Introduction

Unlike other guided interview systems, in which the interview developer maps out a decision tree or flowchart to indicate which questions should be asked and in which order, docassemble figures out what questions to ask and when to ask them based on rules that you specify. You specify these rules using YAML blocks.

Simple interviews: all blocks mandatory

The simplest type of rule you can specify is marking a block as mandatory.

mandatory: True
question: |
  Welcome to the interview!
continue button field: intro
---
mandatory: True
question: |
  What is your favorite fruit?
fields:
  - Fruit: favorite_fruit
---
mandatory: True
question: |
  What is your favorite vegetable?
fields:
  - Vegetable: favorite_vegetable
---
mandatory: True
question: |
  Here is your document.
attachment:
  content: |
    Your favorite fruit is ${ favorite_fruit }.

    % if favorite_fruit == 'apple':
    You will never need to see a doctor.

    % endif
    Your favorite vegetable is ${ favorite_vegetable }.

When docassemble runs an interview, it looks at the YAML and tries to run each block that is marked as “mandatory.” It will run them in the order in which they appear in the YAML. In this example, first the “Welcome to the interview!” question is asked. When the user clicks the “Continue” button, docassemble moves on to the second mandatory block, which asks “What is your favorite fruit?” When that question is answered, docassemble asks “What is your favorite vegetable?” When that question is answered docassemble moves on to the final question, “Here is your document,” which lets the user download a document. This is a very simple interview because there is no branching logic.

Suppose that instead of asking for the user’s favorite vegetable, you wanted to ask for the user’s favorite apple, but only if the user said that their favorite fruit is “apple.” In the previous interview, we set mandatory to True every time, but we can actually set mandatory to a Python expression that evaluates to a true or false value. For example:

mandatory: True
question: |
  Welcome to the interview!
continue button field: intro
---
mandatory: True
question: |
  What is your favorite fruit?
fields:
  - Fruit: favorite_fruit
---
mandatory: favorite_fruit == 'apple'
question: |
  What is your favorite type of apple?
fields:
  - Type of apple: favorite_apple
---
mandatory: True
question: |
  Here is your document.
attachment:
  content: |
    Your favorite fruit is ${ favorite_fruit }.

    % if favorite_fruit == 'apple':
    Your favorite type of apple is ${ favorite_apple }.

    % endif

Here, the “What is your favorite type of fruit” question is only “mandatory” if the user says that apple is their favorite type of fruit. Thus, if the favorite_fruit variable is 'banana', then docassemble will skip over the “What is your favorite type of apple?” question and proceed directly to the “Here is your document” question.

By setting the mandatory directive to a Python expression that uses variables defined in previous question blocks, you can write complex interviews that branch in a lot of different directions depending on the interview answers.

However, in a complex interview with a number of nested branches of logic, the Python expressions you will need to write to indicate whether a question should be asked could be very long and complicated. In the next section, we will discuss another way of implementing branching logic that avoids this complication.

Complex interviews: dependency satisfaction

As explained in the previous section, when docassemble runs your interview, it goes through your YAML from beginning to end and attempts to run each block that is marked as mandatory. Marking a question as “mandatory” is one way to tell docassemble you want a question to be displayed.

docassemble can also do “dependency satisfaction.” For example, you can write an interview like this:

mandatory: True
question: |
  Your favorite fruit is ${ favorite_fruit }.
subquestion: |
  % if favorite_fruit == 'grapes':
  Your favorite vineyard is ${ favorite_vineyard }.
  % endif
---
question: |
  Which vineyard do you think produces the best grapes?
fields:
  - Vineyard: favorite_vineyard
---
question: |
  What is your favorite fruit?
fields:
  - Fruit: favorite_fruit

In this interview, there is a mandatory question and then two questions that do not have a mandatory directive on them. If you run the interview, the first question asked is “What is your favorite fruit?” How did docassemble know it needed to ask that question, even though it was not marked as mandatory? What happened was that docassemble tried to display the “Your favorite fruit is …” question, but in the process of doing so, it encountered an undefined variable favorite_fruit. So then it looked for a block that defines the ‘favorite_fruit’ variable, and it found one, so it asked the “What is your favorite fruit?” question.

If the user types grapes in answer to that question, the interview asks a follow-up question, “Which vineyard do you think produces the best grapes?” and then proceeds to the final screen, which says “Your favorite fruit is grapes.” However, if the user says their favorite fruit is “apples,” the interview will skip the “Which vineyard do you think produces the best grapes?” question and will proceed directly to the final screen, which says “Your favorite fruit is apples.” Thus, with a single mandatory question, the interview does branching logic.

The branching logic is a by-product of the attempt to display the single mandatory question. If you were to change the text of the question and remove the reference to favorite_vineyard, the “Which vineyard do you think produces the best grapes?” question would never be asked. Or, if you were to change the question text to Your favorite fruit is ${ favorite_fruit } and your favorite vegetable is ${ favorite_vegetable }. then the order of questions would change, and the question that defines favorite_vegetable question would be asked right after the favorite_fruit question. When dependency satisfaction is used to ask questions, the order of questions is determined by which variables docassemble sees first.

Note that the order in which non-mandatory questions appear in the YAML does not affect the order in which questions are asked. Each block in in your YAML is a just a “rule,” and you can specify as many “rules” in your YAML as you want, in any order.

For example, the following block is a rule that indicates whether the user is eligible for a benefit.

code: |
  if user.age_in_years() >= 60 or user.is_disabled:
    user.eligible = True
  else:
    user.eligible = False

The rule says that the user is eligible if they are 60 or older, or if they are disabled, otherwise the user is not eligible.

Rules in docassemble are instructions for how to define a particular variable. The code block above is a rule for how to define user.eligible. question blocks are also rules. Here is a rule that specifies how to define user.is_disabled:

question: Are you disabled?
yesno: user.is_disabled

This says that the rule for defining user.disabled is to ask the user a yes/no question.

It is possible to specify rules in fairly complex ways, as we will see later; you can write multiple blocks that define the same variable, so you can have alternative rules for different circumstances, or you can have a general rule that is overridden by a more specific rule in certain circumstances. You can write generic rules that apply to a variety of different variables.

When docassemble runs your interview, it will try to run your mandatory blocks, in the order in which the blocks appear in your YAML. In the course of trying to run a block, docassemble might encounter a variable that hasn’t been defined yet. When this happens, docassemble will evaluate the “rules” you have defined, and it will run code blocks, question blocks, or other types of blocks in order to try to obtain a definition of the undefined variable.

In the course of trying to define a variable, docassemble might encounter yet another undefined variable, in which case it will try to obtain a definition of that variable, and in the course of trying to define that variable, it may encounter yet another undefined variable. A rule that defines a variable may “depend on” the values of other variables. docassemble’s logic engine will perform “dependency satisfaction” by automatically figuring out what variable definitions are necessary and running the appropriate code blocks or showing the appropriate question screens to the user.

