Static Assets

An asset is any file contained within a Python package which is not a Python source code file. For example, each of the following is an asset:

  • a GIF image file contained within a Python package or contained within any subdirectory of a Python package.

  • a CSS file contained within a Python package or contained within any subdirectory of a Python package.

  • a JavaScript source file contained within a Python package or contained within any subdirectory of a Python package.

  • A directory within a package that does not have an __init__.py in it (if it possessed an __init__.py it would be a package).

  • a Chameleon or Mako template file contained within a Python package.

The use of assets is quite common in most web development projects. For example, when you create a Pyramid application using the cookiecutter, as described in Creating the Project, the directory representing the application contains a Python package. Within that Python package, there are directories full of files which are static assets. For example, there's a static directory which contains .css, .js, and .gif files. These asset files are delivered when a user visits an application URL.

Understanding Asset Specifications

Let's imagine you've created a Pyramid application that uses a Chameleon ZPT template via the pyramid.renderers.render_to_response() API. For example, the application might address the asset using the asset specification myapp:templates/some_template.pt using that API within a views.py file inside a myapp package:

1from pyramid.renderers import render_to_response
2render_to_response('myapp:templates/some_template.pt', {}, request)

"Under the hood", when this API is called, Pyramid attempts to make sense out of the string myapp:templates/some_template.pt provided by the developer. This string is an asset specification. It is composed of two parts:

  • The package name (myapp)

  • The asset name (templates/some_template.pt), relative to the package directory.

The two parts are separated by a colon : character.

Pyramid uses the Python pkg_resources API to resolve the package name and asset name to an absolute (operating system-specific) file name. It eventually passes this resolved absolute filesystem path to the Chameleon templating engine, which then uses it to load, parse, and execute the template file.

There is a second form of asset specification: a relative asset specification. Instead of using an "absolute" asset specification which includes the package name, in certain circumstances you can omit the package name from the specification. For example, you might be able to use templates/mytemplate.pt instead of myapp:templates/some_template.pt. Such asset specifications are usually relative to a "current package". The "current package" is usually the package which contains the code that uses the asset specification. Pyramid APIs which accept relative asset specifications typically describe to what the asset is relative in their individual documentation.

Serving Static Assets

Pyramid makes it possible to serve up static asset files from a directory on a filesystem to an application user's browser. Use the pyramid.config.Configurator.add_static_view() to instruct Pyramid to serve static assets, such as JavaScript and CSS files. This mechanism makes a directory of static files available at a name relative to the application root URL, e.g., /static, or as an external URL.

Note

add_static_view() cannot serve a single file, nor can it serve a directory of static files directly relative to the root URL of a Pyramid application. For these features, see Advanced: Serving Static Assets Using a View Callable.

Here's an example of a use of add_static_view() that will serve files up from the /var/www/static directory of the computer which runs the Pyramid application as URLs beneath the /static URL prefix.

1# config is an instance of pyramid.config.Configurator
2config.add_static_view(name='static', path='/var/www/static')

The name represents a URL prefix. In order for files that live in the path directory to be served, a URL that requests one of them must begin with that prefix. In the example above, name is static and path is /var/www/static. In English this means that you wish to serve the files that live in /var/www/static as sub-URLs of the /static URL prefix. Therefore, the file /var/www/static/foo.css will be returned when the user visits your application's URL /static/foo.css.

A static directory named at path may contain subdirectories recursively, and any subdirectories may hold files; these will be resolved by the static view as you would expect. The Content-Type header returned by the static view for each particular type of file is dependent upon its file extension.

By default, all files made available via add_static_view() are accessible by completely anonymous users. Simple authorization can be required, however. To protect a set of static files using a permission, in addition to passing the required name and path arguments, also pass the permission keyword argument to add_static_view(). The value of the permission argument represents the permission that the user must have relative to the current context when the static view is invoked. A user will be required to possess this permission to view any of the files represented by path of the static view. If your static assets must be protected by a more complex authorization scheme, see Advanced: Serving Static Assets Using a View Callable.

Here's another example that uses an asset specification instead of an absolute path as the path argument. To convince add_static_view() to serve files up under the /static URL from the a/b/c/static directory of the Python package named some_package, we can use a fully qualified asset specification as the path:

1# config is an instance of pyramid.config.Configurator
2config.add_static_view(name='static', path='some_package:a/b/c/static')

The path provided to add_static_view() may be a fully qualified asset specification or an absolute path.

