be careful: the built phar file will cause issues with the new opcache that's bundled with 5.5 and later due to this issue) in opcache.
A workaround for the issue will be provided at a later time, but until then, per application server (Apache, FPM master) only one version of the file might exist.
Sacy turns
{asset_compile}
<link type="text/css" rel="stylesheet" href="/styles/file1.css" />
<link type="text/x-sass" rel="stylesheet" href="/styles/file2.sass" />
<link type="text/x-scss" rel="stylesheet" href="/styles/file3.scss" />
<link type="text/x-less" rel="stylesheet" href="/styles/file4.less" />
<script type="text/javascript" src="/jslib/file1.js"></script>
<script type="text/coffeescript" src="/jslib/file2.coffee"></script>
<script type="text/javascript" src="/jslib/file3.js"></script>
<script type="text/x-eco" src="/jslib/sometemplate.eco"></script>
<script type="text/x-jsx" src="/jslib/somecomponent.jsx"></script>
{/asset_compile}
into
<link type="text/css" rel="stylesheet" href="/assetcache/many-files-1234abc.css" />
<script type="text/javascript" src="/assetcache/many-files-abc123.js"></script>
It's a well-known fact that every request for an asset on your HTML page increases the loading time, as even with unlimited bandwidth, there's increased latency for making the request, waiting for the server to process it and to send the data back.
So by now it has become good practice to, in production mode, send out all assets (assets being JavaScript and CSS in the course of this description) together in one big file.
Various approaches have been used for this so far, the most common ones are these:
-
Using some sort of Makefile or similar process, collect the assets, combine them and then link them. This works very nicely, but the additional deployment step is easily forgotten and even if you recompile the assets, old versions of the compiled file could (and should!) be stored in the clients cache.
-
Serving all assets together using a specially created gateway script. This allows you to forego the compilation process, but you have to be very careful not to break client side caching when you use any server side scripting language.
Additionally, by serving assets via your application server, you tie up web server processes much better used for handling dynamic data.
Both solutions also require you to keep them in mind during development: The first one requires you to check for the existence of the compiled file and serve that or the individual files while the second one forces you to do some kind of registry to notify the central asset serving script about the whereabouts of the file.
Smarty Asset Compiler (sacy) is (as the name suggests) a Plugin for the widely used PHP templating engine Smarty (sacy works in both Smarty2 and Smarty3) that provides a fresh approach and solves (nearly) all problems with the traditional solutions.
Let's assume that you have an HTML header that looks like this:
<html>
<head>
<title>Title</title>
<link type="text/css" rel="stylesheet" href="/styles/file1.css" />
<link type="text/css" rel="stylesheet" href="/styles/file2.css" />
<link type="text/css" rel="stylesheet" href="/styles/file3.css" />
<link type="text/css" rel="stylesheet" href="/styles/file4.css" />
<script type="text/javascript" src="/jslib/file1.js"></script>
<script type="text/javascript" src="/jslib/file2.js"></script>
<script type="text/javascript" src="/jslib/file3.js"></script>
</head>
<body>
content
</body>
</html>
and let's further assume that these CSS files exists and you want to have them compiled into one, if possible.
If so, place the sacy plugin file in your Smarty plugin folder and change your HTML to look like this:
<html>
<head>
<title>Title</title>
{asset_compile}
<link type="text/css" rel="stylesheet" href="/styles/file1.css" />
<link type="text/css" rel="stylesheet" href="/styles/file2.css" />
<link type="text/css" rel="stylesheet" href="/styles/file3.css" />
<link type="text/css" rel="stylesheet" href="/styles/file4.css" />
<script type="text/javascript" src="/jslib/file1.js"></script>
<script type="text/javascript" src="/jslib/file2.js"></script>
<script type="text/javascript" src="/jslib/file3.js"></script>
{/asset_compile}
</head>
<body>
content
</body>
</html>
At request time, sacy will parse the content of the block, extract the CSS links and script tags sourcing files from the same server, and check whether it has already cached the compiled version.
If not, it'll create one directly on the file system. Note that this process can take a long time as all sourced files are Minified (using jsMin or uglyfier for JavaScript and Minify for CSS) before being stored.
At any rate, it'll remove the old <link>
- and <script>
-tags and add new
ones, so that your HTML will look like this:
<html>
<head>
<title>Title</title>
<link type="text/css" rel="stylesheet" href="/assetcache/file1-file2-file3-file4-abc1234def12345.css" />
<script type="text/javascript" src="/assetcache/file1-file2-file3-deadbeef1234.js"></script>
</head>
<body>
content
</body>
</html>
sacy takes into account the media attribute for CSS files and only groups links together if they share the same media attribute. The reason for this is that you probably do not want your print-style intermixed with your screen style.
