Managing lists of tests

We don’t always want to run all tests, all the time. Sometimes a test may be broken, in other cases we only want to run a test on a specific platform or build of Mozilla. To handle these cases (and more), we created a python library to create and use test “manifests”, which codify this information.

manifestparser — Create and manage test manifests

manifestparser lets you easily create and use test manifests, to control which tests are run under what circumstances.

What manifestparser gives you:

  • manifests are ordered lists of tests
  • tests may have an arbitrary number of key, value pairs
  • the parser returns an ordered list of test data structures, which are just dicts with some keys. For example, a test with no user-specified metadata looks like this:
[{'expected': 'pass',
  'path': '/home/mozilla/mozmill/src/manifestparser/manifestparser/tests/testToolbar/testBackForwardButtons.js',
  'relpath': 'testToolbar/testBackForwardButtons.js',
  'name': 'testBackForwardButtons.js',
  'here': '/home/mozilla/mozmill/src/manifestparser/manifestparser/tests',
  'manifest': '/home/mozilla/mozmill/src/manifestparser/manifestparser/tests/manifest.ini',}]

The keys displayed here (path, relpath, name, here, and manifest) are reserved keys for manifestparser and any consuming APIs. You can add additional key, value metadata to each test.

Why have test manifests?

It is desirable to have a unified format for test manifests for testing [mozilla-central](, etc.

  • It is desirable to be able to selectively enable or disable tests based on platform or other conditions. This should be easy to do. Currently, since many of the harnesses just crawl directories, there is no effective way of disabling a test except for removal from mozilla-central
  • It is desriable to do this in a universal way so that enabling and disabling tests as well as other tasks are easily accessible to a wider audience than just those intimately familiar with the specific test framework.
  • It is desirable to have other metadata on top of the test. For instance, let’s say a test is marked as skipped. It would be nice to give the reason why.

Most Mozilla test harnesses work by crawling a directory structure. While this is straight-forward, manifests offer several practical advantages:

  • ability to turn a test off easily: if a test is broken on m-c currently, the only way to turn it off, generally speaking, is just removing the test. Often this is undesirable, as if the test should be dismissed because other people want to land and it can’t be investigated in real time (is it a failure? is the test bad? is no one around that knows the test?), then backing out a test is at best problematic. With a manifest, a test may be disabled without removing it from the tree and a bug filed with the appropriate reason:
disabled =
  • ability to run different (subsets of) tests on different platforms. Traditionally, we’ve done a bit of magic or had the test know what platform it would or would not run on. With manifests, you can mark what platforms a test will or will not run on and change these without changing the test.
skip-if = os != 'win'
  • ability to markup tests with metadata. We have a large, complicated, and always changing infrastructure. key, value metadata may be used as an annotation to a test and appropriately curated and mined. For instance, we could mark certain tests as randomorange with a bug number, if it were desirable.
  • ability to have sane and well-defined test-runs. You can keep different manifests for different test runs and [include:] (sub)manifests as appropriate to your needs.

Manifest Format

Manifests are .ini file with the section names denoting the path relative to the manifest:


The sections are read in order. In addition, tests may include arbitrary key, value metadata to be used by the harness. You may also have a [DEFAULT] section that will give key, value pairs that will be inherited by each test unless overridden:

type = restart

color = white

color = yellow
type = other
# override type from DEFAULT

color = red

You can also include other manifests:


And reference parent manifests to inherit keys and values from the DEFAULT section, without adding possible included tests.


Manifests are included relative to the directory of the manifest with the [include:] directive unless they are absolute paths.

By default you can use ‘#’ as a comment character. Comments can start a new line, or be inline.

# a valid comment
color = red # another valid comment

Comment characters must be preceded by a space, or they will not be treated as comments.

url =

The ‘#baz’ anchor will not be stripped off, as it wasn’t preceded by a space.

Special variable server-root

There is a special variable called server-root used for paths on the system. This variable is deemed a path and will be expanded into its absolute form.

Because of the inheritant nature of the key/value pairs, if one requires a system path, it must be absolute for it to be of any use in any included file.

server-root = ../data

server-root = test1/data

Manifest Conditional Expressions

The conditional expressions used in manifests are parsed using the ExpressionParser class.

class manifestparser.ExpressionParser(text, valuemapping, strict=False)[source]

A parser for a simple expression language.