Instead of representing a URL prefix, the name argument of a call to add_static_view() can alternately be a URL. Each of the examples we've seen so far have shown usage of the name argument as a URL prefix. However, when name is a URL, static assets can be served from an external webserver. In this mode, the name is used as the URL prefix when generating a URL using pyramid.request.Request.static_url().

For example, add_static_view() may be fed a name argument which is http://example.com/images:

1# config is an instance of pyramid.config.Configurator
2config.add_static_view(name='http://example.com/images',
3                       path='mypackage:images')

Because add_static_view() is provided with a name argument that is the URL http://example.com/images, subsequent calls to static_url() with paths that start with the path argument passed to add_static_view() will generate a URL something like http://example.com/images/logo.png. The external webserver listening on example.com must be itself configured to respond properly to such a request. The static_url() API is discussed in more detail later in this chapter.

Serving Pre-compressed Assets

New in version 2.0.

It's possible to configure Pyramid to serve pre-compressed static assets. This can greatly reduce the bandwidth required to serve assets - most modern browsers support gzip, deflate, and br (brotli) encoded responses. A client declares support for encoded responses using the Accept-Encoding HTTP header. For example, Accept-Encoding: gzip, default, br. The response will then contain the pre-compressed content with the Content-Encoding header set to the matched encoding. This feature assumes that the static assets exist unencoded (identity encoding) as well as in zero or more encoded formats. If the encoded version of a file is missing, or the client doesn't declare support for the encoded version, the unencoded version is returned instead.

In order to configure this in your application, the first step is to compress your assets. For example, update your static asset pipeline to export .gz versions of every file. Second, add content_encodings=['gzip'] when invoking pyramid.config.Configurator.add_static_view().

The encoded file extensions are determined by mimetypes.encodings_map. So, if your desired encoding is missing, you'll need to add it there:

import mimetypes
mimetypes.encodings_map['.br'] = 'br'  # add brotli

It is not necessary for every file to support every encoding, but Pyramid will not serve an encoding that is not declared.

Generating Static Asset URLs

When an add_static_view() method is used to register a static asset directory, a special helper API named pyramid.request.Request.static_url() can be used to generate the appropriate URL for an asset that lives in one of the directories named by the static registration path attribute.

For example, let's assume you create a set of static declarations like so:

1config.add_static_view(name='static1', path='mypackage:assets/1')
2config.add_static_view(name='static2', path='mypackage:assets/2')

These declarations create URL-accessible directories which have URLs that begin with /static1 and /static2, respectively. The assets in the assets/1 directory of the mypackage package are consulted when a user visits a URL which begins with /static1, and the assets in the assets/2 directory of the mypackage package are consulted when a user visits a URL which begins with /static2.

You needn't generate the URLs to static assets "by hand" in such a configuration. Instead, use the static_url() API to generate them for you. For example:

1from pyramid.renderers import render_to_response
2
3def my_view(request):
4    css_url = request.static_url('mypackage:assets/1/foo.css')
5    js_url = request.static_url('mypackage:assets/2/foo.js')
6    return render_to_response('templates/my_template.pt',
7                              dict(css_url=css_url, js_url=js_url),
8                              request=request)

If the request "application URL" of the running system is http://example.com, the css_url generated above would be: http://example.com/static1/foo.css. The js_url generated above would be http://example.com/static2/foo.js.

One benefit of using the static_url() function rather than constructing static URLs "by hand" is that if you need to change the name of a static URL declaration, the generated URLs will continue to resolve properly after the rename.

URLs may also be generated by static_url() to static assets that live outside the Pyramid application. This will happen when the add_static_view() API associated with the path fed to static_url() is a URL instead of a view name. For example, the name argument may be http://example.com while the path given may be mypackage:images:

1config.add_static_view(name='http://example.com/images',
2                      path='mypackage:images')

Under such a configuration, the URL generated by static_url for assets which begin with mypackage:images will be prefixed with http://example.com/images:

1request.static_url('mypackage:images/logo.png')
2# -> http://example.com/images/logo.png

Using static_url() in conjunction with a add_static_view() makes it possible to put static media on a separate webserver during production (if the name argument to add_static_view() is a URL), while keeping static media package-internal and served by the development webserver during development (if the name argument to add_static_view() is a URL prefix).