This also means though, that to achieve the optimum performance, you should group links with the same media attribute together:
{asset_compile}
<link type="text/css" media="screen" rel="stylesheet" href="/styles/file1.css" />
<link type="text/css" media="print" rel="stylesheet" href="/styles/file2.css" />
<link type="text/css" media="screen" rel="stylesheet" href="/styles/file3.css" />
<link type="text/css" media="screen" rel="stylesheet" href="/styles/file4.css" />
{/asset_compile}
will produce
<link type="text/css" media="screen" rel="stylesheet" href="/csscache/file1-hash.css" />
<link type="text/css" media="print" rel="stylesheet" href="/csscache/file2-hash.css" />
<link type="text/css" media="screen" rel="stylesheet" href="/csscache/file2-file3-hash.css" />
whereas a bit of reordering to
{asset_compile}
<link type="text/css" media="screen" rel="stylesheet" href="/styles/file1.css" />
<link type="text/css" media="screen" rel="stylesheet" href="/styles/file3.css" />
<link type="text/css" media="screen" rel="stylesheet" href="/styles/file4.css" />
<link type="text/css" media="print" rel="stylesheet" href="/styles/file2.css" />
{/asset_compile}
will cause only two links to be created:
<link type="text/css" media="screen" rel="stylesheet" href="/csscache/file1-file3-file4-hash.css" />
<link type="text/css" media="print" rel="stylesheet" href="/csscache/file2-hash.css" />
Same, by the way, goes for intermixing css and javascript references because it's possible that javascript code might depend on something that was set in an earlier CSS file.
So for maximum efficiency, group javascript and CSS file together and don't switch media types around between two link tags.
sacy will ignore all referenced resources if the given URL-string contains a scheme and/or a host name.
While PHP's built-in support for networking protocols would allow for the handling of remote files, in most of the cases this is not what the user expects (think "ad-tracking code"). If you need this feature, you will have to patch sacy accordingly, but keep in mind that checking the last-modified-date of remote resources is very costly and possibly inaccurate.
Also, remote resources can change at will, which would cause the whole cache file to be regenerated way too often.
By now, you will probably have heard of either less or sass and you might want to use them for your projects as they really go a long ways in making your life easier.
Sacy has built-in support to do such transformations if it finds (see below) lessphp or PHamlP for CSS or coffeescript-php loaded or you provide it with paths to the respective official tools (also, see below)
Inside an {asset_compile}
tag, just link to these files, or write inline code
like so:
{asset_compile}
<link type="text/x-sass" rel="stylesheet" href="/styles/style.sass" />
<link type="text/x-scss" rel="stylesheet" href="/styles/style.scss" />
<link type="text/x-less" rel="stylesheet" href="/styles/style.less" />
<script type="text/coffeescript" src="/jslib/file.coffee"></script>
<script type="text/coffeescript">
fun = (a)->
alert "Hello #{a}"
fun "World"
</script>
{/asset_compile}
Sacy uses the mime-types you provide with the type attribute to invoke the correct transformer before writing the file to the cache.
If you are translating .sass and .scss files, you can use the merge_tags
parameter to the ´{asset_compile}´ tag. If that is set, sacy will first merge
all sass/scss files (of the same type) together and only then transform them
to css.
This means that variables and mixins defined in one file will be available in another file. It also means that sacy has to do a bit of magic when assembling the load path, so be prepared for surprising results.
In general, it's not good practice to use inline script
- or style
-tags, so
it's recommended you use this feature (as of 0.4) of sacy sparingly. It can
still be useful though if you have to pass data from smarty through to the JS
layer and you'd like to use coffee script for that too.
You would also get free minification of course.
See below for some information about dependencies and how to bundle them.
-
ease of use: with sacy, you do not have to change your templates to make the caching work. It'll work in the background and do its stuff
-
correct caching: Your compiled CSS is written to the disk the web server is using, thus you can offload all the work needed for correct client-side caching (ETag, If-Modified-Since, and so on) to the webserver which should already be king at doing just that.
Also, it keeps serving assets away from your application server (be it Apache behind a reverse proxy or FastCGI), saving you lots of RAM.
-
efficiency: The static file will be directly sent over by your webserver. No need to hit PHP or any other processor to render the CSS - heck you could even configure sacy to place the compiled file on a different server used for serving just static content.