This allows you, as the interview author, to specify rules and use variables in your interview or in your documents as you see fit, while docassemble does all the thinking about which questions need to be asked and in what order to ask them.

docassemble automatically refrains from asking unnecessary questions. For example, consider this example:

code: |
  if user.age_in_years() >= 60 or user.is_disabled:
    user.eligible = True
  else:
    user.eligible = False

If the user is 60 or older, there is no need to ask the user if they are disabled. It would waste the user’s time to ask that question. docassemble infers this from the rule. Thus “how to conduct the interview” and “what the legal rules are” are effectively the same thing, and can be specified in a single location.

Manually specifying the order of questions

Sometimes, you might not want the order of questions in the interview to be implicitly determined by the way docassemble processes rules; you might want to explicitly specify the order of questions. You can do this using a code block.

mandatory: True
code: |
  if favorite_fruit == 'grapes':
    favorite_vineyard
  favorite_vegetable
  final_screen
---
event: final_screen
question: |
  Your favorite fruit is ${ favorite_fruit }.
subquestion: |
  Your favorite vegetable is ${ favorite_vegetable }.

  % if favorite_fruit == 'grapes':
  Your favorite vineyard is ${ favorite_vineyard }.
  % endif
---
question: |
  Which vineyard do you think produces the best grapes?
fields:
  - Vineyard: favorite_vineyard
---
question: |
  What is your favorite fruit?
fields:
  - Fruit: favorite_fruit
---
question: |
  What is your favorite vegetable?
fields:
  - Vegetable: favorite_vegetable

Screenshot of dependency-demo-code example

In this interview, the mandatory code block drives the interview using dependency satisfaction, but in an explicit order. This block contains a few lines of Python code. The first variable encountered is favorite_fruit, which means that the favorite_fruit question will be asked. If favorite_fruit is 'grapes', then favorite_vineyard is evaluated, which means that favorite_vineyard will be asked. Then favorite_vegetable is asked, and then final_screen is sought. Because final_screen is a special screen, and the event directive is set to final_screen, the variable final_screen will actually not be defined; the screen is a dead-end screen with no fields and no “Continue” button.

If the mandatory code block was not present, and instead the final_screen block was marked mandatory, then the questions would have been asked in a different order: first favorite_fruit, then favorite_vegetable, and then favorite_vineyard (if the favorite_fruit was 'grapes'). We were able to instruct docassemble to ask for favorite_vineyard immediately after favorite_fruit by specifying different interview logic in the code block.

This mandatory code block serves as an “outline” for the interview. Instead of ordering blocks in your YAML, you can simply order lines in your mandatory code block. The code block lets you see the order of your interview at a glance, without having to page through a long interview. The indentation of text under if statements makes clear where there is a “branch” in the logic.

If you are familiar with Python, you might think that the mandatory code block is weird, because simply putting the name of a variable by itself on a line doesn’t do anything; it’s not something that programmers normally do. However, it does something in docassemble, because if the variable is undefined, a Python exception will be “raised,” and the raising of that exception will tell docassemble that a definition of that variable needs to be obtained. docassemble’s dependency satisfaction system operates through the triggering of undefined variable exceptions.

So far, we have discussed three different techniques for specifying interview logic in docassemble:

A series of question blocks with mandatory directives on them;
Allowing question blocks to be asked implicitly as a result of dependency satisfaction; and
Writing a code block marked as mandatory containing an explicit outline of the variables that need to be gathered and the conditions under which each variable definition should be sought.

These three techniques are not mutually exclusive; you can use them together. For example, you might have a mandatory question block followed by a mandatory code block, followed by a mandatory question block.

mandatory: True
question: |
  Welcome!
continue button field: intro_screen
---
mandatory: True
code: |
  intro_screen
  user.name.first
  final_screen
---
mandatory: True
question: |
  Your preferences.
subquestion: |
  Your favorite fruit is ${ favorite_fruit }.

  Your favorite vegetable is ${ favorite_vegetable }.

  % if favorite_vegetable == 'turnip' and user.grows_own_turnips:
  I grow turnips too!
  % endif

Or you could have a mandatory code block that only partially specifies the order of questions, and allows many questions to be asked explicitly. For example:

mandatory: True
code: |
  intro_screen
  user.name.first
  final_screen
---
question: |
  Welcome!
continue button field: intro_screen
---
event: final_screen
question: |
  Your preferences.
subquestion: |
  Your favorite fruit is ${ favorite_fruit }.

  Your favorite vegetable is ${ favorite_vegetable }.

  % if favorite_vegetable == 'turnip' and user.grows_own_turnips:
  I grow turnips too!
  % endif

Here, the mandatory code block ensures that intro_screen and user.name.first are asked up front, but then uses dependency satisfaction to trigger the asking of favorite_fruit and favorite_vegetable, as well as the display of the final_screen.

Writing law as code to drive the interview

docassemble’s “rules”-based logic system is particularly well-suited for legal applications. You can write legal logic in Python code, and docassemble will figure out how to ask the necessary questions to arrive at a legal judgment.

For example, suppose your interview needs to determine whether the user has legal standing as a grandparent to seek custody of a child. The relevant statute states that a grandparent can seek custody under the following circumstance:

(3) A grandparent of the child who is not in loco parentis to the child:

   (i) whose relationship with the child began either with the consent
       of a parent of the child or under a court order;

   (ii) who assumes or is willing to assume responsibility for the
        child; and

   (iii) when one of the following conditions is met:

      (A) the child has been determined to be a dependent child under
          42 Pa.C.S. Ch. 63 (relating to juvenile matters);

      (B) the child is substantially at risk due to parental abuse,
          neglect, drug or alcohol abuse or incapacity; or

      (C) the child has, for a period of at least 12 consecutive
          months, resided with the grandparent, excluding brief
          temporary absences of the child from the home, and is
          removed from the home by the parents, in which case the
          action must be filed within six months after the removal of
          the child from the home.

The interview developer can rewrite this statute in Python, converting each legal concept into a variable.

comment: 23 Pa. C.S.A. 5324(3)
code: |
  if relationship == 'Grandparent' \
     and (relationship_began_with_consent \
          or relationship_began_with_court_order) \
     and willing_to_assume_responsibility \
     and (child_is_dependent \
          or child_is_at_risk \
          or cared_for_child_for_a_year):
    has_grandparent_standing = True
  else:
    has_grandparent_standing = False

Note: if you are wondering why there are \ marks at the end of some of the lines, this is Python syntax for formatting source code and avoiding writing a very long line of code. If the \ was not present, there would be a syntax error, because Python would interpret the newline to mean that you were done specifying a condition, and it would think you forgot to write a : to indicate the end of the condition. The \ basically means “ignore the following newline and pretend this is all one long line.”

