-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Hello Internals,
Has anyone given thought to a feature that allows (PHP) developer to
invalidate classes (allowing the developer to re-load the class, with
modified code)? I realize that this is not possible now, and for good
reason. But I think the idea has potential.
I asked Sara about this, yesterday, after the idea occurred to me, and
she graciously whipped up a small extension that does exactly what I'm
looking for, but she told me she thought it was a bad idea to be doing
this (and also allows destruction of functions). I agree: without the
proper safety nets, this is a dangerous thing to be doing.
So, why do I want this functionality, in the first place? Consider a
long-running application (not a web page, but, for instance, a
sockets-based application). It would be very useful for me, as a
developer, to be able to re-load parts of the application, without
tossing the whole current instance of PHP (and all associated resources)
-
-- ie, doesn't release socket connections, but allows re-loading of the
associated handlers. This could be accomplished by invalidating a class
(destroying it, and marking all instanciated objects with a special type
(__PHP_Destroyed_Class comes to mind, after __PHP_Incomplete_Class), and
re-declaring the class. -
-- sample code (working) --
<?php
eval('Class Foo { function name() { echo "Foo\n"; } }');
Foo::name();
shiva_destroy_class('Foo');
eval('Class Foo { function name() { echo "Bar\n"; } }');
Foo::name();
?> -
-- output --
Foo
Bar
I also spoke (typed) with Stig, who told me that this sounds more like a
PHP 6 feature. I understand that class destruction breaks a lot of
assumptions. As Sara pointed out, it will definitely break optimizers.
Also, the potential for segfault is very high:
Class Foo { function name() { echo "Foo\n"; } }; $Foo = new Foo();
shiva_destroy_class('Foo'); eval('Class Foo { function name() { echo
"Bar\n"; } };'); $Foo->name(); results in segfault..
I'm, admittedly, completely out of my league here.. I played with the
API a little, and was reminded of this. (-:
It may seem like a stupid idea at first, but I think there's some value
in re-loading pieces of code, and with the proper object
counting/invalidation, sounds like it could be safe, but like I said, I
don't know what I'm doing.
Is this worth further discussion?
S
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org
iD8DBQFBI9G1WppknrQMxQIRAjPOAJ9gsQ6AZlwd5QI6+CYerGOgtsKwhQCfejxZ
zfMP6MgKuGvHx7dO4mPNh90=
=GR/W
-----END PGP SIGNATURE
Sean Coates wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1Hello Internals,
Has anyone given thought to a feature that allows (PHP) developer to
invalidate classes (allowing the developer to re-load the class, with
modified code)? I realize that this is not possible now, and for good
reason. But I think the idea has potential.
[snip]
Is this worth further discussion?
There is a clever php4 implementation that allows loading/unloading of
methods, info at http://pear.php.net/pepr/pepr-proposal-show.php?id=68
It does not use any engine magic and might be of use to you while you
wait for PHP 6. :)
Greg
Hello Sean,
Thursday, August 19, 2004, 12:01:25 AM, you wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Hello Internals,
Has anyone given thought to a feature that allows (PHP) developer to
invalidate classes (allowing the developer to re-load the class, with
modified code)? I realize that this is not possible now, and for good
reason. But I think the idea has potential.
[................]
Is this worth further discussion?
Not before until we have an application server and php scripts that are
designed to run for several months. And even then we would also need
improved Reflection support and conatiner support and so on and so on.
Best regards,
Marcus mailto:helly@php.net
Marcus Boerger wrote:
Is this worth further discussion?
Not before until we have an application server and php scripts that are
designed to run for several months. And even then we would also need
improved Reflection support and conatiner support and so on and so on.
No disrespect intended, but this sounds like a chicken-and-egg problem
to me.
S
(PS- Wez, I remembered to turn off my signing this time (-: )
Is this worth further discussion?
Not before until we have an application server and php scripts that
are designed to run for several months. And even then we would also
need improved Reflection support and conatiner support and so on and
so on.
dont you think that lacking this destruction/loading causes potential app server developers to drop their idea? im working on such thing and working
on components in write code-[ctrl-c server]-load server isn`t fun at all :]
rash
Hello Robert,
Thursday, August 19, 2004, 10:29:31 AM, you wrote:
Is this worth further discussion?