For example, we may define a custom setting named media_location which we can set to an external URL in production when our assets are hosted on a CDN.

1media_location = settings.get('media_location', 'static')
2
3config = Configurator(settings=settings)
4config.add_static_view(path='myapp:static', name=media_location)

Now we can optionally define the setting in our ini file:

1# production.ini
2[app:main]
3use = egg:myapp#main
4
5media_location = http://static.example.com/

It is also possible to serve assets that live outside of the source by referring to an absolute path on the filesystem. There are two ways to accomplish this.

First, add_static_view() supports taking an absolute path directly instead of an asset spec. This works as expected, looking in the file or folder of files and serving them up at some URL within your application or externally. Unfortunately, this technique has a drawback in that it is not possible to use the static_url() method to generate URLs, since it works based on an asset specification.

New in version 1.6.

The second approach, available in Pyramid 1.6+, uses the asset overriding APIs described in the Overriding Assets section. It is then possible to configure a "dummy" package which then serves its file or folder from an absolute path.

config.add_static_view(path='myapp:static_images', name='static')
config.override_asset(to_override='myapp:static_images/',
                      override_with='/abs/path/to/images/')

From this configuration it is now possible to use static_url() to generate URLs to the data in the folder by doing something like request.static_url('myapp:static_images/foo.png'). While it is not necessary that the static_images file or folder actually exist in the myapp package, it is important that the myapp portion points to a valid package. If the folder does exist, then the overriden folder is given priority, if the file's name exists in both locations.

Cache Busting

New in version 1.6.

In order to maximize performance of a web application, you generally want to limit the number of times a particular client requests the same static asset. Ideally a client would cache a particular static asset "forever", requiring it to be sent to the client a single time. The HTTP protocol allows you to send headers with an HTTP response that can instruct a client to cache a particular asset for an amount of time. As long as the client has a copy of the asset in its cache and that cache hasn't expired, the client will use the cached copy rather than request a new copy from the server. The drawback to sending cache headers to the client for a static asset is that at some point the static asset may change, and then you'll want the client to load a new copy of the asset. Under normal circumstances you'd just need to wait for the client's cached copy to expire before they get the new version of the static resource.

A commonly used workaround to this problem is a technique known as cache busting. Cache busting schemes generally involve generating a URL for a static asset that changes when the static asset changes. This way headers can be sent along with the static asset instructing the client to cache the asset for a very long time. When a static asset is changed, the URL used to refer to it in a web page also changes, so the client sees it as a new resource and requests the asset, regardless of any caching policy set for the resource's old URL.

Pyramid can be configured to produce cache busting URLs for static assets using add_cache_buster():

1import time
2from pyramid.static import QueryStringConstantCacheBuster
3
4# config is an instance of pyramid.config.Configurator
5config.add_static_view(name='static', path='mypackage:folder/static/')
6config.add_cache_buster(
7    'mypackage:folder/static/',
8    QueryStringConstantCacheBuster(str(int(time.time()))))

Adding the cachebuster instructs Pyramid to add the current time for a static asset to the query string in the asset's URL:

1js_url = request.static_url('mypackage:folder/static/js/myapp.js')
2# Returns: 'http://www.example.com/static/js/myapp.js?x=1445318121'

When the web server restarts, the time constant will change and therefore so will its URL.

Note

Cache busting is an inherently complex topic as it integrates the asset pipeline and the web application. It is expected and desired that application authors will write their own cache buster implementations conforming to the properties of their own asset pipelines. See Customizing the Cache Buster for information on writing your own.

Disabling the Cache Buster

It can be useful in some situations (e.g., development) to globally disable all configured cache busters without changing calls to add_cache_buster(). To do this set the PYRAMID_PREVENT_CACHEBUST environment variable or the pyramid.prevent_cachebust configuration value to a true value.

Customizing the Cache Buster

Calls to add_cache_buster() may use any object that implements the interface ICacheBuster.

Pyramid ships with a very simplistic QueryStringConstantCacheBuster, which adds an arbitrary token you provide to the query string of the asset's URL. This is almost never what you want in production as it does not allow fine-grained busting of individual assets.