The values of a variable like relationship_began_with_consent could be determined by asking the user a question.

question: |
  At some point, did one of the child's parents agree to let you care
  for the child?
yesno: relationship_began_with_consent

Other variables, like cared_for_child_for_a_year, might be too complex to reduce to a single question. In that case, rather than using a question as the “rule” for what the variable means, you can use a code block instead, and you can break the legal concept down into smaller pieces.

code: |
  if (not child_lives_with_client) \
     and child_used_to_live_with_client \
     and child_taken_from_client_by_parent \
     and child_taken_within_last_six_months \
     and child_moved_in_at_least_12_months_before \
     and child_lived_with_client_continuously:
       cared_for_child_for_a_year = True
  else:
       cared_for_child_for_a_year = False

The rules for what these variables mean can in turn be specified as question or code blocks:

question: |
  Does the child currently live with you?
yesno: child_lives_with_client
---
code: |
  if as_datetime(date_child_taken_away) >= date_as_of_six_months_ago:
    child_taken_within_last_six_months = True
  else:
    child_taken_within_last_six_months = False

Given YAML rules like this, docassemble can automatically conduct a parsimonious interview; that is, it will not ask any unnecessary questions. For example, if willing_to_assume_responsibility is False, it will not ask child_is_dependent. The only thing you need to do to trigger this process is to set up a mandatory block that requires a definition of has_grandparent_standing.

mandatory: True
code: |
  if relationship == 'Grandparent' and not has_grandparent_standing:
    grandparent_not_eligible
  final_screen
---
event: grandparent_not_eligible
question: |
  Sorry, you do not have standing as a grandparent to seek custody
  under Pennsylvania law.

Many beginners find this style of rule-based logic specificiation to be confusing; they would rather specify exactly which questions are asked, and exactly what happens as a result of the answer to each question. However, when there are numerous legal rules and the interaction of the legal rules leads to a large number of possible scenarios, planning in advance the interview process for each one of these scenarios is time-consuming, and the work involved is mechanical rather than substantive. If you are going to offer users the ability to spot-edit any of their prior answers to questions in the middle of the interview, you will need to think about exactly which follow-up processes are necessary when the user makes such changes.

The “declarative” style of logic is very useful in these circumstances. All you need to do is the work of a lawyer – concentrate on specifying rules that are legally correct. You can specify multiple overlapping rules, covering special cases and general cases.

How rules determine interview process

Many people envision a guided interview as a process whereby an interviewee starts at the beginning screen, then moves through a series of screens and then arrives at the end screen of the interview. At any point in time, the interviewee is envisioned as being located at a certain “place” in the interview process.

However, when the interview is driven by rules, this way of envisioning the interview process is misleading. For example, consider the following structure for an interview:

Ask “What is your name?”
Ask “When were you injured?”
If the injury took place more than two years ago, say, “Sorry, the statute of limitations has expired, so you cannot file a complaint.”
Ask “Where did the injury take place?”
Ask “How much did you pay in medical bills?”
Ask “How much time did you have to take off from work?”
etc.
Here is a complaint you can file in court.

Suppose the user started the interview one day before the statute of limitations expired, and proceeded as far as the “How much did you pay in medical bills?” question, but then took a few days to locate their medical bills, and didn’t complete the interview until a week after the statute of limitations expired. Should the guided interview allow the user to download a complaint, or should it tell the user, “Sorry, the statute of limitations has expired, so you cannot file a complaint.”? If you think of the interview process as one where the interviewee is “located” at a particular “place” in the interview, you would say that since the user has “gone past” the part of the interview process that checked for a statute of limitations problem, the user should be allowed to proceed.

The philosophy behind docassemble is that a robust interview process is one where the “current question” in the interview is determined not by which question “comes after” the previous question, but rather by the application of a set of rules to a set of facts. If there is a legal rule about the statute of limitations, it should be applied every time the screen loads, not just once at the beginning of the interview.

In docassemble, the interview logic is envisioned more as a “checklist” than a process. Each time the screen loads, docassemble reviews the checklist. What it does next depends on the application of the checklist to the current state of affairs.

By analogy, an airplane pilot will go through a checklist prior to takeoff. Whether the airline pilot turns onto the runway or goes back to the gate depends on the application of the checklist to the state of the aircraft and external factors like the weather. Likewise, what docassemble does when the screen loads depends on the application of the interview logic (specified in the YAML) to the current state of the interview answers and external factors like the date.

From the user’s perspective, the docassemble interview process looks like something that has a beginning, an end, and a “current location,” but this is really just the by-product of docassemble running through a checklist every time the screen loads. “Do we know the user’s name? Check. Has the statute of limitations expired? Check. Do we know where the injury took place? Check.”

If you observe an airplane pilot at work, you might think, “gosh, the pilot spends so much time going through boring repetitive checklists, can’t he just grab the controls and fly the plane?” Similarly, you might look at what docassemble does every time the screen loads, and you might think, “ugh, why is it wasting time going through all of this logic, can’t it just move on to the next question?” Although the checklist method is repetitive, it is robust and is capable of catching hard-to-foresee problems.

The docassemble interview developer’s job is to design the checklist that leads to the interview process, not to specify the process directly. In most situations, this is a distinction without a difference, because the developer can write something like this:

mandatory: True
code: |
  user.name.first
  favorite_fruit
  favorite_vegetable
  final_screen

This is a checklist for what should be considered every time the screen loads: “if the name is not known, ask the name. If the favorite fruit is not known, ask for the favorite fruit. If the favorite vegetable is not known, ask for the favorite vegetable. Then show the ‘final_screen’ screen.” This translates directly into a process: “first ask for the name, then the favorite fruit, then the favorite vegetable, then show the final screen.”

In more complicated interviews, the connection between the checklist and the process is less explicit. For example:

mandatory: True
code: |
  user.name.first
  plaintiffs.gather()
  defendants.gather()
  if not jurisdiction_is_proper:
    kick_out_user
  final_screen

In this interview, after the user is asked for their name, docassemble gathers a list of plaintiffs and then gathers a list of defendants. The process of gathering groups is complex and involves multiple question blocks. The line plaintiffs.gather() is effectively a checklist item that means “make sure the plaintiffs are gathered.” Groups can be gathered in a variety of ways. The questions might be “What is the name of the first plaintiff?”, “Are there any other plaintiffs?”, “What is the name of the second plaintiff?”, “Are there any other plaintiffs?”, etc. The line if not jurisdiction_is_proper implicitly triggers the defining of jurisdiction_is_proper, which is defined by a code block.

By specifying a checklist, you can ensure the integrity of your interview’s logic, control the order of questions, and use to trigger the asking of questions that it would be too tedious to specify individually. There are things you need to think about, however, to ensure that your checklist results in a process that makes sense.