Not before until we have an application server and php scripts that
are designed to run for several months. And even then we would also
need improved Reflection support and conatiner support and so on and
so on.
don
t you think that lacking this destruction/loading causes potential app server developers to drop their idea? im working on such thing and working
on components in write code-[ctrl-c server]-load server isn`t fun at all :]
Could you share some more infos about that project?
marcus
Could you share some more infos about that project?
sure
the project itself is my university thesis, its topic might clear some
things: 'Hive -distributed server for message-driven componental PHP
applications'. main idea is to be able to create php applications in
RAD-way. for this to be possible i designed componental architecture based
on messaging system. it is exacly the way php is NOT designed to be used ;)
all this with transparent component distribution in mind. server runs as
standalone appliacation. http daemon uses sockets to transfer request
parameters and fetch results - later i`ll probably build into app server
simple httpd. server is ('will be' mostly) coded in php, so changes can be
made easily, but if my idea will be useful i have in mind rewriting code in
c. no stress tests have been made so far - to much coding is still to be
done.
it is mix of ideas from java, windows messaging and few very theoretical
books. behind my back lies 1 meter high mountain of books :)
from practical point of view - its something like srm, at least i think so -
i`ve never used it. from the beginning its designed for php5 new super duper
mechanisms.
by now you probably realized why ive showed up in this thread - ability to manage components flexible is crucial in my project. i need for example way to detect class name collisions before compiler dies with error - there is no way to control naming conventions for components. i thought about some nasty ideas like forcing names in HIVE_UUID form, but as you surely understand - its horrible solution. probably ill have to use second process
to load component and use reflection to check for collisions before loading
in main app server process.
rash
ps. no thread support sucks :]
On Wed, 18 Aug 2004 18:01:25 -0400
Sean Coates sean@caedmon.net wrote:
<skip> > - -- sample code (working) -- > <?php > eval('Class Foo { function name() { echo "Foo\n"; } }'); > Foo::name(); > shiva_destroy_class('Foo'); > eval('Class Foo { function name() { echo "Bar\n"; } }'); > Foo::name(); > ?> > - -- output -- > Foo > Bar-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1Hello Internals,
Has anyone given thought to a feature that allows (PHP) developer to
invalidate classes (allowing the developer to re-load the class, with
modified code)? I realize that this is not possible now, and for good
reason. But I think the idea has potential.
Well, some time ago I wrote a small functions, that does exactly the same:
PHP_FUNCTION(unregister_class)
{
zval **class;
zend_class_entry *ce = NULL, **pce;
char *lowercase_name = NULL;
if (ZEND_NUM_ARGS() != 1 || zend_get_parameters_ex(1, &class)==FAILURE) {
ZEND_WRONG_PARAM_COUNT();
}
convert_to_string_ex(class);
if (zend_lookup_class(Z_STRVAL_PP(class), Z_STRLEN_PP(class), &pce TSRMLS_CC) == SUCCESS) {
ce = *pce;
}
if (!ce) {
RETURN_FALSE;
}
lowercase_name = emalloc(ce->name_length+1);
zend_str_tolower_copy(lowercase_name, ce->name, ce->name_length);
if (zend_hash_del(CG(class_table), lowercase_name, ce->name_length+1) == FAILURE) {
efree(lowercase_name);
RETURN_FALSE;
}
efree(lowercase_name);
RETURN_TRUE;
}
Dunno if it's a good implementation, but the idea was to be able to reload classes in the looong running PHP application.
However, I don't think it should be implemented in PHP ever, 'cos this (and many more) functionality is/should be implemented in SRM.
--
Wbr,
Antony Dovgal aka tony2001
tony2001@phpclub.net || antony@dovgal.com
Well, some time ago I wrote a small functions, that does exactly the same:
PHP_FUNCTION(unregister_class)
{
snip
Barring some stylectic differences and a check to make sure it's not an
internal class, that's precisely what I gave sean in
shiva_destroy_class()....
But just to be clear, I don't think it's a good idea, and I'll wager dollars
to doughnuts that it destablizes the active request if not the entire
engine.