In order to implement your own cache buster, you can write your own class from scratch which implements the ICacheBuster interface. Alternatively you may choose to subclass one of the existing implementations. One of the most likely scenarios is you'd want to change the way the asset token is generated. To do this just subclass QueryStringCacheBuster and define a tokenize(pathspec) method. Here is an example which uses Git to get the hash of the current commit:

 1import os
 2import subprocess
 3from pyramid.static import QueryStringCacheBuster
 4
 5class GitCacheBuster(QueryStringCacheBuster):
 6    """
 7    Assuming your code is installed as a Git checkout, as opposed to an egg
 8    from an egg repository like PYPI, you can use this cachebuster to get
 9    the current commit's SHA1 to use as the cache bust token.
10    """
11    def __init__(self, param='x', repo_path=None):
12        super(GitCacheBuster, self).__init__(param=param)
13        if repo_path is None:
14            repo_path = os.path.dirname(os.path.abspath(__file__))
15        self.sha1 = subprocess.check_output(
16            ['git', 'rev-parse', 'HEAD'],
17            cwd=repo_path).strip()
18
19    def tokenize(self, pathspec):
20        return self.sha1

A simple cache buster that modifies the path segment can be constructed as well:

 1import posixpath
 2
 3class PathConstantCacheBuster(object):
 4    def __init__(self, token):
 5        self.token = token
 6
 7    def __call__(self, request, subpath, kw):
 8        base_subpath, ext = posixpath.splitext(subpath)
 9        new_subpath = base_subpath + self.token + ext
10        return new_subpath, kw

The caveat with this approach is that modifying the path segment changes the file name, and thus must match what is actually on the filesystem in order for add_static_view() to find the file. It's better to use the ManifestCacheBuster for these situations, as described in the next section.

Path Segments and Choosing a Cache Buster

Many caching HTTP proxies will fail to cache a resource if the URL contains a query string. Therefore, in general, you should prefer a cache busting strategy which modifies the path segment rather than methods which add a token to the query string.

You will need to consider whether the Pyramid application will be serving your static assets, whether you are using an external asset pipeline to handle rewriting urls internal to the css/javascript, and how fine-grained do you want the cache busting tokens to be.

In many cases you will want to host the static assets on another web server or externally on a CDN. In these cases your Pyramid application may not even have access to a copy of the static assets. In order to cache bust these assets you will need some information about them.

If you are using an external asset pipeline to generate your static files you should consider using the ManifestCacheBuster. This cache buster can load a standard JSON formatted file generated by your pipeline and use it to cache bust the assets. This has many performance advantages as Pyramid does not need to look at the files to generate any cache busting tokens, but still supports fine-grained per-file tokens.

Assuming an example manifest.json like:

{
    "css/main.css": "css/main-678b7c80.css",
    "images/background.png": "images/background-a8169106.png"
}

The following code would set up a cachebuster:

1from pyramid.static import ManifestCacheBuster
2
3config.add_static_view(
4    name='http://mycdn.example.com/',
5    path='mypackage:static')
6
7config.add_cache_buster(
8    'mypackage:static/',
9    ManifestCacheBuster('myapp:static/manifest.json'))

It's important to note that the cache buster only handles generating cache-busted URLs for static assets. It does NOT provide any solutions for serving those assets. For example, if you generated a URL for css/main-678b7c80.css then that URL needs to be valid either by configuring add_static_view properly to point to the location of the files or some other mechanism such as the files existing on your CDN or rewriting the incoming URL to remove the cache bust tokens.

CSS and JavaScript source and cache busting

Often one needs to refer to images and other static assets inside CSS and JavaScript files. If cache busting is active, the final static asset URL is not available until the static assets have been assembled. These URLs cannot be handwritten. Below is an example of how to integrate the cache buster into the entire stack. Remember, it is just an example and should be modified to fit your specific tools.

  • First, process the files by using a precompiler which rewrites URLs to their final cache-busted form. Then, you can use the ManifestCacheBuster to synchronize your asset pipeline with Pyramid, allowing the pipeline to have full control over the final URLs of your assets.

Now that you are able to generate static URLs within Pyramid, you'll need to handle URLs that are out of our control. To do this you may use some of the following options to get started:

  • Configure your asset pipeline to rewrite URL references inline in CSS and JavaScript. This is the best approach because then the files may be hosted by Pyramid or an external CDN without having to change anything. They really are static.