Beware of non-idempotency

When designing the checklist that docassemble runs every time the screen loads, you need to be careful about how you specify the checklist items. For example, you wouldn’t want the checklist to be the following:

Ask for the user’s name.
Ask for the user’s date of birth.
Give the user an assembled document.

That would mean that every time the screen loads, it would ask for the user’s name. Instead, the checklist should be:

If the user’s name is unknown, ask them for it.
If the user’s date of birth is unknown, ask them for it.
Give the user an assembled document.

When you write a checklist in Python format, it looks like this:

mandatory: True
code: |
  user.name.first
  user.birthdate
  final_screen

In docassemble, referencing the name of a variable like user.birthdate effectively means “if user.birthdate is undefined, stop what we are doing and seek out a definition of user.birthdate; otherwise, proceed to the next line.”

By contrast, if you use the force_ask() function, it will always ask the question:

mandatory: True
code: |
  force_ask('user.name.first')
  force_ask('user.birthdate')
  final_screen

Here, the first “checklist” item says that docassemble must ask a question to determine the value of user.name.first even if the variable is already defined. This is not what you want to do in a checklist; the user will be confused about why the interview asks for their name again when they just provided it. (This is is one of the reasons why the force_ask() function is rarely used.)

docassemble allows you to run Python functions inside of a checklist, and in most situations this works as expected. For example:

mandatory: True
code: |
  user.name.first
  user.birthdate
  if not record_exists_in_database_for(user):
    error_screen
  final_screen

Here, there is a function called record_exists_in_database_for() that looks up the user based on the user’s name and birthdate. It is ok if this function runs every time the screen loads.

However, beginning developers sometimes assume that they can do this:

mandatory: True
code: |
  user.name.first
  user.email
  send_email(to=user, subject="Welcome", body="Welcome to the interview!")
  user.birthdate
  final_screen

This means that after the user provides their name and e-mail address, docassemble will send them an e-mail. However, it also means that every time the screen loads thereafter, docassemble will send another e-mail! The checklist item should have been written in such a way that the e-mail is only sent once:

mandatory: True
code: |
  user.name.first
  user.email
  if not task_performed('welcome_email'):
    send_email(to=user, subject="Welcome", body="Welcome to the interview!", task='welcome_email')
  user.birthdate
  final_screen

The task_performed() function, combined with the task paramater of the send_email() function, is one way to ensure that code only runs once. Another method is to use a separate code block that defines a variable:

mandatory: True
code: |
  user.name.first
  user.email
  email_sent
  user.birthdate
  final_screen
---
code: |
  send_email(to=user, subject="Welcome", body="Welcome to the interview!")
  email_sent = True

The logic behind the email_sent line is: “if email_sent is not defined, run the code block in order to define it; otherwise continue to the next line.”

Another mistake that beginning developers sometimes make is writing a checklist that results in the user seeing a different screen if they refresh the screen without providing input. For example, consider this interview:

mandatory: True
code: |
  user.name.first
  user.email
  if not task_performed('data_stored'):
    store_data(user)
    mark_task_as_performed('data_stored')
    user.wants_email
    if user.wants_email:
      send_email(to=user, template=confirmation_email)
  user.birthdate
  final_screen
---
question: |
  Do you want a confirmation e-mail?
yesno: user.wants_email

In this example, after the user provides their e-mail address, docassemble will run the store_data() function, then it will mark the data_stored “task” as having been completed, and then it will see that user.wants_email is undefined, so it will ask the user if they would like to receive a confirmation e-mail. Suppose that the user, instead of answering the question, refreshes the screen. The interview logic will be evaluated again. Now, since the data_stored “task” has been marked as complete, the Python code skips the if clause and asks the user for user.birthdate. But this defeats the user’s expectation; the user reasonably expects that when they refresh the screen, they will see the user.wants_email question again. The problem is with the interview logic.

Software developers use the term “idempotent” to describe a system that produces the same result if an action is repeated. The interview logic in this circumstance is not idempotent because when it is repeated, a different result is produced.

Normally, your “checklist” should be designed to result in idempotent behavior. The exception would be if the passage of time has made the “current question” obsolete. For example, if the user started the interview before the statute of limitations period expired, and then tried to continue with the interview after the statute of limitations period expired, it would be reasonable for the user to see a different screen when they refreshed the screen.

Another consequence of non-idempotent logic is that users might see a pop-up message saying “Input not processed.” This is because of a security feature in docassemble: if the browser tries to submit input for a question that is different from what the current question is according to the interview logic, docassemble will reject the browser’s attempt to change the interview answers. In the example above, if the user clicked “Yes” or “No” in response to the question “Do you want a confirmation e-mail?”, the user would have seen an “Input not processed” error and been sent to the user.birthdate question.

To fix the idempotency problem, you could take the e-mail sending code out of the conditional statement:

mandatory: True
code: |
  user.name.first
  user.email
  if not task_performed('data_stored'):
    store_data(user)
    mark_task_as_performed('data_stored')
  user.wants_email
  if user.wants_email:
    send_email(to=user, template=confirmation_email)
  user.birthdate
  final_screen

Writing idempotent logic is also important because of the way that docassemble runs code blocks. Consider the following interview, which has a code block for calculating the user’s total income:

mandatory: True
question: |
  Tell me about your income and expenses.
fields:
  - Benefits income: benefits_income
    datatype: currency
  - Business income: business_income
    datatype: currency
  - Business expenses: business_expenses
    datatype: currency
---
mandatory: True
code: |
  total_income = 0.0
---
mandatory: True
code: |
  total_income = total_income + benefits_income
  total_income = total_income + net_business_income
---
code: |
  net_business_income = business_income - business_expenses
---
mandatory: True
question: |
  Your total income is ${ currency(total_income) }.

At first glance, this logic looks correct; the interview gathers information from the user, initializes total_income to zero, then adds the benefits income and the net business income to total_income. However, you will find that the calculation is incorrect; benefits_income will be counted twice.

The problem is that this code block is not idempotent:

mandatory: True
code: |
  total_income = total_income + benefits_income
  total_income = total_income + net_business_income

If this code runs more than once, the total_income will be increased each time. If you try to run this interview, this code block will run more than once. The first time it runs, it adds benefits_income to total_income, but then stops because net_business_income is undefined. docassemble obtains a definition of net_business_income in microseconds by running the code block that defines net_business_income. But after it does that, it does not resume where it left off (adding net_business_income to total_income). It will repeat the code block again, from the beginning. So benefits_income will be added to total_income a second time, and then net_business_income will be added, and then the “mandatory” block will be marked as having been completed, because it ran through all the way to the end.

If you are familiar with computer programming, docassemble works by trapping exceptions. When Python encounters an undefined variable, it raises an exception. docassemble traps that exception, figures out what variable was undefined, and then tries to define it. It does this either by asking the user a question or by running a ‘code’ block. Either way, the exception halts code execution, and Python is unable pick up exactly where it left off when the exception was raised.