The expression language can be described as follows:

OP ::= '==' | '!=' | '<' | '>' | '<=' | '>=' | '&&' | '||'
BOOL ::= 'true' | 'false'
INT ::= [0-9]+
IDENT ::= [a-zA-Z_]\w*
STRING ::= '"' [^"] '"' | ''' [^'] '''

At its core, expressions consist of booleans, integers, identifiers and. strings. Booleans are one of true or false. Integers are a series of digits. Identifiers are a series of English letters and underscores. Strings are a pair of matching quote characters (single or double) with zero or more characters inside.

Expressions can be combined with operators: the equals (==) and not equals (!=) operators compare two expressions and produce a boolean. The and (&&) and or (||) operators take two expressions and produce the logical AND or OR value of them, respectively. An expression can also be prefixed with the not (!) operator, which produces its logical negation.

Finally, any expression may be contained within parentheses for grouping.

Identifiers take their values from the mapping provided.

Consumers of this module are expected to pass in a value dictionary for evaluating conditional expressions. A common pattern is to pass the dictionary from the mozinfo module.


Manifest Destiny gives tests as a list of dictionaries (in python terms).

  • path: full path to the test

  • relpath: relative path starting from the root directory. The root directory

    is typically the location of the root manifest, or the source repository. It can be specified at runtime by passing in rootdir to TestManifest. Defaults to the directory containing the test’s ancestor manifest.

  • name: file name of the test

  • here: the parent directory of the manifest

  • manifest: the path to the manifest containing the test

This data corresponds to a one-line manifest:


If additional key, values were specified, they would be in this dict as well.

Outside of the reserved keys, the remaining key, values are up to convention to use. There is a (currently very minimal) generic integration layer in manifestparser for use of all harnesses, manifestparser.TestManifest. For instance, if the ‘disabled’ key is present, you can get the set of tests without disabled (various other queries are doable as well).

Since the system is convention-based, the harnesses may do whatever they want with the data. They may ignore it completely, they may use the provided integration layer, or they may provide their own integration layer. This should allow whatever sort of logic is desired. For instance, if in yourtestharness you wanted to run only on mondays for a certain class of tests:

tests = []
for test in manifests.tests:
    if 'runOnDay' in test:
       if calendar.day_name[calendar.weekday(*[:3])].lower() == test['runOnDay'].lower():

To recap: * the manifests allow you to specify test data * the parser gives you this data * you can use it however you want or process it further as you need

Tests are denoted by sections in an .ini file (see

Additional manifest files may be included with an [include:] directive:


The path to included files is relative to the current manifest.

The [DEFAULT] section contains variables that all tests inherit from.

Included files will inherit the top-level variables but may override in their own [DEFAULT] section.

manifestparser Architecture

There is a two- or three-layered approach to the manifestparser architecture, depending on your needs:

1. ManifestParser: this is a generic parser for .ini manifests that facilitates the [include:] logic and the inheritence of metadata. Despite the internal variable being called self.tests (an oversight), this layer has nothing in particular to do with tests.

2. TestManifest: this is a harness-agnostic integration layer that is test-specific. TestManifest faciliates skip-if logic.

3. Optionally, a harness will have an integration layer than inherits from TestManifest if more harness-specific customization is desired at the manifest level.

See the source code at and in particular.

Filtering Manifests

After creating a TestManifest object, all manifest files are read and a list of test objects can be accessed via TestManifest.tests. However this list contains all test objects, whether they should be run or not. Normally they need to be filtered down only to the set of tests that should be run by the test harness.

To do this, a test harness can call TestManifest.active_tests:

tests = manifest.active_tests(exists=True, disabled=True, **tags)

By default, active_tests runs the filters found in DEFAULT_FILTERS. It also accepts two convenience arguments:

  1. exists: if True (default), filter out tests that do not exist on the local file system.
  2. disabled: if True (default), do not filter out tests containing the ‘disabled’ key (which can be set by skip-if manually).

This works for simple cases, but there are other built-in filters, or even custom filters that can be applied to the TestManifest. To do so, add the filter to TestManifest.filters:

from manifestparser.filters import subsuite
import mozinfo

filters = [subsuite('devtools')]
tests = manifest.active_tests(filters=filters, **

A filter is a callable that accepts an iterable of test objects and a dictionary of values, and returns a new iterable of test objects. It is possible to define custom filters if the built-in ones are not enough.

class manifestparser.filters.chunk_by_dir(this_chunk, total_chunks, depth)[source]

Basic chunking algorithm that splits directories of tests evenly at a given depth.