  • Templatize JS and CSS, and call request.static_url() inside their template code. While this approach may work in certain scenarios, it is not recommended because your static assets will not really be static and are now dependent on Pyramid to be served correctly. See Advanced: Serving Static Assets Using a View Callable for more information on this approach.

If your CSS and JavaScript assets use URLs to reference other assets it is recommended that you implement an external asset pipeline that can rewrite the generated static files with new URLs containing cache busting tokens. The machinery inside Pyramid will not help with this step as it has very little knowledge of the asset types your application may use. The integration into Pyramid is simply for linking those assets into your HTML and other dynamic content.

Advanced: Serving Static Assets Using a View Callable

For more flexibility, static assets can be served by a view callable which you register manually. For example, if you're using URL dispatch, you may want static assets to only be available as a fallback if no previous route matches. Alternatively, you might like to serve a particular static asset manually, because its download requires authentication.

Note that you cannot use the static_url() API to generate URLs against assets made accessible by registering a custom static view.

Root-Relative Custom Static View (URL Dispatch Only)

The pyramid.static.static_view helper class generates a Pyramid view callable. This view callable can serve static assets from a directory. An instance of this class is actually used by the add_static_view() configuration method, so its behavior is almost exactly the same once it's configured.

Warning

The following example will not work for applications that use traversal; it will only work if you use URL dispatch exclusively. The root-relative route we'll be registering will always be matched before traversal takes place, subverting any views registered via add_view (at least those without a route_name). A static_view static view cannot be made root-relative when you use traversal unless it's registered as a Not Found View.

To serve files within a directory located on your filesystem at /path/to/static/dir as the result of a "catchall" route hanging from the root that exists at the end of your routing table, create an instance of the static_view class inside a static.py file in your application root as below.

1from pyramid.static import static_view
2static_view = static_view('/path/to/static/dir', use_subpath=True)

Note

For better cross-system flexibility, use an asset specification as the argument to static_view instead of a physical absolute filesystem path, e.g., mypackage:static, instead of /path/to/mypackage/static.

Subsequently, you may wire the files that are served by this view up to be accessible as /<filename> using a configuration method in your application's startup code.

1# .. every other add_route declaration should come
2# before this one, as it will, by default, catch all requests
3
4config.add_route('catchall_static', '/*subpath')
5config.add_view('myapp.static.static_view', route_name='catchall_static')

The special name *subpath above is used by the static_view view callable to signify the path of the file relative to the directory you're serving.

Registering a View Callable to Serve a "Static" Asset

You can register a simple view callable to serve a single static asset. To do so, do things "by hand". First define the view callable.

1import os
2from pyramid.response import FileResponse
3
4def favicon_view(request):
5    here = os.path.dirname(__file__)
6    icon = os.path.join(here, 'static', 'favicon.ico')
7    return FileResponse(icon, request=request)

The above bit of code within favicon_view computes "here", which is a path relative to the Python file in which the function is defined. It then creates a pyramid.response.FileResponse using the file path as the response's path argument and the request as the response's request argument. pyramid.response.FileResponse will serve the file as quickly as possible when it's used this way. It makes sure to set the right content length and content_type, too, based on the file extension of the file you pass.

You might register such a view via configuration as a view callable that should be called as the result of a traversal:

1config.add_view('myapp.views.favicon_view', name='favicon.ico')

Or you might register it to be the view callable for a particular route:

1config.add_route('favicon', '/favicon.ico')
2config.add_view('myapp.views.favicon_view', route_name='favicon')

Because this is a simple view callable, it can be protected with a permission or can be configured to respond under different circumstances using view predicate arguments.

Overriding Assets

It can often be useful to override specific assets from "outside" a given Pyramid application. For example, you may wish to reuse an existing Pyramid application more or less unchanged. However, some specific template file owned by the application might have inappropriate HTML, or some static asset (such as a logo file or some CSS file) might not be appropriate. You could just fork the application entirely, but it's often more convenient to just override the assets that are inappropriate and reuse the application "as is". This is particularly true when you reuse some "core" application over and over again for some set of customers (such as a CMS application, or some bug tracking application), and you want to make arbitrary visual modifications to a particular application deployment without forking the underlying code.

To this end, Pyramid contains a feature that makes it possible to "override" one asset with one or more other assets. In support of this feature, a Configurator API exists named pyramid.config.Configurator.override_asset(). This API allows you to override the following kinds of assets defined in any Python package:

  • Individual template files.

