this can happen when the config for the router is unset, but this method
does not need to depend on routing. reading an unset config would raise an exception.
Commits
-------
cd24fb8 change explode's limit parameter based on known variable content
b3cc270 minor optimalisations for explode
Discussion
----------
[FrameworkBundle][CssSelector][HttpFoundation][HttpKernel] [Security][Validator] Minor optimizations for "explode" function
Bug fix: no
Feature addition: no
Backwards compatibility break: no
Symfony2 tests pass: yes
Fixes the following tickets: -
Todo: -
I added limit parameter in some places, where it may be usefull. I did not check the context of what values may have been exploded. So to not break anything, I added +1 to limit parameter.
If you find out that in some places limit (or limit+1) is not important or meaningless, write a comment please and I will fix it.
---------------------------------------------------------------------------
by fabpot at 2011/12/07 06:56:49 -0800
Adding +1 just to be sure to not break anything is clearly something we won't do. What is the benefit of doing that anyway?
---------------------------------------------------------------------------
by pulzarraider at 2011/12/07 13:50:24 -0800
The main idea of making this PR was to notify about some places that may run faster with just adding one parameter to explode function.
If in code is someting like: ```list($a, $b) = explode(':', $s);```
Function ```explode``` will create n-items (depends on ```$s```), but we need in code only the first two items. There is no reason to let ```explode``` create more items in memory that are NEVER used in our code. The limit parameter is there for these situations, so let's use it.
I know that it is microoptimization and may look unimportant, but we are writing a framework - so people expect that code will be as fast as possible without this kind of mistakes.
As I've noticed above, I know that +1 is not ideal solution, but the fastest without debugging the code. I expect that someone (with good knowledge of that code) will look at it and write in comments if variable may contain 1 comma (dot or someting on what is doing the explode) or maybe 2 in some situations or more.
Anyway, +1 will not break anything, because same items are created as it is now, but no unnecessary item is created.
---------------------------------------------------------------------------
by fabpot at 2011/12/07 23:14:59 -0800
I'm +1 for adding the number to avoid problems but I'm -1 on the optimization side of things as it won't optimize anything.
---------------------------------------------------------------------------
by helmer at 2011/12/08 12:46:49 -0800
*.. The main idea of making this PR was to notify about some places that **may** run faster ..*
I am also unsure the optimization is really an optimization, care to benchmark (with meaningful inputs)? As for the limit+1 thing, why would you want to +1 it? The number of ``list`` arguments should always reflect the ``limit`` parameter, no?
---------------------------------------------------------------------------
by pulzarraider at 2011/12/08 23:11:34 -0800
@helmer please try this simple benchmark:
```
<?php
header('Content-Type: text/plain; charset=UTF-8');
define('COUNT', 10000);
$source_string = 'aaaaaaaaaaaaaaaaaaaa:bbbbbbbbbbbbbbbbbbbbb:cccccccccccccccccccccccc:dddddddddddddddddddddd:eeeeeeeeeeeeeeeeeeeeeeeee:fffffffffffffffffffffffffff';
$start = microtime(true);
for ($i = 0; $i < COUNT; $i++) {
list($a, $b) = explode(':', $source_string);
}
$end = microtime(true)-$start;
echo 'without limit: '.$end."\n";
$start = microtime(true);
for ($i = 0; $i < COUNT; $i++) {
list($a, $b) = explode(':', $source_string, 2);
}
$end = microtime(true)-$start;
echo 'with limit: '.$end."\n";
```
My results are:
```
without limit: 0.057228803634644
with limit: 0.028676986694336
```
That is 50% difference (with APC enabled). Of course the result depends on the length of source string and if it's too short, the difference may be none or very very small. That's why I said, that it **may** run faster and is just a micro optimization.
---------------------------------------------------------------------------
by pulzarraider at 2011/12/08 23:18:12 -0800
@helmer And why +1? It depends on a code:
```
$source_string = 'aaaaaaaaaaaaaaaaaaaa:bbbbbbbbbbbbbbbbbbbbb:cccccccccccccccccccccccc';
list($a, $b) = explode(':', $source_string, 2);
var_dump($a, $b);
```
and
```
$source_string = 'aaaaaaaaaaaaaaaaaaaa:bbbbbbbbbbbbbbbbbbbbb:cccccccccccccccccccccccc';
list($a, $b) = explode(':', $source_string, 3);
var_dump($a, $b);
```
gives different results. That's why the content of the variable must be known.
