Tasks

Intro

doit is all about automating task dependency management and execution. Tasks can execute external shell commands/scripts or python functions (actually any callable). So a task can be anything you can code :)

Tasks are defined in plain python module with some conventions.

Note

You should be comfortable with python basics. If you don’t know python yet check Python tutorial.

A function that starts with the name task_ defines a task-creator recognized by doit. These functions must return (or yield) dictionaries representing a task. A python module/file that defines tasks for doit is called dodo file (that is something like a Makefile for make).

Take a look at this example (file dodo.py):

def task_hello():
    """hello"""

    def python_hello(targets):
        with open(targets[0], "a") as output:
            output.write("Python says Hello World!!!\n")

    return {
        'actions': [python_hello],
        'targets': ["hello.txt"],
        }

When doit is executed without any parameters it will look for tasks in a file named dodo.py in the current folder and execute its tasks.

$ doit
.  hello

On the output it displays which tasks were executed. In this case the dodo file has only one task, hello.

Actions

Every task must define actions. It can optionally define other attributes like targets, file_dep, verbosity, doc ...

Actions define what the task actually does. Actions is always a list that can have any number of elements. The actions of a task are always run sequentially. There are 2 basic kinds of actions: cmd-action and python-action. The action “result” is used to determine if task execution was successful or not.

python-action

If action is a python callable or a tuple (callable, *args, **kwargs) - only callable is required. The callable must be a function, method or callable object. Classes and built-in functions are not allowed. args is a sequence and kwargs is a dictionary that will be used as positional and keywords arguments for the callable. see Keyword Arguments.

The result of the task is given by the returned value of the action function.

For successful completion it must return one of:

  • True
  • None
  • a dictionary
  • a string

For unsuccessful completion it must return one of:

  • False indicates the task generally failed
  • if it raises any exception, it will be considered an error
  • it can also explicitly return an instance of TaskFailed or TaskError

If the action returns a type other than the types already discussed, the action will be considered a failure, although this behavior might change in future versions.

def task_hello():
    """hello py """

    def python_hello(times, text, targets):
        with open(targets[0], "a") as output:
            output.write(times * text)

    return {'actions': [(python_hello, [3, "py!\n"])],
            'targets': ["hello.txt"],
            }

The function task_hello is a task-creator, not the task itself. The body of the task-creator function is always executed when the dodo file is loaded.

task-creators vs actions

The body of task-creators are executed even if the task is not going to be executed. The body of task-creators should be used to create task metadata only, not execute tasks! From now on when the documentation says that a task is executed, read “the task’s actions are executed”.

cmd-action

CmdAction’s are executed in a subprocess (using python subprocess.Popen).

If action is a string, the command will be executed through the shell. (Popen argument shell=True).

Note that the string must be escaped according to python string formatting.

It is easy to include dynamic (on-the-fly) behavior to your tasks with python code from the dodo file. Let’s take a look at another example:

def task_hello():
    """hello cmd """
    msg = 3 * "hi! "
    return {
        'actions': ['echo %s ' % msg + ' > %(targets)s',],
        'targets': ["hello.txt"],
        }

Note

The body of the task-creator is always executed, so in this example the line msg = 3 * “hi! “ will always be executed.

If action is a list of strings, by default it will be executed without the shell (Popen argument shell=False).

def task_python_version():
    return {
        'actions': [['python', '--version']]
        }

For complex commands it is also possible to pass a callable that returns the command string. In this case you must explicit import CmdAction.

from doit.action import CmdAction

def task_hello():
    """hello cmd """

    def create_cmd_string():
        return "echo hi"

    return {
        'actions': [CmdAction(create_cmd_string)],
        'verbosity': 2,
        }

You might also explicitly import CmdAction in case you want to pass extra parameters to Popen like cwd. All keyword parameter from Popen can be used on CmdAction (except stdout and stderr).

