YANP (Yet Another Namespace Proposal)

20 years ago by Jessie Hernandez — view source

unread

I have locally made changes to support namespaces, with minimal impact to
the Zend Engine (I based my changes on PHP 5.1.0b2). I have gone through
the archives and read the previous discussions about namespaces and the
outstanding issues, and to the best of my knowledge, I think my changes
solve these.

I have only a few things to finish (as I'll explain below), but so far
everything's working great (I have only enabled classes to be under
namespaces, since that's all I care about, but functions and constants can
be added in a similar manner)! Since I still have some things to finish, I
won't post a patch yet, but I'll give a high-level description of what I
did to see if my approach is acceptable or not.

Basically, my approach involves the following concepts:

Classes may not contain colons (no change here), BUT class name references
CAN contain colons (this is KEY!).
All imports are done at compile time.
Imports only affect the current file (yes, this has been done!)

Here are some more details:

Class Table Manipulation

When the start of a namespace declaration is encountered, the namespace name
is saved in a compile global variable. All classes inside the namespace
block will get the namespace name and the colon prepended to the name (this
is done in "zend_do_begin_class_declaration"). So, for the following code:

<code> namespace my_namespace{ class my_class{ public $mem; } } </code>

What will actually happen here is that the class will get added to the class
table as "my_namespace:my_class".

Referencing Classes

The following two examples are valid:

<code> $a = new my_namespace:myclass(); </code> <code> import my_namespace:myclass; $a = new myclass(); </code>

Imports

The syntax for the import statement will be:

import <full-class-name> [as <alias-name>];

I have not yet added full namespace imports ("wildcard" imports), and am not
yet sure if this is a good idea. Anyways, with the import statement, if no
alias is specified, then the class is imported with the remaining string
after the last colon in the name, and if an alias is specified, then that
name is used (aliases cannot contain colons). For example:

<code> import my_namespace:myclass; $a = new myclass(); </code> <code> import my_namespace:myclass as my_namespace_my_class; $a = new my_namespace_myclass(); </code>

When an import is done, what happens is that the import is saved in an
import hashtable, which holds the "aliases" which are active for the
current file. The alias is also saved in class_table (an additional entry
is added, the key being the alias name and the value being the
zend_class_entry for the actual class). This means that a zend_class_entry
can belong to more than one key in class_table at a time (this is done to
emulate that a class by that name actually exists, and it also prevents
classes in the global namespace to be named the same as import aliases).

Now for the interesting part! How are imports done on a file scope only? The
key to this is that file scopes can be emulated at compile time, but not at
run time. As I said before, imports are saved in a compile global
hashtable. When the end of the file is reached, this hashtable is cleared
and the alias entries are removed from class_table. This is done simply by
modifying the start rule:

start:
top_statement_list { zend_do_unimport_all(TSRMLS_C); }
;

This triggers the "zend_do_unimport_all" function to be called at the end of
the file. At this point (before the function is called), all references to
imported names have been changed internally to full class names, which
include the namespace and colon(s). Since include/require statements are
executed at runtime, when the "zend_compile_file" function is executed for
the included files, the import hashtable will be empty!

I have not yet begun doing this, but I would like the import statement to
actually include the file where the class is defined. A new php.ini
variable called "class_path" will be used for this. When an "import
my_ns:my_class;" statement is encountered, the file
"<class-path>/my_ns/my_class.php" will be included (of course, multiple
directories can be specified in the class_path, just like include_path).

The above changes make a minimal impact to the Zend engine, and since all
the namespace/import translation is done at compile time, execution speed
is not affected.

I would like to gather comments/suggestions/criticisms of the above
approach, in order to see if I should continue making these changes or not.
If there is enough demand, I will post the patches of what I have so far,
even though it's not complete.

Regards,

Jessie Hernandez

20 years ago by dz@bitxtender.com — view source

unread

Sounds sweet. However, I already found a problem there, it has to do with
aliasing. Lots of code (libraries) will, even inside the same class, use the
class name for method calls instead of the "self" keywords for various
reasons. This will break if the user decides to alias the class and use it,
I guess. Other than that, I don't see any problems so far. Does a
My:Nested:Namespace:Class::staticMethod() work?

