docassemble interviews can be packaged into Python packages that bundle together:
- YAML files representing interview files that start interviews, or
YAML files that are
included in other files;
- Any document templates that your interviews use;
- Static content (images, downloadable files) that you want to include with your interview;
- Python modules, which include any classes and other code that you might write to go along with your interviews; and
- Data files (translations, machine learning training data) on which your interview depends.
A Python package also contains metadata such as author information, a brief description, and a list of other Python packages on which the package depends. It also contains a “README” file, typically styled with Markdown, which describes the package in detail.
A package containing docassemble interviews must be a subpackage
docassemble package. In practice, this means that the
official package name with
docassemble package itself is
just a shell (a namespace package) that contains subpackages. These
subpackages include docassemble’s core components
docassemble.webapp) as well as user-created
One of the features of the docassemble web application is the Playground, a browser-based development platform for authoring and testing interviews. This is a “sandbox” in which you can make changes to an interview and immediately test the effect of those changes. You can use the Playground to create a variety of resources (interview files, template files, static files, data files) and then build packages out particular resources.
Every interview that you run on docassemble is part of a package.
When you are running an interview in the Playground, the package is
docassemble.playground1 or something similar; the
1 at the
end is the user ID of the person who is using the Playground.
docassemble.playground1 should be considered temporary
and should only be used for development and testing. When you want
your users to actually use your interview, you should move your
interview out of the Playground and into a package with an
Suppose you have an interview file called
custody.yml, which you are
developing in the Playground. When you right click the “ Share”
button, and copy the link URL, or you look at your browser location
bar when you are testing the interview, you will see that the URL for
the interview contains
means that the interview (
i) is in the
package and is called
custody.yml. This link always runs the
version of your interview that is current in the Playground. So if
you make a change that accidentally introduces a bug, anyone who uses
the interview will see an error message.
When you have actual users, you want to be able to make “upgrades” to your interviews without causing users to see error messages. You need to have a distinction between your “development” version and your “production” version. Here is a workflow for accomplishing this:
- Develop your
custody.ymlinterview in the Playground and keep improving it until it is ready for use by users.
- In the “Packages” folder of the Playground, create a package
johnsmithlaw. Give it a version number like 0.0.1. Include within the package the
custody.ymlinterview file and any other resources on which
- This will create a Python package called
- In the “Packages” folder of the Playground, click the “Install”
button. This will install the
docassemble.johnsmithlawpackage on your server.
- Your users can now access your interview at a URL that ends with
- Now you can continue making changes to the
custody.ymlinterview in the Playground, and even if you break something, your users will not get an error, because they will be using the installed version of the package, not the version in your Playground.
- When the new version of your interview is ready, you can go to the “Packages” folder and press “Install” again.
- Your users will still use the same URL to access the interview (one
that ends with
/interview?i=docassemble.johnsmithlaw:data/questions/custody.yml), but now they will be using the new version of your interview.
From the “Packages” folder, you can also press “ Download” to obtain a ZIP file of the Python package. If you open this file, you can see what the structure of a docassemble extension package looks like.
Here is the file structure of a (fictional) docassemble package
The package is known as
docassemble.baseball within Python code,
but the name
docassemble-baseball, replacing the dot with a hyphen,
is used in other contexts, such as when referring to a package
published on the PyPI site, or a folder on your system that contains
the source code of the package.
There are a lot of subdirectories (this is the nature of namespace packages). There are reasons for all of these subdirectories.
- The top-level directory,
docassemble-baseball, is important because a complete Python package should be all in one directory. If you are publishing a package on GitHub, this directory should be the root of the repository;
docassemble-baseball/.gitwill contain the git-related information for the package.
- Within that, the
docassembledirectory is necessary so that the package is a subpackage of
- Within that, the
baseballdirectory is necessary because when packages within the
docassemblenamespace package are installed on a system, Python needs them to be in a subdirectory under a directory called
baseball, you have
baseballstats.py, which contains Python code. Files in this directory correspond with files in the “Modules” folder of the Playground. The
__init.py__file is necessary for declaring
baseballto be a package; you never have to edit that file. The
__init.py__file is empty but its presence is still important.
- There is also a
datadirectory with subdirectories
templates. These are for interviews, static files, data files, and document templates. The
questionsdirectory contains the YAML files that are in the main part of the Playground.
Note that in the Playground, there is a “Modules” folder along
with a “Templates” folder, a “Static” folder, etc., but in a
Python package, there is no
folder, even though there is a
folder and a
docassemble/baseball/data/static folder. The modules
.py files) are located in a different place: directly under
docassemble/baseball. Since the main purpose of a Python package
is to store Python modules, modules files are the “main attraction”
and everything else is just associated “data.” A Python module that
you refer to as
docassemble.baseball.baseballstats must be a file
baseballstats.py that is located in the subdirectory
docassemble/baseball/. (This is how Python works.)
When installed on the server, the interview
hitters.yml can be run
by going to a link like
In your own interviews, you can include resources from this package by writing things like the following:
The second block uses a file reference to refer to an image file in
data/static directory of the package.
Since a Python package is just a collection of files in a particular structure, you can maintain your docassemble extension packages “offline” (outside of the Playground) if you want. This allows you to expand the directory structure beyond what the Playground supports.
If your package uses code from other Python packages that are not distributed with the standard docassemble installation, you will need to indicate that these packages are “dependencies” of your package.
This will ensure that if you share your package with someone else and they install it on their system, the packages that your package needs will be automatically installed. Otherwise, that person will get errors when they try to use your package.