The solution to this problem is to write the code block so that it can be run repeatedly without making a miscalculation:

mandatory: True
code: |
  total_income = 0
  total_income = total_income + benefits_income
  total_income = total_income + net_business_income

This way, the code will produce the correct total no matter how many undefined variables docassemble encounters along the way.

Inexperienced developers also sometimes make the error of assuming that all code blocks will run to completion. For example, suppose that the above interview was written like this, with only one mandatory block:

question: |
  Tell me about your income and expenses.
fields:
  - Benefits income: benefits_income
    datatype: currency
  - Business income: business_income
    datatype: currency
  - Business expenses: business_expenses
    datatype: currency
---
code: |
  total_income = 0.0
  total_income = total_income + benefits_income
  total_income = total_income + net_business_income
---
code: |
  net_business_income = business_income - business_expenses
---
mandatory: True
question: |
  Your total income is ${ currency(total_income) }.

This interview appears to be reasonable, but actually it contains a flaw. When docassemble tries to show the mandatory question, it encounters an undefined variable total_income, so it seeks out a definition of total_income. It tries to run this code block:

code: |
  total_income = 0.0
  total_income = total_income + benefits_income
  total_income = total_income + net_business_income

docassemble sets total_income to zero, and then encounters an undefined variable, benefits_income, so it asks the question that defines benefits_income. However, what if the user refreshed the screen on the question that asks for the benefits_income? docassemble would attempt to show the mandatory question again, and this time total_income is defined, so docassemble can display the screen, which says that the total income is zero. “But wait,” you say, “it didn’t finish running the code block that defines total_income!” True, but the rule of docassemble’s logic is that it goes through your YAML, runs mandatory blocks that haven’t been run before, and tries to obtain definitions for any undefined variables that are encountered along the way. Nothing in this rule says that it will remember if it left a code block early and go back to it.

The moral of the story is that if you are going to use dependency satisfaction, do not allow your dependencies to be satisfied prematurely. The code block should be written instead as:

code: |
  total_income = benefits_income + net_business_income

code: |
  total_income = benefits_income + net_business_income
  draft_total_income = 0.0
  draft_total_income = total_income + benefits_income
  draft_total_income = total_income + net_business_income
  total_income = draft_total_income
  del draft_total_income

You can think of the undefined-ness of a variable as docassemble’s incentive to obtain the definition of the variable. It is like hiring a busy contractor to do work on your house; if you give the contractor his final payment after he is only halfway done with the job, he might leave at the end of the day and forget to come back later to finish the work.

You might be tempted to combine the definition of several variables in a single code block, perhaps because you think it saves space or is easier to read:

code: |
  subtotal = 0
  for item in asset:
    if item.countable:
      subtotal = subtotal + item.value
  total_assets = subtotal
  temp_list = []
  for item in income:
    if item.included and item.type not in income.items:
      temp_list.append(item.type)
  income_items = temp_list

Think about what will happen if the interview needs a value of total_assets. It will run the code block, and halfway through, the value of total_assets will be obtained. But the code will not stop executing; it will go on to start building the income_items list. This will work fine if the income list has been completely gathered, but what if it has not been? Then the code block may result in the asking of a question about the income list, but if the user refreshes the screen, that question will go away. This introduces an idempotency problem.

It is a much better practice to separate your code into single-purpose code blocks:

code: |
  subtotal = 0
  for item in asset:
    if item.countable:
      subtotal = subtotal + item.value
  total_assets = subtotal
---
code: |
  temp_list = []
  for item in income:
    if item.included and item.type not in income.items:
      temp_list.append(item.type)
  income_items = temp_list

This way, no matter whether your interview needs total_assets first or income_items first, and regardless of whether it has already gathered income or asset, these code blocks will perform their function and deliver a definition without causing any non-idempotent questions to be asked.

As a general rule, Let each code block serve a single purpose, or a set of closely-related purposes, and let it deliver its award (the defining of the variable sought) on the last line. If you get into this habit, you will avoid hard-to-debug logic errors.

Although typically your non-mandatory code blocks should only set one variable at a time, it is ok if they set other variables incidentally. However, in that situation you should probably use the only sets modifier.

For example, suppose you have an interview in which you want to ask the user, “Do you receive income from public benefits?”, and you want to set the variable has_benefits to the answer, but as a double-check on the validity of the answer, you want to set has_benefits to True during the income gathering process if the user indicates that they have income from disability or welfare income.

question: |
  Do you receive income from public benefits?
yesno: has_benefits
---
code: |
  temp_total = []
  for item in income:
    if item.included and item.type not in income.items:
      temp_total.append(item.type)
    if item.type in ['disability', 'welfare']:
      has_benefits = True
  total_income = temp_total

The problem here is that the income gathering question will now be called upon to set has_benefits. If has_benefits is needed before total_income, the interview will start asking about income items, but will mysteriously stop in the middle of the process if the user enters a disability or welfare income item. Then, when total_income is needed later in the interview, it will resume asking questions about the income items. It would be better if you used only sets:

only sets: total_income
code: |
  temp_total = []
  for item in income:
    if item.included and item.type not in income.items:
      temp_total.append(item.type)
    if item.type in ['disability', 'welfare']:
      has_benefits = True
  total_income = temp_total

That way, the code block will only be called upon to define total_income. It can still have the side effect of setting has_benefits to True, but it will not be called upon to define anything other than total_income.

The logical order of an interview

In the previous sections, we have explained that when docassemble runs your interview, it goes through your YAML looking for blocks that are mandatory, and it tries to run them in order. When it encounters an undefined variable, it stops what it is doing and tries to obtain a definition of that undefined variable.

If a mandatory question is answered, or a mandatory code block’s Python code runs all the way through to end, then docassemble remembers that the mandatory block has been completed, and the next time it evaluates the interview logic, it will skip over the block.

(Technical note: how does docassemble remember that a block has been completed? It stores a variable in the interview answers, inside of a special dictionary called _internal. In order to identify the blocks that have been completed, it uses the block’s id. If you do not specify an id on a mandatory block, docassemble will generate an identifier like Question_0 or Question_1 for the first and second blocks in your YAML file. This means that if you have an interview that is “in production” and users have active sessions in that interview, and then you change the YAML to insert new blocks or move them around, you could cause these identifiers to change, and then users who started sessions before you changed the YAML could experience problems where questions they have already answered are re-asked. In order to avoid this problem, make sure to attach a unique id to each mandatory block in your interview. That way, even if you rearrange the YAML, users with existing sessions will not experience problems.)

In addition to mandatory, there is a second type of modifier you can use to force a code block to be processed. If you mark a code block with initial: True, then the block will be run every time the screen loads, even if it has run before. The block is “initial” in the sense that it initializes the interview logic that will be evaluated during the screen load.

mandatory and initial blocks are evaluated in the order they appear in the question file. Therefore, the location in the interview of mandatory and initial blocks, relative to each other, is important.