Nice work man, keep it up, let's hope we can finally have some
namespaces/packaging support in PHP ;) I wish you luck. +1 from me.

David

P.S: check out http://article.gmane.org/gmane.comp.php.devel/29492, might be
of help, dunno

-----Original Message-----
From: Jessie Hernandez [mailto:jrhernandez05@gmail.com]
Sent: Thursday, July 07, 2005 2:44 AM
To: internals@lists.php.net
Subject: [PHP-DEV] YANP (Yet Another Namespace Proposal)

I have locally made changes to support namespaces, with minimal impact to
the Zend Engine (I based my changes on PHP 5.1.0b2). I have gone through
the archives and read the previous discussions about namespaces and the
outstanding issues, and to the best of my knowledge, I think my changes
solve these.

I have only a few things to finish (as I'll explain below), but so far
everything's working great (I have only enabled classes to be under
namespaces, since that's all I care about, but functions and constants can
be added in a similar manner)! Since I still have some things to finish, I
won't post a patch yet, but I'll give a high-level description of what I
did to see if my approach is acceptable or not.

Basically, my approach involves the following concepts:

Classes may not contain colons (no change here), BUT class name
references
CAN contain colons (this is KEY!).

All imports are done at compile time.

Imports only affect the current file (yes, this has been done!)

Here are some more details:

Class Table Manipulation

When the start of a namespace declaration is encountered, the namespace
name
is saved in a compile global variable. All classes inside the namespace
block will get the namespace name and the colon prepended to the name
(this
is done in "zend_do_begin_class_declaration"). So, for the following code:
<code> namespace my_namespace{ class my_class{ public $mem; } } </code>
What will actually happen here is that the class will get added to the
class
table as "my_namespace:my_class".

Referencing Classes

The following two examples are valid:
<code> $a = new my_namespace:myclass(); </code> <code> import my_namespace:myclass; $a = new myclass(); </code>
Imports

The syntax for the import statement will be:

import <full-class-name> [as <alias-name>];

I have not yet added full namespace imports ("wildcard" imports), and am
not
yet sure if this is a good idea. Anyways, with the import statement, if no
alias is specified, then the class is imported with the remaining string
after the last colon in the name, and if an alias is specified, then that
name is used (aliases cannot contain colons). For example:
<code> import my_namespace:myclass; $a = new myclass(); </code> <code> import my_namespace:myclass as my_namespace_my_class; $a = new my_namespace_myclass(); </code>
When an import is done, what happens is that the import is saved in an
import hashtable, which holds the "aliases" which are active for the
current file. The alias is also saved in class_table (an additional entry
is added, the key being the alias name and the value being the
zend_class_entry for the actual class). This means that a zend_class_entry
can belong to more than one key in class_table at a time (this is done to
emulate that a class by that name actually exists, and it also prevents
classes in the global namespace to be named the same as import aliases).

Now for the interesting part! How are imports done on a file scope only?
The
key to this is that file scopes can be emulated at compile time, but not
at
run time. As I said before, imports are saved in a compile global
hashtable. When the end of the file is reached, this hashtable is cleared
and the alias entries are removed from class_table. This is done simply by
modifying the start rule:

start:
top_statement_list { zend_do_unimport_all(TSRMLS_C); }
;

This triggers the "zend_do_unimport_all" function to be called at the end
of
the file. At this point (before the function is called), all references to
imported names have been changed internally to full class names, which
include the namespace and colon(s). Since include/require statements are
executed at runtime, when the "zend_compile_file" function is executed for
the included files, the import hashtable will be empty!

I have not yet begun doing this, but I would like the import statement to
actually include the file where the class is defined. A new php.ini
variable called "class_path" will be used for this. When an "import
my_ns:my_class;" statement is encountered, the file
"<class-path>/my_ns/my_class.php" will be included (of course, multiple
directories can be specified in the class_path, just like include_path).

The above changes make a minimal impact to the Zend engine, and since all
the namespace/import translation is done at compile time, execution speed
is not affected.