Note

Different from subprocess.Popen, CmdAction shell argument defaults to True. All other Popen arguments can also be passed in CmdAction except stdout and stderr

The result of the task follows the shell convention. If the process exits with the value 0 it is successful. Any other value means the task failed.

custom actions

It is possible to create other type of actions, check tools.LongRunning as an example.

keywords on actions

It is common situation to use task information such as targets, dependencies, or changed in its own actions. Note: Dependencies here refers only to file-dependencies.

For cmd-action you can use the python notation for keyword substitution on strings. The string will contain all values separated by a space (” ”).

For python-action create a parameter in the function, doit will take care of passing the value when the function is called. The values are passed as list of strings.

def task_hello():
    """hello"""

    def python_hello(targets):
        with open(targets[0], "a") as output:
            output.write("Python says Hello World!!!\n")

    return {
        'actions': [python_hello],
        'targets': ["hello.txt"],
        }

You can also pass the keyword task to have a reference to all task metadata.

from __future__ import print_function
def who(task):
    print('my name is', task.name)
    print(task.targets)

def task_x():
    return {
        'actions': [who],
        'targets': ['asdf'],
        'verbosity': 2,
        }

Note

Note that the task argument is a Task object instance, not the metadata dict.

It is possible not only to retrieve task’s attributes but also to modify them while the action is running!

task name

By default a task name is taken from the name of the python function that generates the task. For example a def task_hello would create a task named hello.

It is possible to explicitly set a task name with the parameter basename.

def task_hello():
    return {
        'actions': ['echo hello']
        }

def task_xxx():
    return {
        'basename': 'hello2',
        'actions': ['echo hello2']
        }
$ doit
.  hello
.  hello2

When explicit using basename the task-creator is not limited to create only one task. Using yield it can generate several tasks at once. It is also possible to yield a generator that generate tasks. This is useful to write some generic/reusable task-creators.

def gen_many_tasks():
    yield {'basename': 't1',
           'actions': ['echo t1']}
    yield {'basename': 't2',
           'actions': ['echo t2']}

def task_all():
    yield gen_many_tasks()
$ doit
.  t2
.  t1

sub-tasks

Most of the time we want to apply the same task several times in different contexts.

The task function can return a python-generator that yields dictionaries. Since each sub-task must be uniquely identified it requires an additional field name.

def task_create_file():
    for i in range(3):
        filename = "file%d.txt" % i
        yield {'name': filename,
               'actions': ["touch %s" % filename]}
$ doit
.  create_file:file0.txt
.  create_file:file1.txt
.  create_file:file2.txt

avoiding empty sub-tasks

If you are not sure sub-tasks will be created for a given basename but you want to make sure that a task exist, you can yield a sub-task with name equal to None. This can also used to set the task doc and watch attribute.

import glob

def task_xxx():
    """my doc"""
    LIST = glob.glob('*.xyz') # might be empty
    yield {
        'basename': 'do_x',
        'name': None,
        'doc': 'docs for X',
        'watch': ['.'],
        }
    for item in LIST:
        yield {
            'basename': 'do_x',
            'name': item,
            'actions': ['echo %s' % item]
            }
$ doit
$ doit list
do_x   docs for X

Dependencies & Targets

One of the main ideas of doit (and other build-tools) is to check if the tasks/targets are up-to-date. In case there is no modification in the dependencies and the targets already exist, it skips the task execution to save time, as it would produce the same output from the previous run.

Dependency
A dependency indicates an input to the task execution.
Target
A target is the result/output file produced by the task execution.

i.e. In a compilation task the source file is a file_dep, the object file is a target.

def task_compile():
    return {'actions': ["cc -c main.c"],
            'file_dep': ["main.c", "defs.h"],
            'targets': ["main.o"]
            }

doit automatically keeps track of file dependencies. It saves the signature (MD5) of the dependencies every time the task is completed successfully.

So if there are no modifications to the dependencies and you run doit again. The execution of the task’s actions is skipped.

$ doit
.  compile
$ doit
-- compile

Note the -- (2 dashes, one space) on the command output on the second time it is executed. It means, this task was up-to-date and not executed.

file_dep (file dependency)

Different from most build-tools dependencies are on tasks, not on targets. So doit can take advantage of the “execute only if not up-to-date” feature even for tasks that don’t define targets.

