Zend framework.
It's fairly quick to get started and includes a lot of useful libraries and classes as standard. It's good for personal projects if your just looking to gain more experience with OOP.
http://framework.zend.com/
I would look at Kohana. It came out of CodeIgnitor, and loading Models etc is done just the way you proposed.
Check out their core features, many of which relate directly to your question (I've highlighted those):
How is Kohana Different?
Although Kohana reuses many common design patterns and concepts, there are some things that make Kohana stand out:
Community, not company, driven. Kohana development is driven by a team of dedicated people that need a framework for fast, powerful solutions.
Strict PHP 5 OOP. Offers many benefits: visibility protection, automatic class loading, overloading, interfaces, abstracts, and singletons.
Extremely lightweight. Kohana has no dependencies on PECL extensions or PEAR libraries. Large, monolithic libraries are avoided in favor of optimized solutions.
GET, POST, COOKIE, and SESSION arrays all work as expected. Kohana does not limit your access to global data, but offers filtering and XSS protection.
True auto-loading of classes. True on-demand loading of classes, as they are requested in your application.
No namespace conflicts. All classes are suffixed to allow similar names between components, for a more coherent API.
Cascading resources offer unparalleled extensibility. Almost every part of Kohana can be overloaded or extended without editing core system files. Modules allow multi-file plugins to be added to your application, transparently.
Library drivers and API consistency. Libraries can use different "drivers" to handle different external APIs transparently. For example, multiple session storage options are available (database, cookie, and native), but the same interface is used for all of them. This allows new drivers to be developed for existing libraries, which keeps the API consistent and transparent.
Powerful event handler. Observer-style event handlers allow for extreme levels of customization potential.
Rapid development cycle. Rapid development results in faster response to user bugs and requests.
I'm going to write a framework for my web projects in PHP.
Please don't tell me about considering to use some existing framework (Cake, CodeIgniter, Symfony, etc.) - I have already had a look at them and decided to write one for myself.
The framework itself will mainly consist of a module system, a database handler and a template parser. (Many other things too, of course)
With module system I mean that every module has exactly one PHP file and one or more templates associated with it.
An example module would be modules/login.php that uses templates/login.tpl for its design.
These days everyone(?) is talking about the MVC (Model View Controller) concept and most of the existing frameworks use it, too.
So my questions are the following:
Is MVC really effective for a personal framework?
Would it be a bad idea to use a module system?
Did you ever write a framework for yourself? What are your experiences?
Is MVC really effective for a personal framework?
Yes, it can be. Although, it might be a little overkill (which, is not necessarily a bad thing if you are trying to learn)
Would it be a bad idea to use a module system?
This is never a bad idea.
Did you ever write a framework for yourself? What are your experiences?
I wrote a common security framework for my group's PHP applications when I was an intern. I learned alot, but the project as a whole might have benefited more from a pre-built solution.
Of course, I wouldn't have learned as much if I just installed a pre-built solution. So you always have to take that into account, especially for personal projects. Sometimes re-inventing the wheel is the only way you will learn something well.
Is MVC really effective for a personal framework?
What MVC means anymore, due to its vague interpretation, is business logic, presentation, and input handling. So, unless you aim to design an application that does not involve any three of those, MVC is, in its vague sense, very suitable.
Often it can be more formal than you desire, however, as it demands physical separation of ideas into different code files. Quick and dirty tasks or rapid prototyping might be more quickly setup if the formalities are avoided.
In the long term, what MVC asks for is beneficial to the sustainability of the application in ways of maintenance and modification or addition. You will not want to miss this. Not all frameworks encourage the right practices, though. I am not surprised that you find the various implementations you've tried insufficient. My personal favourite is Agavi. To me and others, in a world of PHP frameworks that do not feel right, Agavi emerges to do the right things. Agavi is worth the shot.
Would it be a bad idea to use a module system?
MVC asks you to separate components of business logic, presentation, and input handling, but it does not suggest how to layout the files. I presume this is the challenge you are addressing with a module system. To answer your question: modules serve identically to sub-directories. If the items are few, its probably more hassle to bother with subdirectories even if the files could logically be separated into them. When the number of items grow large, its now cumbersome to locate them all and sub-directories become a better option.