I would like to gather comments/suggestions/criticisms of the above
approach, in order to see if I should continue making these changes or
not.
If there is enough demand, I will post the patches of what I have so far,
even though it's not complete.

Regards,

Jessie Hernandez

20 years ago by Timm Friebe — view source

unread

[...]

All imports are done at compile time.

So:

== Object.php ==
<?php
package lang { class Object {

}}
?>

== script.php ==
<?php
include('Object.php');

import lang:Object;
?>

...will not work. You'll need an opcode for this.

Timm

20 years ago by Jessie Hernandez — view source

unread

Actually, this will work! In the "class_name_reference" rule, the
"zend_do_fetch_class" function is called. Inside this function, class_name
is a znode, and what is assigned to the opcode is "opline->op2 =
*class_name". This function is changed to check the import hashtable, and if
a match is found, then the znode string value is updated to contain the full
class name! Here's some pseudocode:

<?php
include('Object.php');
// include opcode is generated here

import lang:Object;
// "Object" => "lang:Object" added to import hashtable

$o = new Object();
// zend_do_fetch_class checks for "Object" in the import hashtable.
// A match is found, so "class_name->u.constant.value.str" is modified from
"Object" to "lang:Object".
// At runtime, the lang:Object class will be created and added to
class_table,
// and since all opcodes are using lang:Object instead of simply Object,
then everything will work.
?>

--
Jessie

""Timm Friebe"" thekid@thekid.de wrote in message
news:0MKwh2-1DqTBy43Wu-0004sH@mrelayeu.kundenserver.de...

[...]

All imports are done at compile time.

So:

== Object.php ==
<?php
package lang { class Object {

}}
?>

== script.php ==
<?php
include('Object.php');

import lang:Object;
?>

...will not work. You'll need an opcode for this.

Timm

20 years ago by Stanislav Malyshev — view source

unread

JH>>is a znode, and what is assigned to the opcode is "opline->op2 =
JH>>*class_name". This function is changed to check the import hashtable, and if
JH>>a match is found, then the znode string value is updated to contain the full
JH>>class name! Here's some pseudocode:

This of course won't work with variable class names.

Stanislav Malyshev, Zend Products Engineer
stas@zend.com http://www.zend.com/ +972-3-6139665 ext.115

20 years ago by Jessie Hernandez — view source

unread

Stanislav,

Correct, in fact, I began to work on this yesterday. My approach will be to
translate whatever's possible at compile-time, as I'm doing now, and saving
the import alias info for what's needed at runtime, such as in the case of
variable class names. This hashtable will also be needed for namespace
imports, if they are in fact implemented.

--
Jessie

Stanislav Malyshev wrote:

JH>>is a znode, and what is assigned to the opcode is "opline->op2 =
JH>>*class_name". This function is changed to check the import hashtable,
and if JH>>a match is found, then the znode string value is updated to
contain the full JH>>class name! Here's some pseudocode:

This of course won't work with variable class names.

20 years ago by Jessie Hernandez — view source

unread

The attached patch includes variable class name support (test file has also
been updated). To clarify for the benefit of others, this means that the
code below will work. Note that imports are still done on "file-level"
scope, even at runtime, so you can only create objects using variable
import class names for the imports that have been declared in that same
file.

<?php
import my_ns:my_class;

$n = 'my_class';
$c = new $n();
?>

--
Jessie

Stanislav Malyshev wrote:

JH>>is a znode, and what is assigned to the opcode is "opline->op2 =
JH>>*class_name". This function is changed to check the import hashtable,
and if JH>>a match is found, then the znode string value is updated to
contain the full JH>>class name! Here's some pseudocode:

This of course won't work with variable class names.

20 years ago by Xuefer — view source

unread

at compile time? this means u cannot import * (all) symbols from a namespace

20 years ago by Jessie Hernandez — view source

unread

I'm still debating whether importing all symbols from a namespace is a good
idea or not (I sometimes use Java at work, and I never import a full package
myself, and this is not recommended anyways). But, if it were to be
implemented, then below is how I would approach it. The following would be
required:

As I said before, the "class_path" php.ini variable would be used for
imports.
Each file under the class_path is named exactly like the class that it
declares. For example, my_namespace/my_class.php would define the class
"my_class" under the my_namespace namespace (just like in Java).