The order in which non-mandatory and non-initial questions appear is usually not important. If docassemble needs a definition of a variable, it will go looking for a block that defines the variable.

Consider the following example:

question: |
  Do you like penguins?
yesno: user_likes_penguins
---
question: |
  Do you like elephants?
yesno: user_likes_elephants
---
mandatory: true
question: |
  Hello!
buttons:
  - Hello to you: continue
---
mandatory: true
question: |
  What is your name?
fields:
  - Name: user_name
---
mandatory: true
question: |
  Your favorite food is
  ${ favorite_food }.

  % if user_likes_penguins:
  You are a fan of penguins.
  % else:
  You detest penguins, for some
  strange reason.
  % endif
---
question: |
  What is your favorite food?
fields:
  - Favorite food: favorite_food

The order of the questions is:

Hello!
What is your name?
What is your favorite food?
Do you like penguins?

The first two questions are asked because the corresponding question blocks are marked as mandatory. They are asked in the order in which they are asked because of the way the question blocks are ordered in the YAML file.

The next two questions are asked implicitly. The third and final mandatory block makes reference to two variables: favorite_food and user_likes_penguins. Since the questions that define these variables are not mandatory, they can appear anywhere in the YAML file, in any order you want. In this case, the favorite_food question block is at the end of the YAML file, and the user_likes_penguins question block is at the start of the YAML file.

The order in which these two questions are asked is determined by the order of the variables in the text of the final mandatory question. Since favorite_food is referenced first, and user_likes_penguins is referenced afterwards, the user is asked about food and then asked about penguins.

Note that there is also an extraneous question in the interview that defines user_likes_elephants; the presence of this question block in the YAML file has no effect on the interview.

Generally, you can order non-mandatory blocks in your YAML file any way you want. You may want to group them by subject matter into separate YAML files that you include in your main YAML file. When your interviews get complicated, there is no natural order to questions. In some situations, a question may be asked early, and in other situations, a question may be asked later.

Overriding one question with another

The order in which non-mandatory blocks appear in the YAML file is only important if you have multiple blocks that each offer to define the same variable. In that case, the order of these blocks relative to each other is important. When looking for blocks that offer to define a variable, docassemble will use later-defined blocks first. Later blocks “supersede” the blocks that came before.

This allows you to include “libraries” of questions in your interview while retaining the ability to customize how any particular question is asked.

As explained in the initial blocks section, the effect of an include block is basically equivalent to copying and pasting the contents of the included file into the original file.

For example, suppose that there is a YAML file called question-library.yml, which someone else wrote, which consists of the following questions:

question: |
  Nice evening, isn't it?
yesno: user_agrees_it_is_a_nice_evening
---
question: |
  Interested in going to the dance tonight?
yesno: user_wants_to_go_to_dance

You can write an interview that uses this question library:

include:
  - question-library.yml
---
mandatory: True
code: |
  if user_agrees_it_is_a_nice_evening and user_wants_to_go_to_dance:
    good_news
---
mandatory: True
question: |
  Say, I have to run.  Bye!
---
event: good_news
question: |
  That is splendid news!

Screenshot of use-question-library example

When docassemble needs to know the definition of user_agrees_it_is_a_nice_evening or user_wants_to_go_to_dance, it will be able to find a block in question-library.yml that offers to define the variable.

Suppose, however, that you thought of a better way to ask the user_wants_to_go_to_dance question, but you don’t want to get rid of question-library.yml entirely. You could override the user_wants_to_go_to_dance question in question-library.yml by doing the following:

include:
  - question-library.yml
---
question: |
  So, about that dance tonight . . .
  wanna go?
yesno: user_wants_to_go_to_dance
---
mandatory: True
code: |
  if user_agrees_it_is_a_nice_evening and user_wants_to_go_to_dance:
    good_news
---
mandatory: True
question: |
  Say, I have to run.  Bye!
---
event: good_news
question: |
  That is splendid news!

This interview file loads the two questions defined in question-library.yml, but then, later in the list of questions, provides a different way to get the value of user_wants_to_go_to_dance. When docassemble goes looking for a question to provide a definition of user_wants_to_go_to_dance, it starts with the questions that were defined last, and it will prioritize your question over the question in question-library.yml. Your question block takes priority because it is located later in the YAML file.

This is similar to the way law works: old laws do not disappear from the law books, but they can get superseded by newer laws. “Current law” is simply “old law” that has not yet been superseded.

A big advantage of this feature is that you can include “libraries” written by other people without having to edit those other files in order to tweak them. You can use another person’s work without taking on the responsibility of maintaining that person’s work over time; you can just incorporate by reference that person’s file, which they continue to maintain.

For example, if someone else has developed interview questions that determine a user’s eligibility for food stamps, you can incorporate by reference that developer’s YAML file into an interview that assesses whether a user is maximizing his or her public benefits. When the law about food stamps changes, that developer will be responsible for updating his or her YAML file; your interview will not need to change. This allows for a division of labor. All you will need to do is make sure that the docassemble package containing the food stamp YAML file gets updated on the server when the law changes.

Fallback questions

If a code block does not, for whatever reason, actually define the variable, docassemble will “fall back” to a block that is located earlier in the YAML file. For example:

include:
  - question-library.yml
---
question: |
  I forgot, did we already agree to go to the dance together?
yesno: we_already_agreed_to_go
---
code: |
  if we_already_agreed_to_go:
    user_wants_to_go_to_dance = True
---
mandatory: True
code: |
  if user_agrees_it_is_a_nice_evening and user_wants_to_go_to_dance:
    good_news
---
mandatory: True
question: |
  Say, I have to run.  Bye!
---
event: good_news
question: |
  That is splendid news!

In this case, when docassemble tries to get a definition of user_wants_to_go_to_dance, it will first try running the code block, and then it will encounter we_already_agreed_to_go and seek its definition. If the value of we_already_agreed_to_go turns out to be false, the code block will run its course without setting a value for user_wants_to_go_to_dance. Not giving up, docassemble will keep going backwards through the blocks in the YAML file, looking for one that offers to define user_wants_to_go_to_dance. It will find such a question among the questions included by reference from question_library.yml, namely the question “Interested in going to the dance tonight?”

This “fall back” process can also happen with special question blocks that use the continue option.

include:
  - question-library.yml
---
question: Which of these statements is true?
choices:
  - "I am old-fashioned":
      question: |
        My darling, would you do me the
        honor of accompanying me to
        the dance this fine evening?
      yesno: user_wants_to_go_to_dance
  - "I don't care for flowerly language": continue
---
mandatory: True
code: |
  if user_agrees_it_is_a_nice_evening and user_wants_to_go_to_dance:
    good_news
---
mandatory: True
question: |
  Say, I have to run.  Bye!
---
event: good_news
question: |
  That is splendid news!