-
automatically up-to-date: Because a new static file is generated whenever your assets change, you'll never have to worry about convincing clients that the files have changed. Even if one CSS-file of the compilation changed, the URL will change completely and will cause the clients to request the new file
-
ease of development: You can keep all assets in non-minified development form to work with them. You can even disable the plugin and have the assets included the traditional way during development, leaving you with all the methods for debugging you would want.
There are other solutions for this around, but sacy has a few really unique features:
-
Transformations: Sacy can transform less, sass, scss and coffee script files for you without making you remember to run deployment scripts or keeping an additional daemon running that recompiles stuff.
-
Fallback: If at any time there is an issue in generating the cached copy, sacy will not alter the existing link tags. Sure: More requests will be sent to the server, but nothing will break.
-
Concurrency: If two requests come in at the same time and the cache-file does not exist, sacy will neither create a corrupted cache file, nor will it block any request. Any request being processed while the cache is being written will have the individual links in the code. Any subsequent request will link to the compiled file.
-
Being Helpful: The unique name of the cache file contains the base names of the CSS files used to create the compilation which helps you to debug this quickly. If you look inside the file, you'll find a list of the full path names (minus the
DOCUMENT_ROOT
as to not expose private data) -
url()-rewriting: If you are using relative urls in your css files, they would break if they are used in css-files in hierarchies deeper than what sacy exposes. This would cause background images not to load and lots of other funny mistakes. Thus, sacy looks at the CSS files and tries to rewrite all url()'s it finds using information from
ASSET_COMPILE_URL_ROOT
to point to the correct files again. This is now done by Minify which is included in the package. -
inline tags since version 0.4, sacy supports transforming tags with inline content. This means that you can now have coffee-script and sass or all other types supported by sacy directly inside your page. While not recommended practice, sometimes, you need to pass data from smarty to the JS code and it would be cool to use coffee there too!
Because the the inline script passed into sacy inherently changes depending on its input and because you can nest smarty code inside such inline scripts, sacy has to hash the whole untransformed inlined code before it can check whether it already has transformed that specific piece of code.
This costs (comparatively) more performance than simply check a file for its last modification time, so using sacy to deal with inlined code segments is potentially slower than using it with external files (which is the recommended practice anyways).
Make sure that you define two constants somewhere before {asset_compile} is evaluated the first time:
ASSET_COMPILE_OUTPUT_DIR
is the filesystem path where you want the
files written to.
ASSET_COMPILE_URL_ROOT
is how that directoy is accessible
as an URL relative to the server root.
It's recommended to set OUTPUT_DIR
to some place that's not publicly
accessible and then using symlinks or a webserver level alias directory to
make just that directory accessible to the outside. You do not want
world-writable directories exposed more than what's absolutely needed.
Let's say you want to create one single smarty plugin file (for easy deployment) that contains all dependencies for both minification and transformation.
Using the power of PHP 5.3's phar module you can do that using the
build.php
script that comes with the sacy source code.
build.php is going to write a phar file, so you need to have phar write
support enabled in your php.ini (set phar.readonly
to off). Then you
can execute the build script from the command line.
It supports the following parameters:
-c
will add CSSMin to the bundle
-j
will add JSMin support into the bundle
Note that CSSMin is required by all means for sacy to work. If you decide not to bundle it, that means that you need to have it available and loaded somewhere else or sacy will blow up.
JSMin is optional if you use an external processor, but in general, sacy won't work without any means for CSS- and JS minification (which was its initial purpose).
Then, the script has two more optional parameters:
--with-phamlp=<dir>
: pass the path to the extracted source code archive of
PHamlP. This will enable sacy to transform both sass and scss files. The
currently released phamlp has an issue with @import
- the build script patches the library accordingly
--with-lessphp=<dir>
: pass the path to the extracted source code archive of
lessphp. This will enable sacy to transform less files.
--with-coffeescript-php=<dir>
: pass the path to the extracted source code
repository clone of coffeescript-php (https://github.com/alxlit/coffeescript-php).
If you provide this, sacy will transform coffee script files to javascript
for you.
As before: If you don't bundle these three but you have them loaded somewhere before sacy is used, then sacy will use your already loaded copy to do the thing.
After the script has run, you will find the compiled
block.asset_compile.php
in the build/
subdirectory. Place that (yes
just that one file) in your smarty plugins directory and you're done.
You can provide a -o
parameter to build.php
to have
block_asset_compile.php
written in another directory of your choice.