If an "import namespace my_namespace;" (the syntax is still tentative)
statement is found, then the directory <class-path>/my_namespace will be
opened (at compile-time). Each file under this directory would be imported.
So, assuming our class_path is "/usr/local/php-classes" and the
"/usr/local/php-classes/my_namespace" directory had two files, "class1.php"
and "class2.php", then this is what would happen:

<?php
import namespace my_namespace;

// Directory "/usr/local/php-classes/my_namespace" is opened.
// There are two files, class1.php and class2.php.
//
// 1) "my_namespace:class1" is added to the import hashtable.
// 2) The opcode for
"require_once('/usr/local/php-classes/my_namespace/class1.php')" is
generated.
// 3) "my_namespace:class2" is added to the import hashtable.
// 4) The opcode for
"require_once('/usr/local/php-classes/my_namespace/class2.php')" is
generated.
?>

There would also need to be another hashtable to indicate if a full
namespace import for a particular namespace has already been done or not, to
avoid a directory from being read over and over again.

--
Jessie

"Xuefer" xuefer@gmail.com wrote in message
news:28139bc05070705364cc572f7@mail.gmail.com...
at compile time? this means u cannot import * (all) symbols from a namespace

20 years ago by Stanislav Malyshev — view source

unread

JH>>// 1) "my_namespace:class1" is added to the import hashtable.
JH>>// 2) The opcode for
JH>>"require_once('/usr/local/php-classes/my_namespace/class1.php')" is
JH>>generated.
JH>>// 3) "my_namespace:class2" is added to the import hashtable.
JH>>// 4) The opcode for
JH>>"require_once('/usr/local/php-classes/my_namespace/class2.php')" is
JH>>generated.
JH>>?>

This means, of course, that such file could not be used with bytecode
caches, because between runs contents of the directory can change.

--
Stanislav Malyshev, Zend Products Engineer
stas@zend.com http://www.zend.com/ +972-3-6139665 ext.115

20 years ago by Jessie Hernandez — view source

unread

Could you elaborate on this? Let's say we have the following situation:

a.php does a full namespace import of "my_namespace".
The <class_path>/my_namespace directory has two files (classes),
class1.php and class2.php.
a.php has the following code:
<?php
import namespace my_namespace;
$c1 = new class1();
$c2 = new class2();
?>
After a.php is first run, a new file, class3.php, is added.

With the above scenario, nothing will be affected if there is an opcode
cache or not. The only situation I can think of where an opcode cache can
affect this is when the script DYNAMICALLY creates classes from the imported
namespace. Maybe there's other situations, but I can't come up with any at
the moment.

Either way, I'm not sure if full namespace imports are a good idea anyways.
Issues like the one you presented are one reason, and the other is that it
is bad practice. If multiple namespaces are imported, readability is
affected (which class belongs to which namespace?). Also, if two namespaces
are imported, and one day a new class is added to one which has the same
name as another, then there will suddenly be a compile error, and the error
will be elusive.

--
Jessie

"Stanislav Malyshev" stas@zend.com wrote in message
news:Pine.LNX.4.62.0507072119540.24702@mail.zend.com...

JH>>// 1) "my_namespace:class1" is added to the import hashtable.
JH>>// 2) The opcode for
JH>>"require_once('/usr/local/php-classes/my_namespace/class1.php')" is
JH>>generated.
JH>>// 3) "my_namespace:class2" is added to the import hashtable.
JH>>// 4) The opcode for
JH>>"require_once('/usr/local/php-classes/my_namespace/class2.php')" is
JH>>generated.
JH>>?>

This means, of course, that such file could not be used with bytecode
caches, because between runs contents of the directory can change.

--
Stanislav Malyshev, Zend Products Engineer
stas@zend.com http://www.zend.com/ +972-3-6139665 ext.115

20 years ago by Stanislav Malyshev — view source

unread

JH>>With the above scenario, nothing will be affected if there is an
JH>>opcode cache or not. The only situation I can think of where an opcode
JH>>cache can affect this is when the script DYNAMICALLY creates classes
JH>>from the imported namespace. Maybe there's other situations, but I
JH>>can't come up with any at the moment.