In this case, the special continue choice causes docassemble to skip the question block and look elsewhere for a definition of user_wants_to_go_to_dance. docassemble will “fall back” to the version of the question that exists within question-library.yml. When looking for a block that offers to define a variable, docassemble starts at the bottom and works its way up.

(Note that questions using continue are of limited utility because they cannot use the generic object modifier or index variables. However, code blocks do not have this limitation.)

So, to recapitulate: when docassemble considers what blocks it must process, it goes from top to bottom through your interview YAML file, looking for mandatory and initial blocks; if a block is later in the file, it is processed later in time. However, when docassemble considers what question it should ask to define a particular variable, it goes from bottom to top; if a block is later in the file, it is considered to “supersede” blocks that are earlier in the file.

As explained below, however, instead of relying on relative placement of blocks in the YAML file, you can explicitly indicate which blocks take precedence over other blocks.

How docassemble runs your code

docassemble goes through your interview YAML file from start to finish, incorporating included files as it goes. It always executes initial code when it sees it. It executes any mandatory code blocks that have not been successfully executed yet. If it encounters a mandatory question that it has not been successfully asked yet, it will stop and ask the question.

If at any time it encounters a variable that is undefined, for example while trying to formulate a question, it will interrupt itself in order to go find the a definition for that variable.

Whenever docassemble comes back from one of these excursions to find the definition of a variable, it does not pick up where it left off; it starts from the beginning again.

Therefore, when writing code for an interview, you need to keep in mind that any particular block of code may be re-run from the beginning multiple times.

For example, consider the following code:

---
mandatory: True
code: |
  if user_has_car:
    user_net_worth = user_net_worth + resale_value_of_user_car
    if user_car_brand == 'Toyota':
      user_is_sensible = True
    elif user_car_is_convertible:
      user_is_sensible = False
---

The intention of this code is to increase the user’s net worth by the resale value of the user’s car, if the user has a car. If the code only ran once, it would work as intended. However, because of docassemble’s design, which is to ask questions “as needed,” the code actually runs like this:

docassemble starts running the code; it encounters user_has_car, which is undefined. It finds a question that defines user_has_car and asks it. (We will assume user_has_car is set to True.)
docassemble runs the code again, and tries to increment the user_net_worth (which we can assume is already defined); it encounters resale_value_of_user_car, which is undefined. It finds a question that defines resale_value_of_user_car and asks it.
docassemble runs the code again. The value of user_net_worth is increased. Then the code encounters user_car_brand, which is undefined. It finds a question that defines user_car_brand and asks it.
docassemble runs the code again. The value of user_net_worth is increased (again). If user_car_brand is equal to “Toyota,” then user_is_sensible is set. In that case, the code runs successfully to the end, and the mandatory code block is marked as completed, so that it will not be run again.
However, if user_car_brand is not equal to “Toyota,” the code will encounter user_car_is_convertible, which is undefined. docassemble will find a question that defines user_car_is_convertible and ask it. docassemble will then run the code again, the value of user_net_worth will increase yet again, and then (finally) the code will run successfully to the end.

The solution here is to make sure that your code is prepared to be stopped and restarted. For example, you could have a separate code block to compute user_net_worth:

---
mandatory: True
code: |
  user_net_worth = 0
  if user_has_car:
    user_net_worth = user_net_worth + resale_value_of_user_car
  if user_has_house:
    user_net_worth = user_net_worth + resale_value_of_user_house
---

Note that mandatory must be true for this to work sensibly. If this were an optional code block, it would not run to completion because user_net_worth would already be defined when docassemble came back from asking whether the user has a car.

How docassemble finds questions for variables

There can be multiple questions or code blocks in an interview that can define a given variable. You can write generic object questions in order to define attributes of objects, and you can use index variables to refer to any given item in a DAList or DADict (or a subtype of these objects). Which one will be used?

In general, if you have multiple questions or code blocks that are capable of defining a variable, docassemble will try the more specific ones first, and then the more general ones.

For example, if the interview needs a definition of fruit['a'].seed_info.tally['b'].molecules[4].name, it will look for questions that offer to define the following variables, in this order:

fruit['a'].seed_info.tally['b'].molecules[4].name
fruit[i].seed_info.tally['b'].molecules[4].name
fruit['a'].seed_info.tally[i].molecules[4].name
fruit['a'].seed_info.tally['b'].molecules[i].name
fruit[i].seed_info.tally[j].molecules[4].name
fruit[i].seed_info.tally['b'].molecules[j].name
fruit['a'].seed_info.tally[i].molecules[j].name
fruit[i].seed_info.tally[j].molecules[k].name

Then it will look for generic object blocks that offer to define the following variables, in this order:

x['a'].seed_info.tally['b'].molecules[4].name
x[i].seed_info.tally['b'].molecules[4].name
x['a'].seed_info.tally[i].molecules[4].name
x['a'].seed_info.tally['b'].molecules[i].name
x[i].seed_info.tally[j].molecules[4].name
x[i].seed_info.tally['b'].molecules[j].name
x['a'].seed_info.tally[i].molecules[j].name
x[i].seed_info.tally[j].molecules[k].name
x.seed_info.tally['b'].molecules[4].name
x.seed_info.tally[i].molecules[4].name
x.seed_info.tally['b'].molecules[i].name
x.seed_info.tally[i].molecules[j].name
x.tally['b'].molecules[4].name
x.tally[i].molecules[4].name
x.tally['b'].molecules[i].name
x.tally[i].molecules[j].name
x['b'].molecules[4].name
x[i].molecules[4].name
x['b'].molecules[i].name
x[i].molecules[j].name
x.molecules[4].name
x.molecules[i].name
x[4].name
x[i].name
x.name

Moreover, when docassemble searches for a generic object question for a given variable, it first look for generic object questions with the object type of x (e.g., Individual). Then it will look for generic object questions with the parent type of object type of x (e.g., Person). It will keep going through the ancestors, stopping at the most general object type, DAObject.

Note that the order of questions or code blocks in the YAML matters where the variable name is the same; the blocks that appear later in the YAML will be tried first. But when the variable name is different, the order of the blocks in the YAML does not matter. If your interview has a question that offers to define seeds['apple'] and another question that offers to define seeds[i], the seeds['apple'] question will be tried first, regardless of where the question is located in the the YAML.

Here is an example in which a relatively specific question, which sets veggies[i][1], will be used instead of a more general question, which sets veggies[i][j], when applicable:

question: |
  How much does the other
  ${ i } weigh?
fields:
  - Grams: veggies[i][1]
    datatype: number
---
question: |
  How much does the
  ${ ordinal(j) }
  ${ i } weigh?
fields:
  - Grams: veggies[i][j]
    datatype: number

Screenshot of nested-veggies-override example

These rules about which blocks are tried before other blocks can be overriden using the order initial block or the id and supersedes modifiers. You can use the if modifier to indicate that a given question should only be asked under certain conditions. You can use the scan for variables modifier to indicate that a question or code block should only be considered when looking to define a particular variable or set of variables, even though it is capable of defining other variables.