Let’s say you work with a dynamic language (python in this example). You don’t need to compile anything but you probably want to apply a lint-like tool (i.e. pyflakes) to your source code files. You can define the source code as a dependency to the task.

def task_checker():
    return {'actions': ["pyflakes sample.py"],
            'file_dep': ["sample.py"]}
$ doit
.  checker
$ doit
-- checker

Note the -- again to indicate the execution was skipped.

Traditional build-tools can only handle files as “dependencies”. doit has several ways to check for dependencies, those will be introduced later.

targets

Targets can be any file path (a file or folder). If a target doesn’t exist the task will be executed. There is no limitation on the number of targets a task may define. Two different tasks can not have the same target.

Lets take the compilation example again.

def task_compile():
    return {'actions': ["cc -c main.c"],
            'file_dep': ["main.c", "defs.h"],
            'targets': ["main.o"]
            }
  • If there are no changes in the dependency the task execution is skipped.
  • But if the target is removed the task is executed again.
  • But only if it does not exist. If the target is modified but the dependencies do not change the task is not executed again.
$ doit
.  compile
$ doit
-- compile
$ rm main.o
$ doit
.  compile
$ echo xxx > main.o
$ doit
-- compile

execution order

If your tasks interact in a way where the target (output) of one task is a file_dep (input) of another task, doit will make sure your tasks are executed in the correct order.

def task_modify():
    return {'actions': ["echo bar > foo.txt"],
            'file_dep': ["foo.txt"],
            }

def task_create():
    return {'actions': ["touch foo.txt"],
            'targets': ["foo.txt"]
            }
$ doit
.  create
.  modify

Task selection

By default all tasks are executed in the same order as they were defined (the order may change to satisfy dependencies). You can control which tasks will run in 2 ways.

Another example

DOIT_CONFIG = {'default_tasks': ['t3']}

def task_t1():
    return {'actions': ["touch task1"],
            'targets': ['task1']}

def task_t2():
    return {'actions': ["echo task2"]}

def task_t3():
    return {'actions': ["echo task3"],
            'file_dep': ['task1']}

DOIT_CONFIG -> default_tasks

dodo file defines a dictionary DOIT_CONFIG with default_tasks, a list of strings where each element is a task name.

$ doit
.  t1
.  t3

Note that only the task t3 was specified to be executed by default. But its dependencies include a target of another task (t1). So that task was automatically executed also.

command line selection

From the command line you can control which tasks are going to be execute by passing its task name. Any number of tasks can be passed as positional arguments.

$ doit t2
.  t2

You can also specify which task to execute by its target:

$ doit task1
.  t1

sub-task selection

You can select sub-tasks from the command line specifying its full name.

def task_create_file():
    for i in range(3):
        filename = "file%d.txt" % i
        yield {'name': filename,
               'actions': ["touch %s" % filename]}
$ doit create_file:file2.txt
.  create_file:file2.txt

wildcard selection

You can also select tasks to be executed using a glob like syntax (it must contains a *).

$ doit create_file:file*
.  create_file:file1.txt
.  create_file:file2.txt
.  create_file:file3.txt

private/hidden tasks

If task name starts with an underscore ‘_’, it will not be included in the output.

title

By default when you run doit only the task name is printed out on the output. You can customize the output passing a “title” function to the task:

def show_cmd(task):
    return "executing... %s" % task.name

def task_custom_display():
    return {'actions':['echo abc efg'],
            'title': show_cmd}
$ doit
.  executing... Cmd: echo abc efg

verbosity

By default the stdout from a task is captured and its stderr is sent to the console. If the task fails or there is an error the stdout and a traceback (if any) is displayed.

There are 3 levels of verbosity:

0:
capture (do not print) stdout/stderr from task.
1 (default):
capture stdout only.
2:
do not capture anything (print everything immediately).

You can control the verbosity by:

  • task attribute verbosity
def task_print():
    return {'actions': ['echo hello'],
            'verbosity': 2}
$ doit
.  print
hello