Frameworks will tack on functionality that allows you to deal with modules as their own configurable entity. The same functionality could just as well exist without modules, perhaps in a more cumbersome manor. Nonetheless, do not consider modules primarily as a system. Systems are so wonderfully vague that you can adapt them to whatever setup you find suitable.
Did you ever write a framework for yourself? What are your experiences?
Yes I have wrote several frameworks with various approaches to solving the issues of web applications. Every such framework I wrote became nothing but a vital learning curve. In each framework I made I discovered more and more the issues with building software. After failing to create anything interesting, I still gained because when asked to make a program I could fully do so with justice.
I recommend you continue if this is the sort of learning experience you want. Otherwise, give Agavi a shot. If that too fails, ensure that you have a clear and detailed specification of what your framework will do. The easiest way to barge into making software, work really hard, and accomplish nothing is to not decide before-hand what exactly your software will do. Every time I ran into making code the only thing in my mind was I will do it right. What happened was a different story: oh, well I need to make a routing system as that seems logical; hmm, okay, now I need a good templating system; alright, now time for the database abstraction; but gee, what a lot of thinking; I should look to the same system from software XXY for inspiration. Therein is the common cry that pleads to use existing software first.
The reason I thought I could do it right was not because all the nuts and bolts of the framework felt wrong. In fact, I knew nothing about how right or wrong they were because I never worked with them. What I did work with was the enamel, and it felt wonky. The quickest way to derive your own framework is really to steal the nuts and bolts from another and design your own enamel. That is what you see when building an application and frankly is the only part that matters. Everything else is a waste of your time in boilerplate. For learning how to build software, however, its not a waste of time.
If you have any other questions, please ask. I am happy to answer with my own experience.
I am also actually writing a php framework with a friend of mine. I absolutely can understand what you do.
I thing what you are doing is near mvc. You have the templates as views. And the modules as controller. So I think that is ok. The only thing you need is the model. That would be some kind of active records.
In my framework there are simular concepts, except we are writing our own active records engine at the moment. I think what you do isn't bad. But it's hard to say without seeing code.
I see only one problem you have to solve. A framework should be perfectly integrated. It is always a complicated to make your module look nice integrated without always have to think of module while you are coding application.
Is MVC really effective for a personal framework?
Would it be a bad idea to use a module system?
Yes it is. But MVC is such a loosy-goosy design pattern that you can draw the line between model, view, and controller anywhere you want. To me, the most important parts are the model and the view. I simply have pages, php modules, that generate html by filling in a template from a database. The pages are the view and the database is the model. Any common application-specific code can be factored out into "controllers". An example might be a common, sophisticated query that multiple pages must use to render data.
Other than that I have utilities for safe database access, simple templating, and other stuff.
Did you ever write a framework for yourself? What are your experiences?
Yes. I'm very glad I did. I can keep it simple. I know intimately how it works. I'm not dependent on anyone but myself. I can keep it simple yet useful.
Some pointers (0x912abe25...):
Every abstraction comes with a cost.
Don't get to fancy. You might regret not keeping it simple. Add just the right amount of abstraction. You may find you over-abstracted and something that should be simple became excessively complex. I know I've made this mistake. Remember You-aint-gonna-need-it.
Scope your variables well
Don't load your pages by doing
include_once('...page file ...');
where it's expected that page file will have a bunch of inline php to execute looking up different global variables. You lose all sense of scope. This can get nasty if you load your page file from inside a function:
function processCredentials()
{
if (credentialsFail)
{
include_once('loginpage.php');
}
}
Additionally, when it comes to scoping, treat anything plugged into templates as variables with scope. Be careful if you fill in templates from something outside the page file associated with that template (like a master index.php or something). When you do this it's not clear exactly what's filled in for you and what you are required to plug into the template.
Don't over-model your database with OO.
For simple access to the database, create useful abstractions. This could be something as simple as fetching a row into an object by a primary index.
For more complex queries, don't shy away from SQL. Use simple abstractions to guarantee sanitization and validation of your inputs. Don't get too crazy with abstracting away the database. KISS.