No, more simple - somebody removes one of the classes. Even if nothing in
your code ever used it, you'd start getting errors about missing files.

JH>>Either way, I'm not sure if full namespace imports are a good idea anyways.

Without this, only thing you are achieving with imports is saving a couple
of keystrokes. If you have to declare each imported class anyway, why not
name it by the full name? Just saving a dozen of keystrokes (which any
good code-completing editor would save anyway) and one require statement
is not worth the trouble, IMHO.

JH>>Issues like the one you presented are one reason, and the other is that it
JH>>is bad practice. If multiple namespaces are imported, readability is
JH>>affected (which class belongs to which namespace?). Also, if two namespaces

That's a general problem with all namespace imports.

JH>>are imported, and one day a new class is added to one which has the same
JH>>name as another, then there will suddenly be a compile error, and the error
JH>>will be elusive.

Right, that's one more problem with namespace imports.

Stanislav Malyshev, Zend Products Engineer
stas@zend.com http://www.zend.com/ +972-3-6139665 ext.115

20 years ago by Al Baker — view source

unread

I think we're trying to boil the ocean here, and in doing so failing to
see both the problems that need to be solved, and the possible
solutions. I'd like to break things down a little bit here to address
what the real problem is with a lack of namespaces in PHP.

In my eyes, there are really only two problems that arise because of
lack of namespaces. The first is the one true problem we should try to
solve and that is class naming conflicts. Now that PHP is growing and
make inroads into big business, several large development groups may
collaborate on a project. Or, they may want to use a library, only to
find out a class naming conflict. After all they love about PHP, it
seems like this problem makes their lives more difficult by having to
manually prefix all of their classes and use those prefixes throughout
their code.

So, the most basic namespace proposal should just be a prefix to
classes, allowing the compiler to differentiate between two classes of
the same name, but from different packages. $myobj = new
MySpace:MyClass(); seems good to enough to me. If this is all
namespaces are, an extra label to prefix classes with - no function
namespace, no import, then I say no problem. A namespace hierarchy
should be problem free as well, allowing developers a little more
structure. At compile time, it just results in a longer prefix than if
it were a single namespace.

This would also solve a slightly different problem, and that is one of
perception to enterprise developers. I know that productivity wise, the
above namespace proposal hardly saves the developer more time than
having to hardcode prefixes into class names. But what it does do, is
it makes sense to a large population of developers, especially those
'enterprise' (ie Java guys). Honestly, when I pitch how productive PHP
is on my project (100+ developers), the lack of namespaces is a good jab
against PHP. I know people cringe at the term enterprise a little
because accomodating the enterprise doesn't always mean accomodating
everyone. However, in this case, I would say we have critical mass of
interest in namespaces. Having a better prefixing mechanism wouldn't
break legacy code, would be optional, and would make a bunch of people
happy.

Now, I would assert that Java's import system (of which this proposal
and others often try to follow) is only partially about namespaces, but
is more of a mix of their compile system, source structure, and even
naming conventions. We could do the namespace proposal and not do
imports.

The whole concept of Java's imports is only partially related to
namespaces. There are a lot of assumptions made in Java's import
system, such as a configuration setting where to look, having one class
per file, having that class as the filename for that file, etc. This
file including system for the compile matches up with the namespace
syntax. Though there's an impression that this is a great system, I
would say that it still results in confusion. The dot notation in Java
can be used in the context of prefixing a class for proper compilation,
accessing a function, a member variable (including those that are
classes), a function, and you can string them together like
mypackage.myclass.function.returnfunctionsobject.function;. There's no
Java.ini to set a directory, you have to set environment variables.

If we want to build a more advanced include system, then I would say
that the labels should map to sub-directories in some directory
specified in php.ini, and the last token in the import statement just
performs a normal PHP include. E.g. if it's import mypackage.*; it
would go to the predefined directory, go to the my package subdirectory,
and perform an include on all the class files in that directory. Of
course, we'd always want to have strict class/file names for this to
work as well and only one class per file.