If you use -z g
or -z b
, the files inside the archive will be compressed
using either gzip (g) or bzip2 (b). Be mindful though that for such a .phar
file to be usable, the target machine needs the corresponding compression
extension to be loaded.
Note: If you indend to use the official, external tools instead of the PHP ports (see below), you don't have to include any of the PHP native tools into the bundle (aside of CSSMin for which there are no external replacements)
´{asset_tag}´ supports three parameters:
-
query_strings = ("ignore"|"force-handle")
Specifies how sacy should handle relations that contain query-strings in their location:
"ignore" will decline handling tags whose locations contain query strings
"force-handle" will handle them.
"ignore" is the default.
-
write_headers = (true|false)
Specifies whether sacy should write a header enumerating the source files into the compiled files (it will never expose the DOCUMENT_ROOT though).
This can be helpful for debugging purposes, but one might want to turn it off for file size reasons (ridiculous considering the size of the headers) or to not expose information.
true is the default, so headers are written
debug_toggle = (<string>|false)
If $_GET[<debug_toggle>]
or $_COOKIE[<debug_toggle>]
is set to either
1 or 2, then the following will happen:
"1" will make sacy decline all processing of a blocks content.
"2" will make sacy re-process all files as if the cache was empty.
"3" will make sacy only do transformations (less, sass, scss), but leave
all other files alone and write one output file per input.
Use this for development. This option is not available if you have enabled
asset merging with merge_tags
.
debug_toggle is set to _sacy_debug
per default. If it's set to false
sacy
will never do any debug handling regardless of the request.
If you want to specify a default value for all tags on a page, you can define SACY_(any of the above in uppercase) before you use the tag the first time. In that case, the value of your define() will be used as default.
Example:
define(SACY_WRITE_HEADERS, false);
Will cause {asset_compile} without parameters not to emit the headers.
In order to correctly deal with inline tags (<style>
...</style>
or
<script>
..</script>
tags not referring to an external resource), sacy
needs read/write access to a fragment cache where it stores the output of its
transformation under a key derived by hashing the original tag contents.
By default, sacy uses a very trivial filesystem based cache in a folder called
"fragments" which it auto-creates inside of ASSET_COMPILE_OUTPUT_DIR
.
If you have a better caching facility you'd like sacy to use, that's fine and
even recommended in order to safe the server a lot of stat()
-ing.
In order to do that, create a class that implements two methods:
<?php
class MyOwnFragmentCache{
function __construct(){}
function get($key){}
function set($key, $value){}
}
?>
Key is guaranteed to match /^[0-9a-z]+$/
. If get returns a value considered
false
by PHP, sacy assumes a cache miss. The constructor must not require
any arguments.
In order to not mess with Smarty's autoloading of plugins, there's no specific interface your class has to implement (as that would require you to pre-load the block plugin in order for PHP to find the interface).
Once that class exists, define SACY_FRAGMENT_CACHE_CLASS
to the name of
your class:
<?php
define('SACY_FRAGMENT_CACHE_CLASS', 'MyOwnFragmentCache');
?>
If that define is set, sacy is going to use that class instead of its own trivial file based cache.
If you have loaded my fork of sassphp, then the extension will be used to transform .scss files. The advantage of using the extension is that the compilation is done in native code and in-process. You won't get any faster .scss compilation than that.
The extension is currently based on libsass 1.0
Otherwise, by default, sacy will use pure PHP implementations for all of the transformations it supports. This has the advantage that sacy will work on any webserver - even on shared hosting environments.
The disadvantage is that you will have to fight with additional bugs that are specific to the respective PHP ports. It also means that you will lag behind the latest releases of the various tools (which can be especially annoying for the transformations like Coffee Script or SASS)
So if you are capable of providing the required dependencies (all of them require either node.js or ruby to be installed on the machine that sacy runs on), you can tell sacy to use these.
