Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:91654
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: pass (pb1.pair.com: domain gmail.com designates 209.85.161.178 as permitted sender)
MIME-Version: 1.0
In-Reply-To: <56E6A761.8070005@zend.com>
References: <BY2PR0201MB1784EE8F89DDDF36FC01328CBF880@BY2PR0201MB1784.namprd02.prod.outlook.com>
	<CAF+90c8NDPio+8iTBXavJ5K945fNh9=uVccLwm5gK+YDZZsVDQ@mail.gmail.com>
	<56E6A761.8070005@zend.com>
Date: Mon, 14 Mar 2016 13:23:02 +0100
Message-ID: <CAF+90c-6kBysspifAZRsgq+E_V_srUM3ZWDoEG5+uu1Txpjj9Q@mail.gmail.com>
To: Dmitry Stogov <dmitry@zend.com>
Cc: Xinchen Hui <xinchen.h@zend.com>, Bob Weinand <bobwei9@hotmail.com>, 
	"internals@lists.php.net" <internals@lists.php.net>
Content-Type: multipart/alternative; boundary=94eb2c0356ec8c4601052e0157cd
Subject: Re: Remove useless opcode handlers
From: nikita.ppv@gmail.com (Nikita Popov)

--94eb2c0356ec8c4601052e0157cd
Content-Type: text/plain; charset=UTF-8

On Mon, Mar 14, 2016 at 12:58 PM, Dmitry Stogov <dmitry@zend.com> wrote:

>
>
> On 03/14/2016 02:29 PM, Nikita Popov wrote:
>
> On Mon, Mar 14, 2016 at 11:56 AM, Dmitry Stogov <dmitry@zend.com> wrote:
>
>> Please take a quick look and let me know if you see any problems.
>>
>>
>> https://gist.github.com/dstogov/23cc318dd3e411904e10
>>
>>
>> I'm going to commit this tomorrow.
>>
>>
>> Thanks. Dmitry.
>>
> ZEND_ADD is not commutative for array operands.
>
>
> Unfortunately, you are right :(
>
> ZEND_MUL may not be commutative if operator overloading is involved (e.g.
> matrix multiplication is not abelian).
>
>
> Right again.
>
> The NO_CONST_CONST cases are problematic: Opcache currently, in my eyes
> incorrectly, evaluates constant expressions even if they throw notices
> (currently I think mostly by suppressing them). If we do not evaluate
> these, we may end up with CONST_CONST operations. (This is not a huge
> problem just now because constant folding in opcache is limited -- however
> with full constant propagation this leads to many test failures.)
>
>
> The patch adds NO_CONST_CONST handling to opcache. All tests are passed, I
> also didn't see any problems running ~10 real-life apps.
>
> The opcodes affected by the last point (limited to those included in your
> patch) are SUB, MUL, POW, CONCAT (array operands) and BW_NOT (various).
> However with Andrea's invalid numeric string RFC more will fall in this
> category (evaluable but throwing).
>
>
> I didn't get what is wrong with SUB, MUL, etc, because they are evaluated
> at compile-time
> Anyway, I see - the patch can't be committed in current state.
> Thanks for review.
>

What I mean is, for example:

var_dump(~(bool)true);
echo "Done\n";

Will throw an exception without opcache, will result in no output (and no
*visible* error) with opcache. My point here is that const-operand-only
cases might be necessary to handle error cases like this.

Nikita

--94eb2c0356ec8c4601052e0157cd--