Thanks,
Al

20 years ago by Al Baker — view source

unread

One more thing, here's an example of how things go in the enterprise and
where namespaces make or break PHP being used.

Every now and then, someone says "Let's merge projects A and B", and
those projects could be in different parts of the world, each having 10s
if not 100s of developers, and very large code bases. These projects
may rarely communicate if ever if they sell to different markets.

In Java, you can start writing code to tie them together, because each
of them has defined their own package, and you can re-use that code
without modifying it inside new code. You just do an import, perhaps
use the full namespace prefix (e.g. package) when using classes that
conflict otherwise.

In PHP today, there are two possibilities. Either someone was smart and
enforced some coding guidelines (e.g. PEAR) and people have prefixed
their classnames (unlikely as this is an uncommon programming aspect) OR
we have a deal breaker issue. Project A has 10,000 lines of code, and
Project B has 10,000 lines of code and we have 50 classes of the same
name. A global replace doesn't do it, because I may have SQL, other
variables, function names similar to that class name. So, one project
may get a rewrite or invest a large number of resources to make things
compatible and hopefully one project doesn't get the axe. If management
gets word of what happened, PHP will be made out to be the villian and
there'll be a big fuss and a technology change to Java or something
else.

I just wanted to throw this out there to set some context on why it's
important to the enterprise. I know this feels like a corner case to
everyone else, but this is what happens when you have huge organizations
developing vast amounts of code. PHP is already a tough sell to a
population that has been trained (thanks to colleges) and experienced in
something like Java. Thankfully, there are pragmmatic developers out
there who see the light of PHP, adopt it, and watch their productivity
shoot through the roof for web applications.

Anyways, I'm +1 on this YANP sans the import feature as defined in
Jessie's proposal.

Thanks!
Al

I think we're trying to boil the ocean here, and in doing so failing to
see both the problems that need to be solved, and the possible
solutions. I'd like to break things down a little bit here to address
what the real problem is with a lack of namespaces in PHP.

In my eyes, there are really only two problems that arise because of
lack of namespaces. The first is the one true problem we should try to
solve and that is class naming conflicts. Now that PHP is growing and
make inroads into big business, several large development groups may
collaborate on a project. Or, they may want to use a library, only to
find out a class naming conflict. After all they love about PHP, it
seems like this problem makes their lives more difficult by having to
manually prefix all of their classes and use those prefixes throughout
their code.

So, the most basic namespace proposal should just be a prefix to
classes, allowing the compiler to differentiate between two classes of
the same name, but from different packages. $myobj = new
MySpace:MyClass(); seems good to enough to me. If this is all
namespaces are, an extra label to prefix classes with - no function
namespace, no import, then I say no problem. A namespace hierarchy
should be problem free as well, allowing developers a little more
structure. At compile time, it just results in a longer prefix than if
it were a single namespace.

This would also solve a slightly different problem, and that is one of
perception to enterprise developers. I know that productivity wise, the
above namespace proposal hardly saves the developer more time than
having to hardcode prefixes into class names. But what it does do, is
it makes sense to a large population of developers, especially those
'enterprise' (ie Java guys). Honestly, when I pitch how productive PHP
is on my project (100+ developers), the lack of namespaces is a good jab
against PHP. I know people cringe at the term enterprise a little
because accomodating the enterprise doesn't always mean accomodating
everyone. However, in this case, I would say we have critical mass of
interest in namespaces. Having a better prefixing mechanism wouldn't
break legacy code, would be optional, and would make a bunch of people
happy.

Now, I would assert that Java's import system (of which this proposal
and others often try to follow) is only partially about namespaces, but
is more of a mix of their compile system, source structure, and even
naming conventions. We could do the namespace proposal and not do
imports.