Not to be too blunt, but I gave him that extension so he'd have some rope to
hang himself with (or maybe learn something useful...)
-Sara
At 08:32 19/08/2004, Sara Golemon wrote:
Well, some time ago I wrote a small functions, that does exactly the same:
PHP_FUNCTION(unregister_class)
{
snipBarring some stylectic differences and a check to make sure it's not an
internal class, that's precisely what I gave sean in
shiva_destroy_class()....But just to be clear, I don't think it's a good idea, and I'll wager dollars
to doughnuts that it destablizes the active request if not the entire
engine.
From the technical perspective, it's actually easy to implement in a
rock-solid manner. Don't kill the class completely but just clean any
functions/properties inside of it, and mark it as dead. Then patch the
class declaration begin code to check for this flag, and reuse the class
structure instead of complaining about reregistering a class. There might
be some more cleanup necessary in PHP 5, but if any, it shouldn't be too
complicated.
But I agree with you that it's a bad idea :)
Zeev
From the technical perspective, it's actually easy to implement in a
rock-solid manner. Don't kill the class completely but just clean any
functions/properties inside of it, and mark it as dead. Then patch the
class declaration begin code to check for this flag, and reuse the class
structure instead of complaining about reregistering a class. There might
be some more cleanup necessary in PHP 5, but if any, it shouldn't be too
complicated.
I took this in a slightly different direction:
http://pecl.php.net/classkit Any and all thoughts appreciated.
Function Entries:
bool classkit_method_add(string classname, string methodname, string
args, string code[, long flags])
bool classkit_method_redefine(string classname, string methodname,
string args, string code[, long flags])
Note: flags parameter requires PHP >= 5.0.0
bool classkit_method_remove(string classname, string methodname)
bool classkit_method_rename(string classname, string methodname, string
newname)
Constants (passed in flags parameter):
CLASSKIT_ACC_PUBLIC
CLASSKIT_ACC_PROTECTED
CLASSKIT_ACC_PRIVATE
-Sara
Sara Golemon wrote:
I took this in a slightly different direction:
http://pecl.php.net/classkit Any and all thoughts appreciated.
Sara, seems very interesting to me. Anyway, can you please provide me
your shiva_destroy_class() extension?
I need something similar in order to reload classes for an open source
EJB like system I built, when redeploying code with already declared
classes.
Rasmus was right when he said "PHP is getting stronger in order to
provide more powerful tools to solve the same set of problems people
have been using PHP for all along", when talking about the "PHP becoming
Java".
I love PHP and I use it wherever possible, coding in PHP is fun for me;
and more features may means more new possibilities, such as that project
I built. I know that it is using PHP in a way it wasn't originally meant
for; but it works, and PHP made it possible, and if it allowed me to do
it, then it is "legal" to do such sort of things ;)
PS. For the interested ones, this is a list of current features:
- Encapsulates business logic in a Carthag Business Module (CBM)
- Delivers optional persistance, with Value Objects and Data Access Objects
- CBM deployer
- Transparent access to remote CBMs as if they were local
- Transparent access to local CBMs without using CBM server
- Location of CBMs through a CBM locator descriptor
- Own socket server for receiving requests and serving classes
- Uses XML-RPC as communication protocol
- Uses PHP internal XML-RPC server implementation for faster speed
- Watchdog for monitoring CBM server and restarting it in case of failure
- Supports CBM session for the entire lifetime of a remote object
Alex
Please excuse me for repeated messages, I got some timeouts when posting
from the news server gateway.
I took this in a slightly different direction:
http://pecl.php.net/classkit Any and all thoughts appreciated.Sara, seems very interesting to me. Anyway, can you please provide me
your shiva_destroy_class() extension?
Shiva's not anything approaching stable or trustable. It's just a five
minute exercise I wrote for Sean Coates to illustrate some of the inherent
dangers of futzing around with the class table.
If you need class redefinition, I would much sooner recommend
classkit_import() which will aggregate all the methods from a new class
definition over the top of an old class entry. (Note: It's been documented
in CVS, it just hasn't shown in the manual build yet:
http://cvs.php.net/co.php/phpdoc/en/reference/classkit/functions/classkit-import.xml )
I'll also note that while this extensions is more trustable than Shiva, it's
also marked as beta. Sean (who uses this far more than I actually do)
mentioned a bug (non-crashing, but it raises E_ERROR and halts your current
script) when classkit_import() is used under certain conditions (something
involving importing multiple classes in a single file, we're working through
the specifics still).