  • A directory containing multiple template files.

  • Individual static files served up by an instance of the pyramid.static.static_view helper class.

  • A directory of static files served up by an instance of the pyramid.static.static_view helper class.

  • Any other asset (or set of assets) addressed by code that uses the Setuptools pkg_resources API.

The override_asset API

An individual call to override_asset() can override a single asset. For example:

1config.override_asset(
2    to_override='some.package:templates/mytemplate.pt',
3    override_with='another.package:othertemplates/anothertemplate.pt')

The string value passed to both to_override and override_with sent to the override_asset API is called an asset specification. The colon separator in a specification separates the package name from the asset name. The colon and the following asset name are optional. If they are not specified, the override attempts to resolve every lookup into a package from the directory of another package. For example:

1config.override_asset(to_override='some.package',
2                      override_with='another.package')

Individual subdirectories within a package can also be overridden:

1config.override_asset(to_override='some.package:templates/',
2                     override_with='another.package:othertemplates/')

If you wish to override a directory with another directory, you must make sure to attach the slash to the end of both the to_override specification and the override_with specification. If you fail to attach a slash to the end of a specification that points to a directory, you will get unexpected results.

You cannot override a directory specification with a file specification, and vice versa; a startup error will occur if you try. You cannot override an asset with itself; a startup error will occur if you try.

Only individual package assets may be overridden. Overrides will not traverse through subpackages within an overridden package. This means that if you want to override assets for both some.package:templates, and some.package.views:templates, you will need to register two overrides.

The package name in a specification may start with a dot, meaning that the package is relative to the package in which the configuration construction file resides (or the package argument to the Configurator class construction). For example:

1config.override_asset(to_override='.subpackage:templates/',
2                      override_with='another.package:templates/')

Multiple calls to override_asset which name a shared to_override but a different override_with specification can be "stacked" to form a search path. The first asset that exists in the search path will be used; if no asset exists in the override path, the original asset is used.

Asset overrides can actually override assets other than templates and static files. Any software which uses the pkg_resources.get_resource_filename(), pkg_resources.get_resource_stream(), or pkg_resources.get_resource_string() APIs will obtain an overridden file when an override is used.

New in version 1.6: As of Pyramid 1.6, it is also possible to override an asset by supplying an absolute path to a file or directory. This may be useful if the assets are not distributed as part of a Python package.

Cache Busting and Asset Overrides

Overriding static assets that are being hosted using pyramid.config.Configurator.add_static_view() can affect your cache busting strategy when using any cache busters that are asset-aware such as pyramid.static.ManifestCacheBuster. What sets asset-aware cache busters apart is that they have logic tied to specific assets. For example, a manifest is only generated for a specific set of pre-defined assets. Now, imagine you have overridden an asset defined in this manifest with a new, unknown version. By default, the cache buster will be invoked for an asset it has never seen before and will likely end up returning a cache busting token for the original asset rather than the asset that will actually end up being served! In order to get around this issue, it's possible to attach a different pyramid.interfaces.ICacheBuster implementation to the new assets. This would cause the original assets to be served by their manifest, and the new assets served by their own cache buster. To do this, pyramid.config.Configurator.add_cache_buster() supports an explicit option. For example:

 1from pyramid.static import ManifestCacheBuster
 2
 3# define a static view for myapp:static assets
 4config.add_static_view('static', 'myapp:static')
 5
 6# setup a cache buster for your app based on the myapp:static assets
 7my_cb = ManifestCacheBuster('myapp:static/manifest.json')
 8config.add_cache_buster('myapp:static', my_cb)
 9
10# override an asset
11config.override_asset(
12    to_override='myapp:static/background.png',
13    override_with='theme:static/background.png')
14
15# override the cache buster for theme:static assets
16theme_cb = ManifestCacheBuster('theme:static/manifest.json')
17config.add_cache_buster('theme:static', theme_cb, explicit=True)

In the above example there is a default cache buster, my_cb, for all assets served from the myapp:static folder. This would also affect theme:static/background.png when generating URLs via request.static_url('myapp:static/background.png').

The theme_cb is defined explicitly for any assets loaded from the theme:static folder. Explicit cache busters have priority and thus theme_cb would be invoked for request.static_url('myapp:static/background.png'), but my_cb would be used for any other assets like request.static_url('myapp:static/favicon.ico').