The whole concept of Java's imports is only partially related to
namespaces. There are a lot of assumptions made in Java's import
system, such as a configuration setting where to look, having one class
per file, having that class as the filename for that file, etc. This
file including system for the compile matches up with the namespace
syntax. Though there's an impression that this is a great system, I
would say that it still results in confusion. The dot notation in Java
can be used in the context of prefixing a class for proper compilation,
accessing a function, a member variable (including those that are
classes), a function, and you can string them together like
mypackage.myclass.function.returnfunctionsobject.function;. There's no
Java.ini to set a directory, you have to set environment variables.

If we want to build a more advanced include system, then I would say
that the labels should map to sub-directories in some directory
specified in php.ini, and the last token in the import statement just
performs a normal PHP include. E.g. if it's import mypackage.*; it
would go to the predefined directory, go to the my package subdirectory,
and perform an include on all the class files in that directory. Of
course, we'd always want to have strict class/file names for this to
work as well and only one class per file.

Thanks,
Al

20 years ago by Andi Gutmans — view source

unread

Did you make sure to real all the past threads on the issue?
There are a variety of problems which arise with such a namespaces
implementation, including ambiguity, problems with multiple symbol table
lookups, dynamic variable creation, cross-file dependencies. I believe this
solution suffers from mostly the same problems.

Andi

At 08:44 PM 7/6/2005 -0400, Jessie Hernandez wrote:

I have locally made changes to support namespaces, with minimal impact to
the Zend Engine (I based my changes on PHP 5.1.0b2). I have gone through
the archives and read the previous discussions about namespaces and the
outstanding issues, and to the best of my knowledge, I think my changes
solve these.

I have only a few things to finish (as I'll explain below), but so far
everything's working great (I have only enabled classes to be under
namespaces, since that's all I care about, but functions and constants can
be added in a similar manner)! Since I still have some things to finish, I
won't post a patch yet, but I'll give a high-level description of what I
did to see if my approach is acceptable or not.

Basically, my approach involves the following concepts:

Classes may not contain colons (no change here), BUT class name references
CAN contain colons (this is KEY!).

All imports are done at compile time.

Imports only affect the current file (yes, this has been done!)

Here are some more details:

Class Table Manipulation

When the start of a namespace declaration is encountered, the namespace name
is saved in a compile global variable. All classes inside the namespace
block will get the namespace name and the colon prepended to the name (this
is done in "zend_do_begin_class_declaration"). So, for the following code:
<code> namespace my_namespace{ class my_class{ public $mem; } } </code>
What will actually happen here is that the class will get added to the class
table as "my_namespace:my_class".

Referencing Classes

The following two examples are valid:
<code> $a = new my_namespace:myclass(); </code> <code> import my_namespace:myclass; $a = new myclass(); </code>
Imports

The syntax for the import statement will be:

import <full-class-name> [as <alias-name>];

I have not yet added full namespace imports ("wildcard" imports), and am not
yet sure if this is a good idea. Anyways, with the import statement, if no
alias is specified, then the class is imported with the remaining string
after the last colon in the name, and if an alias is specified, then that
name is used (aliases cannot contain colons). For example:
<code> import my_namespace:myclass; $a = new myclass(); </code> <code> import my_namespace:myclass as my_namespace_my_class; $a = new my_namespace_myclass(); </code>
When an import is done, what happens is that the import is saved in an
import hashtable, which holds the "aliases" which are active for the
current file. The alias is also saved in class_table (an additional entry
is added, the key being the alias name and the value being the
zend_class_entry for the actual class). This means that a zend_class_entry
can belong to more than one key in class_table at a time (this is done to
emulate that a class by that name actually exists, and it also prevents
classes in the global namespace to be named the same as import aliases).

Now for the interesting part! How are imports done on a file scope only? The
key to this is that file scopes can be emulated at compile time, but not at
run time. As I said before, imports are saved in a compile global
hashtable. When the end of the file is reached, this hashtable is cleared
and the alias entries are removed from class_table. This is done simply by
modifying the start rule:

start:
top_statement_list { zend_do_unimport_all(TSRMLS_C); }
;

This triggers the "zend_do_unimport_all" function to be called at the end of
the file. At this point (before the function is called), all references to
imported names have been changed internally to full class names, which
include the namespace and colon(s). Since include/require statements are
executed at runtime, when the "zend_compile_file" function is executed for
the included files, the import hashtable will be empty!