For example, a depth of 2 means all test directories two path nodes away from the base are gathered, then split evenly across the total number of chunks. The number of tests in each of the directories is not taken into account (so chunks will not contain an even number of tests). All test paths must be relative to the same root (typically the root of the source repository).

  • this_chunk – the current chunk, 1 <= this_chunk <= total_chunks
  • total_chunks – the total number of chunks
  • depth – the minimum depth of a subdirectory before it will be considered unique
class manifestparser.filters.chunk_by_runtime(this_chunk, total_chunks, runtimes, default_runtime=0)[source]

Chunking algorithm that attempts to group tests into chunks based on their average runtimes. It keeps manifests of tests together and pairs slow running manifests with fast ones.

  • this_chunk – the current chunk, 1 <= this_chunk <= total_chunks
  • total_chunks – the total number of chunks
  • runtimes – dictionary of test runtime data, of the form {<test path>: <average runtime>}
  • default_runtime – value in seconds to assign tests that don’t exist in the runtimes file
class manifestparser.filters.chunk_by_slice(this_chunk, total_chunks, disabled=False)[source]

Basic chunking algorithm that splits tests evenly across total chunks.

  • this_chunk – the current chunk, 1 <= this_chunk <= total_chunks
  • total_chunks – the total number of chunks
  • disabled – Whether to include disabled tests in the chunking algorithm. If False, each chunk contains an equal number of non-disabled tests. If True, each chunk contains an equal number of tests (default False)
manifestparser.filters.enabled(tests, values)[source]

Removes all tests containing the disabled key. This filter can be added by passing disabled=False into active_tests.

manifestparser.filters.exists(tests, values)[source]

Removes all tests that do not exist on the file system. This filter is added by default, but can be removed by passing exists=False into active_tests.

manifestparser.filters.fail_if(tests, values)[source]

Sets expected to ‘fail’ on all tests containing the fail-if tag and whose condition is True. This filter is added by default.

class manifestparser.filters.pathprefix(paths)[source]

Removes tests that don’t start with any of the given test paths.

Parameters:paths – A list of test paths to filter on
manifestparser.filters.run_if(tests, values)[source]

Sets disabled on all tests containing the run-if tag and whose condition is False. This filter is added by default.

manifestparser.filters.skip_if(tests, values)[source]

Sets disabled on all tests containing the skip-if tag and whose condition is True. This filter is added by default.

class manifestparser.filters.subsuite(name=None)[source]

If name is None, removes all tests that have a subsuite key. Otherwise removes all tests that do not have a subsuite matching name.

It is possible to specify conditional subsuite keys using:
subsuite = foo,condition

where ‘foo’ is the subsuite name, and ‘condition’ is the same type of condition used for skip-if. If the condition doesn’t evaluate to true, the subsuite designation will be removed from the test.

Parameters:name – The name of the subsuite to run (default None)
class manifestparser.filters.tags(tags)[source]

Removes tests that don’t contain any of the given tags. This overrides InstanceFilter’s __eq__ method, so multiple instances can be added. Multiple tag filters is equivalent to joining tags with the AND operator.

To define a tag in a manifest, add a tags attribute to a test or DEFAULT section. Tests can have multiple tags, in which case they should be whitespace delimited. For example:

[test_foobar.html] tags = foo bar

Parameters:tags – A tag or list of tags to filter tests on

By default active_tests() will run the skip_if(), run_if() and fail_if() filters.

For example, suppose we want to introduce a new key called timeout-if that adds a ‘timeout’ property to a test if a certain condition is True. The syntax in the manifest files will look like this:

timeout-if = 300, os == 'win'

The value is <timeout>, <condition> where condition is the same format as the one in skip-if. In the above case, if os == ‘win’, a timeout of 300 seconds will be applied. Otherwise, no timeout will be applied. All we need to do is define the filter and add it:

from manifestparser.expression import parse
import mozinfo

def timeout_if(tests, values):
    for test in tests:
        if 'timeout-if' in test:
            timeout, condition = test['timeout-if'].split(',', 1)
            if parse(condition, **values):
                test['timeout'] = timeout
        yield test

tests = manifest.active_tests(filters=[timeout_if], **

Creating Manifests

manifestparser comes with a console script, manifestparser create, that may be used to create a seed manifest structure from a directory of files. Run manifestparser help create for usage information.