If you don't so-much want to redefine the class as undefine it, I suppose it
could be done safely by following the idea in Zeev's earlier post which
involves "hiding" the class rather than outright destroying it. That way if
there are any instances hiding in some obscure symbol table somewhere (e.g.
another instantiated object, a static variable hash or a local variable hash
in an upstream function/method) they'll still have a class_entry to refer
to. This is something I could add to classkit, but because the class
definitions don't really disappear but are rather just "hidden", it'll
appear to "leak" a little bit of memory each time it's used so again, I'd
recommend classkit_import() over it.
Before you ask: No, it's not "simple enough" to just go looking through
those symbol tables to see if any instances exists. Possible yes, practical
no.
-Sara
Sara Golemon wrote:
If you need class redefinition, I would much sooner recommend
classkit_import() which will aggregate all the methods from a new class
definition over the top of an old class entry. (Note: It's been
documented
Sara, thank you very much for your suggestion. This should be enough for
my needs, I'm going to study/test it. Redefining only is ok for me,
anyway having the option of removing (hiding) a class may be interesting
too.
Alex
Sara Golemon wrote:
I took this in a slightly different direction:
http://pecl.php.net/classkit Any and all thoughts appreciated.
Sara, seems very interesting to me. Anyway, can you please provide me
your shiva_destroy_class() extension?
I need something similar in order to reload classes for an open source
EJB like system I built, when redeploying code with already declared
classes.
Rasmus was right when he said "PHP is getting stronger in order to
provide more powerful tools to solve the same set of problems people
have been using PHP for all along", when talking about the "PHP becoming
Java".
I love PHP and I use it wherever possible, coding in PHP is fun for me;
and more features may means more new possibilities, such as that project
I built. I know that it is using PHP in a way it wasn't originally meant
for; but it works, and PHP made it possible, and if it allowed me to do
it, then it is "legal" to do such sort of things ;)
PS. For the interested ones, this is a list of current features:
- Encapsulates business logic in a Carthag Business Module (CBM)
- Delivers optional persistance, with Value Objects and Data Access Objects
- CBM deployer
- Transparent access to remote CBMs as if they were local
- Transparent access to local CBMs without using CBM server
- Location of CBMs through a CBM locator descriptor
- Own socket server for receiving requests and serving classes
- Uses XML-RPC as communication protocol
- Uses PHP internal XML-RPC server implementation for faster speed
- Watchdog for monitoring CBM server and restarting it in case of failure
- Supports CBM session for the entire lifetime of a remote object
Alex
Sara Golemon wrote:
I took this in a slightly different direction:
http://pecl.php.net/classkit Any and all thoughts appreciated.
Sara, seems very interesting to me. Anyway, can you please provide me
your shiva_destroy_class() extension?
I need something similar in order to reload classes for an open source
EJB like system I built, when redeploying code with already declared
classes.
Rasmus was right when he said "PHP is getting stronger in order to
provide more powerful tools to solve the same set of problems people
have been using PHP for all along", when talking about the "PHP becoming
Java".
I love PHP and I use it wherever possible, coding in PHP is fun for me;
and more features may means more new possibilities, such as that project
I built. I know that it is using PHP in a way it wasn't originally meant
for; but it works, and PHP made it possible, and if it allowed me to do
it, then it is "legal" to do such sort of things ;)
PS. For the interested ones, this is a list of current features:
- Encapsulates business logic in a Carthag Business Module (CBM)
- Delivers optional persistance, with Value Objects and Data Access Objects
- CBM deployer
- Transparent access to remote CBMs as if they were local
- Transparent access to local CBMs without using CBM server
- Location of CBMs through a CBM locator descriptor
- Own socket server for receiving requests and serving classes
- Uses XML-RPC as communication protocol
- Uses PHP internal XML-RPC server implementation for faster speed
- Watchdog for monitoring CBM server and restarting it in case of failure
- Supports CBM session for the entire lifetime of a remote object
Alex