If you maintain your package in the Packages area of the Playground, you can indicate the dependencies of your package by selecting them from a multi-select list. (If you the Python package you need is not listed, you need to install it on your system using “Package Management” on the menu.)
If you maintain your package off-line, you will need to edit the
setup.py file and change the line near the end that begins with
install_requires. This refers to a list of Python packages. For
This line indicates that the package relies on the docassemble
docassemble.helloworld, as well as the Python
kombu. When someone tries to install
on their system,
kombu will be
installed first, and any packages that these packages depend on will
also be installed.
Note that if your package depends on a package that exists on GitHub
but not on PyPI, you will also need to add an extra line so that the
system knows where to find the package. For example, if
docassemble.helloworld did not exist on PyPI, you would need to
You can install a docassemble extension package, or any other Python package, using the docassemble web application.
From the menu, go to “Package Management.”
docassemble installs packages using the pip package manager. This installation process may take a long time. A log of the output of pip will be shown when the installation is complete. The server will restart so that any old versions of the package that are still in memory will be refreshed.
- Find the GitHub URL of the package you want to install. This is
in the location bar when you are looking at a GitHub repository.
For example, the GitHub URL of the
docassemble-baseballpackage may be
https://github.com/jhpyle/docassemble-baseball. (No such package actually exists.)
- In the docassemble web app, go to Package Management.
https://github.com/jhpyle/docassemble-baseballinto the “GitHub URL” field.
- The “GitHub Branch” field will be updated with the default branch
of the repository (usually
master). You can select another branch if you wish to install a different branch of the repository.
- Click “Update.”
You can also install Python packages from ZIP files. For example,
if you have a package
docassemble-baseball, the ZIP file
will be called
docassemble-baseball.zip. It will contain
a single directory called
docassemble-baseball, which in
setup.py, a subdirectory called
- In the docassemble web app, go back to Package Management.
- Under “Zip File,” upload the
.zipfile you want to install.
- Click “Update.”
You can also install Python packages from PyPI. PyPI is the
central repository for Python software. Anyone can register on
PyPI and upload software to it. For example, if you want to install
- Make sure the
docassemble-baseballpackage does not already exist on PyPI (note: it doesn’t; it is just a fictional package).
- In the docassemble web app, go to Package Management.
docassemble.baseballinto the “Package on PyPI” field.
- Click “Update.”
The “Package Management” system remembers which user installed which
package. Thus, if developer Fred installs a package called
docassemble.guardianship, and developer George tries to install a
docassemble.guardianship, George will get an error.
These permissions do not apply to package dependencies, however.
A user with
admin privileges can install or uninstall any package.
The purpose of the permission system is to facilitate the use of a
single server by a team of developers by implementing some minor
safeguards that prevent developers from interfering with each others’
work. It is not the case that each developer works in an insulated
“sandbox.” All interviews on the system run with the same system
Once a docassemble extension package is installed, you can start
using its interviews. For example, if you installed
docassemble.baseball, and there was an interview file in that
questions.yml, you would point your browser to
(substituting the actual domain and base URL of your docassemble
site). Note that a URL like this is different from the URL you see
when you are running an interview in the Playground (see
For more information about starting docassemble interviews, see how you run a docassemble interview.
To upgrade a package that you installed from a GitHub URL or from PyPI, you can press the “Update” button next to the package name on the “Package Management” screen. You will only see these Update buttons if you are an administrator or if you are the person who caused the packages to be installed. Also, the “Update” buttons will not appear if the package was installed using a ZIP file.
Then, go to “Configuration” on the menu and enable the PyPI publishing feature in docassemble configuration like so:
Press the button to publish the package to PyPI.
When the publishing is done, you will see an informational message with the output of the uploading commands. Check this message to see if there are any errors.
If the publishing was successful, then at the bottom of the page describing your package, you should see a message that the package now exists on PyPI.
You can click on the link to see what the package looks like on PyPI.
Now, on the docassemble menu (of this server or another server), you can go to Package Management and install the package by typing in “docassemble.baseball” into the “Package on PyPI” field.
You can publish your package on GitHub in addition to (or instead of) publishing it on PyPI. (Publishing on both sites is recommended. PyPI is the simplest and safest way to distribute Python packages, while GitHub is a version control system with many features for facilitating sharing and collaboration.)
When that configuration is done, each user who is a developer will need to connect their GitHub accounts with their accounts on your docassemble server. From the menu, the user should go to “Profile,” click “GitHub integration,” and follow the instructions. If the user is not currently logged in to GitHub in the same browser, GitHub will ask for log in information. (Users without GitHub accounts can create one.) Users will need to give consent to giving the docassemble server to have privileges of making changes to repositories and SSH keys within the GitHub account.
(Note: it is not possible to connect more than one docassemble
account on a single docassemble server with the same GitHub
account. However, it is possible to connect accounts on multiple
servers with the same GitHub account, so long as the
each docassemble server is different.)
To publish a package on GitHub, go to the Packages area of the Playground and press the button. You will be asked for a “commit message.” This is a brief, one-line message that describes the changes made to your package since the last time you “committed” changes. Each “commit” is like a snapshot, and the history of “commit” messages is a record of the development of your project.
When you press the “Commit” button after writing the commit message, your package will be “pushed” to a GitHub repository in your account. If a repository does not already exist on GitHub with the name of your package, a new repository will be created.
You can follow the hyperlink to your package’s page on GitHub.
After your first commit, GitHub reports that there have been two
commits; this is because the initial creation of the repository caused
a commit (containing a
LICENSE file only) and then the addition of
the files of your package caused a second commit.
It is also a good practice to separate your interview into at least three files, separately containing:
This way, other people can take advantage of your work product in interviews that might have a very different purpose.