I would say that MVC makes more sense to me, since it feels better, but the only practical difference is that your login.php would contain both the model (data structure definitions) and the controller (code for page actions). You could add one file to the module, e.g. class.login.php and use __autoload() for that, which would essentially implement an MVC structure.
I have refactored a big PHP project to make it more MVC compliant.
I found especially usefull to create a DAO layer to centralize all database accesses. I created a daoFactory function, which creates the DAO and injects the database handle into it (also the logger, I used log4php, got injected).
For the DAO, i used a lot the functionalities of the database (mysql), like stored procedure and triggers. I completly agree with Doug T. about avoid over-abstraction, especially for database access : if you use the DB properly (prepared statements, etc.) you don't need any ORM and your code will be much faster. But of course you need to learn mysql (or postgress) and you become dependant on it (especially if you use a lot of stored procedure, like I tend to do).
I am currently refactoring a step further, using the Slim php framework and moving toward a restfull api : in this case there is no view anymore because everything is outputted as json. But I still use smarty because its caching works well and I know it.
Writing a framework could be a rewarding experience. The important thing to consider is that you do not write a framework for its own sake. The reason one writes a framework is to make development easy.
Since it is a personal framework you should think in terms of how it could help you develop with less hassle.
I do not think a template system is a good idea. Think of it - what is the major benefit of using a template system? The answer is that it helps teams with different skill sets jointly develop an application. In other words, some members of the team can work on the user interface and they do not need to be PHP coders. Now, a personal framework will most likely be used by a single person and the benefit of template system becomes irrelevant.
All in all, you should look at your own coding habits and methods and discover tasks that take most of your time on a typical project. Then you should ask yourself how you can automate those tasks to require less time and effort. By implementing those automation mechanisms you will have to stick to some sort of conventions (similar to an API). The sum of the helper mechanisms and the conventions will be your personal framework.
Good luck.
MVC doesn't work
you don't want to be constrained in the structure of your "modules"; also, keep templates close to the code (the templates directory is a bad idea)
no
re 1.: see Allen Holub's Holub on Patterns. briefly: MVC basically requires you to give up object oriented principles.
Tell Don't Ask is a catchy name for a mental trick that helps you keep the data and code that acts on it together. Views cause the Model to degrade into a heap of getters and setters, with few if any meaningful operations defined on them. Code that naturally belongs in the Model is then in practice spread among Controllers and Views(!), producing the unhealthy Distant Action and tight coupling.
Model objects should display themselves, possibly using some form of Dependency Injection:
interface Display
{
function display($t, array $args);
}
class SomePartOfModel
...
{
function output(Display $d)
{
$d->display('specific.tpl', array(
'foo' => $this->whatever,
...
));
}
}
OTOH, in practice I find most web applications call for a different architectural pattern, where the Model is replaced with Services. An active database, normalized schema and application specific views go a long way: you keep the data and code that acts on it together, and the declarative nature makes it much shorter than what you could do in PHP.
Ok, so SQL is a terribly verbose language. What prevents you from generating it from some concise DSL? Mind you, I don't necessarily suggest using an ORM. In fact, quite the opposite. Without Model, there's little use for an ORM anyway. You might want to use something to build queries, though those should be very simple, perhaps to the point of obviating such a tool...
First, keep the interface your database exposes to the application as comfortable for the application as possible. For example, hide complex queries behind views. Expose update-specific interfaces where required.
Most web applications are not only the owners of their respective underlying databases, they're their only consumers. Despite this fact, most web applications access their data through awkward interfaces: either a normalized schema, bare-bones, or a denormalized schema that turned out to make one operation easier at the price of severe discomfort elsewhere (various csv-style columns etc). That's a bit sad, and needlessly so.
re 2.: it's certainly good to have a unified structure. what you don't want to do is to lock yourself into a situation where a module cannot use more than one file.
templates should be kept close to code that uses them for the same reason that code that works together should be kept together. templates are a form of code, the V in MVC. you'll want fine-grained templates to allow (re)use. there's no reason the presentation layer shouldn't be as DRY as other parts of code.