[RFC] opcache.no_cache: Opcache optimization without any caching

3 years ago by tyson andre — view source

unread

Hi internals,

I've created the RFC https://wiki.php.net/rfc/opcache.no_cache to make the opcode optimizer and JIT available without opcode caching, through a new setting opcache.no_cache=1.

So far, the feedback I've gotten is that having the ability to optimize without caching would be useful,
but that moving the optimizer into Zend is possibly a better solution.
(https://github.com/php/php-src/pull/5504 ,
announced two weeks ago in https://externals.io/message/109959
"opcache.no_cache prototype: Opcode optimization without caching")

To summarize my reasons for creating the RFC without moving the optimizer:

Even with the optimizer moved into PHP's core,
I still think there'd be the same use case for a system ini configuration setting to disable caching.
(in designs I can imagine)
I consider this useful functionality to have, so I'd rather have a good solution than an ideal solution
that currently isn't being worked on (to my knowledge)
I'd prefer creating an RFC with small changes here.
The discussions would be useful for future RFCs in this area.

If a viable RFC&implementation to move opcache into core is created
(and would conflict with this), I'll postpone this RFC.

Thanks,

Tyson

3 years ago by Derick Rethans — view source

unread

I've created the RFC https://wiki.php.net/rfc/opcache.no_cache to make
the opcode optimizer and JIT available without opcode caching, through
a new setting opcache.no_cache=1.

In my opinion, configuration names should never be "in the negative".
Instead of defaulting to opcache.no_cache=0, default to opcache.cache=1.

cheers,
Derick

--
PHP 7.4 Release Manager
Host of PHP Internals News: https://phpinternals.news
Like Xdebug? Consider supporting me: https://xdebug.org/support
https://derickrethans.nl | https://xdebug.org | https://dram.io
twitter: @derickr and @xdebug

3 years ago by tyson andre — view source

unread

In my opinion, configuration names should never be "in the negative".
Instead of defaulting to opcache.no_cache=0, default to opcache.cache=1.

I see your point - there are only a few negative ini names I see in phpinfo() (disable_functions, disable_cache), and those are for lists of strings.

There's a mix of on by default and off by default in
zend.enable_gc, opcache.enable, opcache.enable_cli, opcache.enable_file_override,
enable_dl, and enable_post_data_reading.

opcache.enable_cache=1 as the default would probably be more consistent with those.
But the setting isn't actually enabling cache, it's allowing cache (i.e. opposite of disabling),
so making this "opcache.allow_cache" might be the best. I'll probably go with that.
(similar to allow_url_fopen, allow_url_include)

opcache.cache could be confused with the caching strategy or the path to the cache,
or being sufficient on its own to enable caching

Thanks,

Tyson

3 years ago by tyson andre — view source

unread

Hi internals,

https://wiki.php.net/rfc/opcache.no_cache has been updated.
Instead of opcache.no_cache=1, opcache.enable_cache=0 is now used to disable caching.

In the RFC, I wrote that opcache_get_status() now includes the following new booleans

optimizations_enabled, which is true if any optimization passes will get run.
allow_cache, which will be true if any form of opcode caching is enabled.
Maybe cache_opcodes or opcode_cache_enabled would be a better name.

Any thoughts on whether opcache_get_status() should be changed at all here?
It's probably possible but inconvenient to determine what's being done with extension_loaded() and multiple calls to ini_get(),
and the intended meaning of existing flags such as opcache_enabled was never documented
in https://www.php.net/opcache_get_status

Any other thoughts on the RFC?

In my opinion, configuration names should never be "in the negative".
> Instead of defaulting to opcache.no_cache=0, default to opcache.cache=1.

... the setting isn't actually enabling cache, it's allowing cache (i.e. opposite of disabling),
so making this "opcache.allow_cache" might be the best. I'll probably go with that.
(similar to allow_url_fopen, allow_url_include)

... opcache.cache could be confused with the caching strategy or the path to the cache,
or being sufficient on its own to enable caching

Thanks,

Tyson

3 years ago by Albert Casademont — view source

unread

Hi tyson,

Thanks for your RFC. As ReactPHP users this is a very welcome change. Right
now we have to enable file_cache only for the optimizations, as there's no
real benefit of the full opcache in long-running CLI apps. Having this
setting would allow us to even stop using file_cache but get the opcache
optimizations. It would be even better if these optimizations moved into
PHP Core.

Thank you!

Albert

On Thu, May 21, 2020 at 4:07 PM tyson andre tysonandre775@hotmail.com
wrote:

Hi internals,

https://wiki.php.net/rfc/opcache.no_cache has been updated.
Instead of opcache.no_cache=1, opcache.enable_cache=0 is now used to
disable caching.

In the RFC, I wrote that opcache_get_status() now includes the following
new booleans

optimizations_enabled, which is true if any optimization passes will
get run.

allow_cache, which will be true if any form of opcode caching is
enabled.
Maybe cache_opcodes or opcode_cache_enabled would be a better name.

Any thoughts on whether opcache_get_status() should be changed at all here?
It's probably possible but inconvenient to determine what's being done
with extension_loaded() and multiple calls to ini_get(),
and the intended meaning of existing flags such as opcache_enabled was
never documented
in https://www.php.net/opcache_get_status

Any other thoughts on the RFC?

In my opinion, configuration names should never be "in the negative".
Instead of defaulting to opcache.no_cache=0, default to
opcache.cache=1.

... the setting isn't actually enabling cache, it's allowing cache (i.e.
opposite of disabling),
so making this "opcache.allow_cache" might be the best. I'll probably go
with that.
(similar to allow_url_fopen, allow_url_include)

... opcache.cache could be confused with the caching strategy or the
path to the cache,
or being sufficient on its own to enable caching

Thanks,

Tyson

3 years ago by tyson andre — view source

unread

Hi internals,

If there are no unexpected issues,
then voting will start on https://wiki.php.net/rfc/opcache.no_cache in 2 days, on May 30th.

Thanks,

Tyson