Copying Manifests

To copy tests and manifests from a source:

manifestparser [options] copy from_manifest to_directory -tag1 -tag2 `key1=value1 key2=value2 ...

Updating Tests

To update the tests associated with with a manifest from a source directory:

manifestparser [options] update manifest from_directory -tag1 -tag2 `key1=value1 `key2=value2 ...

Usage example

Here is an example of how to create manifests for a directory tree and update the tests listed in the manifests from an external source.

Creating Manifests

Let’s say you want to make a series of manifests for a given directory structure containing .js test files:


You can use manifestparser create to do this:

$ manifestparser help create
Usage: [options] create directory <directory> <...>

     create a manifest from a list of directories

  -p PATTERN, `pattern=PATTERN
                        glob pattern for files
  -i IGNORE, `ignore=IGNORE
                        directories to ignore
  -w IN_PLACE, --in-place=IN_PLACE
                        Write .ini files in place; filename to write to

We only want .js files and we want to skip the restartTests directory. We also want to write a manifest per directory, so I use the –in-place option to write the manifests:

manifestparser create . -i restartTests -p '*.js' -w manifest.ini

This creates a manifest.ini per directory that we care about with the JS test files:


The top-level manifest.ini merely has [include:] references to the sub manifests:


Each sub-level manifest contains the (.js) test files relative to it.

Updating the tests from manifests

You may need to update tests as given in manifests from a different source directory. manifestparser update was made for just this purpose:

Usage: manifestparser [options] update manifest directory -tag1 -tag2 `key1=value1 --key2=value2 ...

    update the tests as listed in a manifest from a directory

To update from a directory of tests in ~/mozmill/src/mozmill-tests/firefox/ run:

manifestparser update manifest.ini ~/mozmill/src/mozmill-tests/firefox/


manifestparser includes a suite of tests.

test_manifest.txt is a doctest that may be helpful in figuring out how to use the API. Tests are run via mach python-test testing/mozbase/manifestparser.


Please file any bugs or feature requests at

Or contact jhammel or in #ateam on


Run manifestparser help for usage information.

To create a manifest from a set of directories:

manifestparser [options] create directory <directory> <...> [create-options]

To output a manifest of tests:

manifestparser [options] write manifest <manifest> <...> -tag1 -tag2 --key1=value1 --key2=value2 ...

To copy tests and manifests from a source:

manifestparser [options] copy from_manifest to_manifest -tag1 -tag2 `key1=value1 key2=value2 ...

To update the tests associated with with a manifest from a source directory:

manifestparser [options] update manifest from_directory -tag1 -tag2 --key1=value1 --key2=value2 ...

Design Considerations

Contrary to some opinion, and the associated .ini format were not magically plucked from the sky but were descended upon through several design considerations.

  • test manifests should be ordered. While python 2.6 and greater has a ConfigParser that can use an ordered dictionary, it is a requirement that we support python 2.4 for the build + testing environment. To that end, a read_ini function was implemented in that should be the equivalent of the .ini dialect used by ConfigParser.
  • the manifest format should be easily human readable/writable. While there was initially some thought of using JSON, there was pushback that JSON was not easily editable. An ideal manifest format would degenerate to a line-separated list of files. While .ini format requires an additional [] per line, and while there have been complaints about this, hopefully this is good enough.
  • python does not have an in-built YAML parser. Since it was undesirable for to have any dependencies, YAML was dismissed as a format.
  • we could have used a proprietary format but decided against it. Everyone knows .ini and there are good tools to deal with it. However, since read_ini is the only function that transforms a manifest to a list of key, value pairs, while the implications for changing the format impacts downstream code, doing so should be programmatically simple.
  • there should be a single file that may easily be transported. Traditionally, test harnesses have lived in mozilla-central. This is less true these days and it is increasingly likely that more tests will not live in mozilla-central going forward. So should be highly consumable. To this end, it is a single file, as appropriate to mozilla-central, which is also a working python package deployed to PyPI for easy installation.

Historical Reference

Date-ordered list of links about how manifests came to be where they are today: