Everything

php

About assert()

php.internals

Pedro Lacerda

8 years ago

Hi developers,

Trying to resolve the bug #75950 (that after long hours I found that I
couldn't reproduce), I observed that if `zend.assertions >= 0` the
generated code inside `assert()` was indeed executed even if `assert.active
= off`. Naturally the function arguments were evaluated before entering
into the `assert()` function.

https://bugs.php.net/bug.php?id=75950

The point is that will be possible to fully disable assertions setting
`assert.active = false` after initialization if we turn `assert` into a
statement. This can mantain fully backward compatibility. Calling
`assert_options(ASSERT_ACTIVE, true)` or `false` after initialization would
fully enable or disable the functionality. It seems the most sensible thing
to do.

By the way `assert` in Java and Python is an statement, and in C it isn't a
function.

So my question is what is the purpose of mantaining `assert()` a function,
there are any drawbacks of fully disabling it when `zend.assertions >= 0`?


PS: Strange that nobody cared in recent emails about the proposal to a
small increase of the testing coverage by doubling the testing time
duration, was somewhat interesting.

--
Atenciosamente,
Pedro Lacerda

Michael Morris

8 years ago

On Wed, Feb 14, 2018 at 1:16 AM Pedro Lacerda <pslacerda@gmail.com> wrote:

> Hi developers,
>
> Trying to resolve the bug #75950 (that after long hours I found that I
> couldn't reproduce), I observed that if `zend.assertions >= 0` the
> generated code inside `assert()` was indeed executed even if `assert.active
> = off`. Naturally the function arguments were evaluated before entering
> into the `assert()` function.
>
> https://bugs.php.net/bug.php?id=75950
>
> The point is that will be possible to fully disable assertions setting
> `assert.active = false` after initialization if we turn `assert` into a
> statement. This can mantain fully backward compatibility. Calling
> `assert_options(ASSERT_ACTIVE, true)` or `false` after initialization would
> fully enable or disable the functionality. It seems the most sensible thing
> to do.
>
> By the way `assert` in Java and Python is an statement, and in C it isn't a
> function.
>
> So my question is what is the purpose of mantaining `assert()` a function,
> there are any drawbacks of fully disabling it when `zend.assertions >= 0`?
>
>
> PS: Strange that nobody cared in recent emails about the proposal to a
> small increase of the testing coverage by doubling the testing time
> duration, was somewhat interesting.
>

Assert.active is the legacy PHP 5 control. Don’t use it.

zend.assertions -1 has the behavior you want. 0 emulates PHP 5 and
earlier’s broken implementation of assert. If you don’t have legacy
software to babysit, don’t use it.

Pedro Lacerda

8 years ago

The same beharviour, but `assert` as statement also uses 1 character less.

Em 14 de fev de 2018 10:13 AM, "Michael Morris" <tendoaki@gmail.com>
escreveu:

On Wed, Feb 14, 2018 at 1:16 AM Pedro Lacerda <pslacerda@gmail.com> wrote:

> Hi developers,
>
> Trying to resolve the bug #75950 (that after long hours I found that I
> couldn't reproduce), I observed that if `zend.assertions >= 0` the
> generated code inside `assert()` was indeed executed even if `assert.active
> = off`. Naturally the function arguments were evaluated before entering
> into the `assert()` function.
>
> https://bugs.php.net/bug.php?id=75950
>
> The point is that will be possible to fully disable assertions setting
> `assert.active = false` after initialization if we turn `assert` into a
> statement. This can mantain fully backward compatibility. Calling
> `assert_options(ASSERT_ACTIVE, true)` or `false` after initialization would
> fully enable or disable the functionality. It seems the most sensible thing
> to do.
>
> By the way `assert` in Java and Python is an statement, and in C it isn't a
> function.
>
> So my question is what is the purpose of mantaining `assert()` a function,
> there are any drawbacks of fully disabling it when `zend.assertions >= 0`?
>
>
> PS: Strange that nobody cared in recent emails about the proposal to a
> small increase of the testing coverage by doubling the testing time
> duration, was somewhat interesting.
>

Assert.active is the legacy PHP 5 control. Don’t use it.

zend.assertions -1 has the behavior you want. 0 emulates PHP 5 and
earlier’s broken implementation of assert. If you don’t have legacy
software to babysit, don’t use it.

Michael Morris

8 years ago

My earlier reply was through my phone. I've since read the bug report and
saw your note where you couldn't reproduce it.

Assert is a statement in PHP 7.  Try this code

function foo() {
debug_print_backtrace();
return true;
}

assert(foo());

In PHP 7 only foo() will be in the backtrace.  In PHP 5 assert will be
listed as well.

What you are proposing has already been done.  It's also why
zend.assertions has three settings: 1: on, -1: off, 0: emulate PHP 5.
Unless you have code on your hands that is breaking when the -1 setting is
used, the 0 setting should never be used.

On Wed, Feb 14, 2018 at 7:43 AM, Pedro Lacerda <pslacerda@gmail.com> wrote:

Niklas Keller

8 years ago

> In PHP 7 only foo() will be in the backtrace.  In PHP 5 assert will be
> listed as well.
>
> What you are proposing has already been done.  It's also why
> zend.assertions has three settings: 1: on, -1: off, 0: emulate PHP 5.
> Unless you have code on your hands that is breaking when the -1 setting is
> used, the 0 setting should never be used.

I don't think that's true. 0 can be used to dynamically switch
assertions on and off, which can be useful in some applications.

Regards, Niklas

Pedro Lacerda

8 years ago

2018-02-14 12:40 GMT-03:00 Michael Morris <tendoaki@gmail.com>:

> Assert is a statement in PHP 7.  Try this code
>
> function foo() {
>   debug_print_backtrace();
>   return true;
> }
>
> assert(foo());
>
> In PHP 7 only foo() will be in the backtrace.  In PHP 5 assert will be
> listed as well.
>

I saw all that `PHP_FUNCTION` and `ZEND_PARSE_PARAMETERS` so thought that
was just a regular exposed function with compilation toggled by setting.
Anyway doing it directly in the lexer would avoid the argument parsing step
(and turning parentesis optional).


2018-02-14 13:29 GMT-03:00 Niklas Keller <me@kelunik.com>:

>
> I don't think that's true. 0 can be used to dynamically switch
> assertions on and off, which can be useful in some applications.
>

This can be useful for testing frameworks which does contract programming
by assertions but only in development.

Atenciosamente,
Pedro