In order to not hit the file system multiple times per request to check whether the external tools are available or not, you will have to tell sacy by define()'ing some constants pointing sacy to the path of the respecive tools:
-
SACY_COMPRESSOR_UGLIFY
specifies the path to a release of uglify-js which you can install usingnpm install -g uglify-js
. If you do so, uglify is likely to be installed in/usr/local/bin/uglifyjs
. -
SACY_TRANSFORMER_COFFEE
specifies the path to a release of the coffee-script compiler. You can install that usingnpm install -g coffee-script
and if you do so, you'd likely set this to/usr/local/bin/coffee
-
SACY_TRANSFORMER_LESS
specifies the path to the less compiler. At this time, the latest released version (1.1.4) does not support reading the less data from STDIN, so you will have to ask npm to install the git version usingnpm install -g git://github.com/cloudhead/less.js.git
which will make the binary available at/usr/local/bin/lessc
. -
SACY_TRANSFORMER_SASS
specifies the path to the sass compiler. This requires a working ruby installation (1.8.7 or later) and you would install it usinggem install sass
, giving you an executable at/usr/bin/sass
-
SACY_TRANSFORMER_ECO
specifies the path to the eco compiler. You can install it usingnpm install -g eco
, which will likely put the eco executable in/usr/local/bin/eco
. -
SACY_TRANSFORMER_JSX
specifies the path to the jsx compiler. You can install it usingnpm install -g react-tools
, which will likely put the jsx executable in/usr/local/bin/jsx
.
If you are not sure what you are doing, always install these utilities to their global locations. If you install them as a user account, chances are that the user the web server runs as will not be able to find them or if they do, the tools themselves will probably not find their required libraries.
I'm not saying it's impossible - it's just much harder.
At the moment, CSS minification is done only using the native PHP CSSMin, as this is as "official" as all the other tools.
In its default configuration, sacy will write a cache file which is named based on the maximum mtime of all files in a batch. This has the advantage that it's comparably cheap to calculate and it doesn't require any additional infrastructure.
However, once you start adding multiple servers compiling assets for the same application (load balancer in front of multiple app servers), you will run into an issue that you will need to synchonize the mtime on all app servers or sacy will generate different file names on different servers.
This won't work very well with load balancers as the request for an asset might be hitting a different machine than the machine which has generated the asset to begin with.
In the end, the only option to solve this issue in sacy's default configuration is to store the asset cache on a distributed file system.
However, if you define SACY_USE_CONTENT_BASED_CACHE, sacy will name the generated asset files based on the actual contents of the files it processes.
This process is comparably expensive (md5_file()
), so you don't want
to do it for every request hitting your application.
This is why sacy, in SACY_USE_CONTENT_BASED_CACHE mode still uses an
mtime based name, but then looks up a content based key in the fragment
cache (see above). Only if it doesn't find a precached key, it will
hit the file system to do the md5_file()
operation.
Thus, it's highly recommended to use SACY_USE_CONTENT_BASED_CACHE ony if you have configured your fragment cache to something faster than the built-in filesystem cache.
This also is the reason why SACY_USE_CONTENT_BASED_CACHE is off by default.
Because the name of the generated file changes the moment you change any of the dependent files, this also means that a web browser or even a proxy server will never have to re-request a file once it has been downloaded.
To make this perfectly clear to both browsers and proxy-servers, set these the Expires-header far into the future and set Cache-Control to public so proxies will cache the file too. This will greatly decrease the load on your server and the bandwidth consumed.
To do this in Apache, I have used these directives:
<Location /assetcache>
ExpiresActive On
ExpiresDefault "access plus 1 year"
Header merge Cache-Control public
</Location>
If you are using nginx, you can accomplish the same thing using
location /assetcache {
expires max;
}
Other browsers will provide similar methods.
-
sacy doesn't handle HTML comments (bug #4), so even if a tag is inside HTML comments, it will still be rendered. Use Smarty comments for now. This also means that sacy will fail in cases where IE's conditional comments are used to fetch additional files. For now, put these conditional comments outside of a
{asset_compile}
tag -
sacy does contain very minimal support for
@import
in SASS/SCSS files in that it tracks dependent files before running the sass compiler. That means that a new compilation process is started whenever a dependent file has changed (the maximum allowed nesting level is 10 imports).However, sacy has no support for @import in pure CSS files and as @import must appear at the beginning for a CSS file, sacyfying CSS files that make use of @import is probably not practical, so if you want to use @import, it's recommended to just use SASS
This is based around the blog entry How we hash our Javascript for better caching and less breakage on updates with some added simplifications and adapted for use with Smarty.
Thanks for that blog entry and the accompanying discussions on Hacker News.
Thanks to http://github.com/rgrove/jsmin-php/ and http://code.google.com/p/minify/ for their implementations of JS and CSS minification respectively. Sacy uses a built-in copy of these if they are not already loaded when sacy is run the first time.
sacy is © 2009-2013 by Philip Hofstetter [email protected] and is licensed under the MIT License which is reprinted in the LICENSE file accompanying this distribution.
If you find this useful, consider dropping me a line, or, if you want, a patch :-)