---------------------------------------------------------------------------
by helmer at 2011/12/09 00:08:28 -0800
@pulzarraider Thanks for the benchmark, seems like a gain enough. Although, we are more likely having a scenario of:
``explode(':', 'a🅱️c')`` vs ``explode(':', 'a🅱️c', 3)`` with a ``COUNT`` of 10, where the difference is not even in microseconds anymore :)
The limit addition alters the behaviour though, ie suddenly you can define a controller [logical name](http://symfony.com/doc/current/book/routing.html#controller-string-syntax) as ´´AcmeBlogBundle:Blog:show:something``, and things go downhill from there on.
All that aside, I'm +1 for setting the limit to the exact number of ``list`` parameters, but certainly not number+1, this is just too wtfy (as you said, this was a safety thing, but I reckon for this PR to be merged it needs to be +0).
---------------------------------------------------------------------------
by drak at 2011/12/09 08:28:58 -0800
Overall `list()` is ugly as it's not very explicit. Even though it would mean extra lines, it's better to `explode()` then explicitly assign variables:
```
$parts = explode(':', $foo);
$name = $parts[0];
$tel = $parts[1];
```
`list()` is one of those bad relics from the PHP past...
---------------------------------------------------------------------------
by fabpot at 2011/12/11 10:07:47 -0800
@drak: why is `list` not explicit? It is in fact as explicit as the more verbose syntax you propose.
---------------------------------------------------------------------------
by pulzarraider at 2011/12/11 13:08:50 -0800
@drak: I agree with @fabpot. In speech of benchmarks ```list``` is faster then using a helper variable.
@fabpot, @helmer I've changed explode's limit to be correct (without +1) and removed some changes from this PR, where I can't find out what the content of variable may be. Unit tests pass, so I think it's ready for merge.
Builds upon aead4a9836180cabae4d47fe27c634dcd79ac8f2, which prematurely removed request scoping from the assets templating helper in all cases. The helper need only be request-scoped if one or more request-scoped packages (e.g. PathPackages) are injected into it. This change makes it possible to utilize the assets helper outside of a request (e.g. during a console script).
To ensure that the assets helper is not assigned a request scope, all asset base URL's must be defined for all packages (default and any named) and both protocols: HTTP and SSL. The included test config fixtures concisely accomplish this by specifying a single HTTPS URL as the base URL for our default and named package, since FrameworkExtension's Configuration conveniently registers this URL for both protocols.
when esi is enabled and internal uris are generated for esi-tags, an
attribute-array consisting entirely of null-values isn't handled correctly.
The reason is that php's `http_build_query()`-method outputs an empty string
for such arrays:
http_build_query(array('foo' => '')) == 'foo='
http_build_query(array('foo' => null)) == ''
In the latter case, the generation of an URI in `HttpKernel::generateInternalUri()`
generates an URI that could not be matched by the corresponding route (ex.
`_internal/Controller/.html` opposed to `_internal/Controller/none.html` which
should be expected).
This commit adds a possible solution as well as a simple test for this issue.
Commits
-------
9bcce9f fix tests
fc4787a fix non-extensible router
Discussion
----------
Router fix
Right now, the router is hard to overwrite (you need always a compiler pass). This commit fixes this.
---------------------------------------------------------------------------
by fabpot at 2011/07/18 01:15:36 -0700
Why do you need a complier pass to override the router?
---------------------------------------------------------------------------
by schmittjoh at 2011/07/18 01:47:47 -0700
How would you suggest to overwrite it?
Basically, I want to do something like this:
```yml
services:
router:
parent: router.default
class: MyClass
calls:
- [moreDeps, []]
```
---------------------------------------------------------------------------
by Seldaek at 2011/07/18 05:07:19 -0700
Then maybe we should somehow support redefining services with the same name while keeping the old one as parent, otherwise we need this foo.default for every service out there?
---------------------------------------------------------------------------
by fabpot at 2011/07/18 06:30:34 -0700
as @Seldeak said, why do that for the router and not all services?
---------------------------------------------------------------------------
by schmittjoh at 2011/07/18 06:38:39 -0700
I have designed the SecurityBundle this way where extension is encouraged.
---------------------------------------------------------------------------
by schmittjoh at 2011/07/18 11:15:57 -0700
I should add that this is mainly a problem for services where you still want to use the semantic configuration that is provided by the bundle. For services which are not configured by the extension, this is not so much of an issue.
Anyway, if you don't want to merge it, just close the PR. I have no problem with using a compiler pass.
---------------------------------------------------------------------------
by fabpot at 2011/07/18 11:55:11 -0700
We already have such a case with translator and translator.real. I will review the existing services to see where it makes sense to implement the same strategy.
---------------------------------------------------------------------------
by Seldaek at 2011/07/18 12:20:55 -0700
I guess you'd do it anyway, but we should pick a winner between .real and .default
---------------------------------------------------------------------------
by lsmith77 at 2011/07/18 12:26:52 -0700
I would prefer ".default" as ".real" always confused me.