I have not yet begun doing this, but I would like the import statement to
actually include the file where the class is defined. A new php.ini
variable called "class_path" will be used for this. When an "import
my_ns:my_class;" statement is encountered, the file
"<class-path>/my_ns/my_class.php" will be included (of course, multiple
directories can be specified in the class_path, just like include_path).

The above changes make a minimal impact to the Zend engine, and since all
the namespace/import translation is done at compile time, execution speed
is not affected.

I would like to gather comments/suggestions/criticisms of the above
approach, in order to see if I should continue making these changes or not.
If there is enough demand, I will post the patches of what I have so far,
even though it's not complete.

Regards,

Jessie Hernandez

20 years ago by Jessie Hernandez — view source

unread

For previewing purposes, I have attached a patch of what I have so far so
that others can look at it and try it out. I have also attached two PHP
files for testing (_testns.php is the script that should be run after the
patch is applied. This script includes a class from class3.php to
demonstrate two things: namespaces may be spread out in files, and classes
do not have to be immediately available for imports to work).

I based my patch on PHP 5.1.0b2 (not CVS, yet). I have not used the diff
utility before, but here is the command I ran to produce the patch (let me
know if there's a better way):

diff -Bbruw -X diff_exclude.txt php-5.1.0b2-bak/Zend php-5.1.0b2/Zend/ >
ns.patch

Anyways, all the features I've mentioned have been enabled in the patch
except the following:

Automatic including of class files.
Wildcard imports (or full namespace imports).

I have been checking the code to see what it would take to apply the current
include/require logic against a different INI variable (class_path), but I
digged deeper and deeper into the code and got lost in the streams
functions. So far, it seems I have to create my own opcode (like
ZEND_INCLUDE_OR_EVAL), and when that's encountered, then call whatever
stream function is needed to perform the include against class_path. It
seems like it will take me some time to figure this out, so that's why I
haven't included it yet (maybe someone can help me?).

Please try it out and let me know what you think. I believe this solution
compiles the code to be the equivalent of prefixing all classes, so I think
most of the issues presented in the archives are solved with this solution.
Let me know if you need anything else from me to get it working on your
installation. Also, I'm not sure if there are any attachment restrictions
in this group (email me privately if you can't open the files).

--
Jessie

20 years ago by Stanislav Malyshev — view source

unread

JH>>- Classes may not contain colons (no change here), BUT class name references
JH>>CAN contain colons (this is KEY!).
JH>>- All imports are done at compile time.
JH>>- Imports only affect the current file (yes, this has been done!)

The problem here is that once you have somewhat developed package - like
PEAR packages - which may contain dozens of internal classes, you'd have
to have a large import statement block at the beginning of each file using
them - or refer to them all by the full name, thus just replacing
underscore-notation with colon-notation.

Stanislav Malyshev, Zend Products Engineer
stas@zend.com http://www.zend.com/ +972-3-6139665 ext.115

20 years ago by Jessie Hernandez — view source

unread

As requested by others, attached is a cvs patch that includes the latest
namespace additions and PHP test files. With this patch, the following
features have been added:

import statements now include the class files, using a new .ini variable,
"class_path"!
Near-full support of private classes! Currently, private classes cannot
be used outside their namespace, but can be used within methods of classes
within the namespace. What's left for me to do is to allow private classes
to be extended from other classes in the same namespace (this is almost
complete in this patch, it works if both classes are defined in the same
file). Private classes are declared simply as "private class_name" within a
namespace (if the "private" keyword is applied to a class in the global
namespace, an error will be generated).

Aside from what's pending in #2, I think the only missing feature is
"namespace imports". I'd like to gather thoughts from others as how this
can be best implemented (some issues have been raised as how namespace
imports will work with opcode caches). I'll post a few possible approaches
some other day this week (I'm tired now ;-)).

BTW, I tried to do a "cvs add" to add my test files, but of course, it
failed with an "cvs add requires write access to repository" error. Hence,
I had to zip up the files separately. How can I add new files locally so
that cvs diff picks them up in the patch?

Regards,

Jessie Hernandez