Hi internals!
The performance optimizations to func_get_args that were introduced in PHP
5.5.6 break BC for call_user_func_array on functions with by-ref args, see
https://bugs.php.net/bug.php?id=66107.
While using that kind of code is wrong in the first place, this seems like
a nasty break for a bugfix release. Maybe revert this for 5.5 and only keep
in 5.6?
Nikita
Hi,
I agree that the change makes BC break for some tricky code :(
and i don't insist to keep it in 5.5.
Thanks. Dmitry.
On Mon, Dec 9, 2013 at 4:26 AM, Stas Malyshev smalyshev@sugarcrm.comwrote:
Hi!
Can we do anything to fix it? Is the problem limited only to
pass-object-by-ref or also for other by-ref scenarios? In the latter
case it does constitute a BC break I think and it may be a good idea to
move it to 5.6...-------- Original Message --------
Subject: [PHP-DEV] func_get_args BC break
Date: Sat, 7 Dec 2013 06:15:51 -0600
From: Nikita Popov nikita.ppv@gmail.com
To: PHP internals internals@lists.php.netHi internals!
The performance optimizations to func_get_args that were introduced in PHP
5.5.6 break BC for call_user_func_array on functions with by-ref args, see
https://bugs.php.net/bug.php?id=66107.While using that kind of code is wrong in the first place, this seems like
a nasty break for a bugfix release. Maybe revert this for 5.5 and only keep
in 5.6?Nikita
Hi,
I agree that the change makes BC break for some tricky code :(
and i don't insist to keep it in 5.5.Thanks. Dmitry.
On Mon, Dec 9, 2013 at 4:26 AM, Stas Malyshev <smalyshev@sugarcrm.com
wrote:
Hi!
Can we do anything to fix it? Is the problem limited only to
pass-object-by-ref or also for other by-ref scenarios? In the latter
case it does constitute a BC break I think and it may be a good idea to
move it to 5.6...
Hi,
as we agreed that this is a BC break, I don't think that it is a good idea
even for 5.6.
could we also create a test for this behavior, so we don't change this
unintentionally later on?
--
Ferenc Kovács
@Tyr43l - http://tyrael.hu
I'm completely disagree.
That PHP code (in bug report) worked not by design.
Actually, it's wrong code that worked just because of luck (zval refcount
became 1).
I don't see any reason to keep such compatibility for 5.6 in cost of
performance.
Thanks. Dmitry.
Hi,
I agree that the change makes BC break for some tricky code :(
and i don't insist to keep it in 5.5.Thanks. Dmitry.
On Mon, Dec 9, 2013 at 4:26 AM, Stas Malyshev <smalyshev@sugarcrm.com
wrote:
Hi!
Can we do anything to fix it? Is the problem limited only to
pass-object-by-ref or also for other by-ref scenarios? In the latter
case it does constitute a BC break I think and it may be a good idea to
move it to 5.6...Hi,
as we agreed that this is a BC break, I don't think that it is a good idea
even for 5.6.
could we also create a test for this behavior, so we don't change this
unintentionally later on?--
Ferenc Kovács
@Tyr43l - http://tyrael.hu
Hi,
I agree that the change makes BC break for some tricky code :(
and i don't insist to keep it in 5.5.
An old enough bug may become a feature.
I am not sure it should be kept for 5.6 either. As of the last "dramatic"
performance loss (about some string funcs), what are you referencing to
here?
Cheers,
Pierre
Hi Pierre,
It was a performance improvement that caused that BC break.
For some reason func_get_arg() made a copy of each argument.
In case some argument is a big array it made a significant useless work
creating and filling new HashTable.
Thanks. Dmitry.
Hi,
I agree that the change makes BC break for some tricky code :(
and i don't insist to keep it in 5.5.An old enough bug may become a feature.
I am not sure it should be kept for 5.6 either. As of the last "dramatic"
performance loss (about some string funcs), what are you referencing to
here?Cheers,
Pierre