Specifiers that control interview logic

`mandatory`

By default, all blocks in an interview are optional; they will be called upon only if needed to retrieve the value of a variable. However, if all blocks are optional, the interview has nothing to do. You can use the mandatory modifier to indicate that a block must be run. The first mandatory block in your interview will be the starting point of the interview logic when the user first starts the interview.

Consider the following as a complete interview file:

---
question: What is the capital of Maine?
fields:
  - Capital: maine_capital
---
question: Are you sitting down?
yesno: user_sitting_down
mandatory: True
---
question: Your socks do not match.
mandatory: True
---

The interview will ask “Are you sitting down” and then it will say “Your socks do not match.” It will not ask “What is the capital of Maine?”

Another way to control the logic of an interview is to have a single, simple mandatory code block that sets the interview in motion.

For example:

---
mandatory: True
code: |
  if user_sitting_down:
    user_informed_that_socks_do_not_match
  else:
    user_will_not_sit_down
---
question: What is the capital of Maine?
fields:
  - Capital: maine_capital
---
question: Are you sitting down?
yesno: user_sitting_down
---
question: Your socks do not match.
sets: user_informed_that_socks_do_not_match
---
question: You really should have sat down.
subquestion: I had something important to tell you.
sets: user_will_not_sit_down
---

Here, the single mandatory block contains simple Python code that contains the entire logic of the interview.

If a mandatory specifier is not present within a block, it is as though mandatory was set to False.

The value of mandatory can be a Python expression. If it is a Python expression, the question or code block will be treated as mandatory if the expression evaluates to a true value.

mandatory: |
  favorite_food == "apples"
question: |
  You have good taste in food.
buttons:
  - Continue: continue
---
mandatory: True
question: |
  Thank you for your input.
---
question: |
  What is your favorite food?
fields:
  - no label: favorite_food

It is a best practice to tag all mandatory blocks with an id.

`initial`

The initial modifier causes the code block to be run every time docassemble processes your interview (i.e., every time the screen loads during an interview). mandatory blocks, by contrast, are never run again during the session if they are successfully “asked” once. docassemble executes the code in an initial block in the same way it executes the code of mandatory blocks, except that running to completion does not mean the block will not be executed again.

initial blocks should be used in the following situations:

“Initializing” the Python context in a multi-user interview depending on who the user is. For example, if your interview uses a variable user that should always refer to an Individual object corresponding to the user, you can write an initial block that looks at user_info().email to figure out who the logged-in user is.
When you are using the actions feature and you want to make sure the actions are processed only in particular circumstances.

Here is an example that illustrates how initial blocks work:

mandatory: True
code: |
  counter = 0
---
initial: True
code: |
  counter = counter + 1
---
question: |
  How many peaches do you have?
subquestion: |
  The value of the counter
  is ${ counter }.
fields:
  - no label: peaches
    datatype: integer
---
code: |
  fruits = peaches + pears
---
question: |
  How many pears do you have?
subquestion: |
  The value of the counter
  is ${ counter }.
fields:
  - no label: pears
    datatype: integer
---
question: |
  You have ${ fruits } pieces of fruit.
subquestion: |
  The value of the counter
  is ${ counter }.
mandatory: True

Note in this example that from screen to screen, the counter increments from 1 to 2 and then to 4. The counter does not count the number of screens displayed, but rather the number of times the interview logic was evaluated. The “passes” through the interview are:

The interview logic is evaluated, but the evaluation stops when the undefined variable fruit is encountered. The interview then tries to run the code block to get fruit, but encounters an undefined variable peaches, so it asks a question to gather peaches.
The interview logic is evaluated, but the evaluation stops when the undefined variable fruit is encountered. The interview then tries to run the code block to get fruit, but encounters an undefined variable pears, so it asks a question to gather pears.
The interview logic is evaluated, but the evaluation stops when the undefined variable fruit is encountered. The interview then runs the code block, and this time, fruit is successfully defined.
The interview logic is evaluated again, and the final question is displayed.

Like mandatory, initial can be set to True, False, or to Python code that will be evaluated to see whether it evaluates to a true or false value.

If your interview has a single mandatory code block and it is incapable of running to completion, then you don’t really need an initial block because you can put the logic that needs to run every time the screen loads at the beginning of that mandatory block.

`need`

The need specifier allows you to manually specify the prerequisites of a question or code block. This can be helpful for tweaking the order in which questions are asked.

mandatory: True
need:
  - number_of_years_old
  - favorite_animal
question: |
  Thank you for that information.
subquestion: |
  My favorite animal is
  ${ favorite_animal },
  too!

  % if number_of_years_old < 10:
  You're growing so fast.  Pretty
  soon you'll be driving a
  ${ favorite_color } car!
  % endif

In this example, the ordinary course of the interview logic would ask “What is your favorite animal?” as the first question. However, everyone knows that the first question you should ask of a child is “How old are you?” The need specifier indicates that before docassemble should even try to present the “Thank you for that information” screen, it should ensure that number_of_years_old old is defined, then ensure that favorite_animal, and then try to present the screen.

The variables listed in a need specifier do not have to actually be used by the question. Also, if your question uses variables that are not mentioned in the need list, docassemble will still pursue definitions of those variables.

If any of the variables listed under need are undefined, docassemble will obtain their definitions before processing the content of the question. For example, suppose you have the following question:

need:
  - favorite_fruit
question: |
  Your favorite apple is ${ favorite_apple }.
continue button field: fruit_verified

If both favorite_fruit and favorite_apple are undefined, the definition of favorite_fruit will be sought first.

What if you wanted favorite_fruit to be sought after favorite_apple? To do this, you can use the following special form of need:

need:
  post:
    - favorite_fruit
question: |
  Your favorite apple is ${ favorite_apple }.
continue button field: fruit_verified

In this case, the definition of favorite_fruit will be sought after all of the prerequisites of the question have been satisfied.

You can organize your need items into pre and post items:

need:
  pre:
    - favorite_vegetable
  post:
    - favorite_fruit

You can also include post among a list of other items:

need:
  - favorite_vegetable
  - post:
      - favorite_fruit

`depends on`

The depends on specifier indicates that if the listed variables change, the results of the question or code block should be invalidated.

question: |
  What is A?
fields:
  - A: a
    datatype: number
---
question: |
  What is the square of ${ a }?
fields:
  - B: b
    datatype: number
depends on:
  - a
---
code: |
  c = a + b
depends on:
  - a
  - b
---
event: review_screen
skip undefined: False
question: |
  Review your answers.
review:
  - label: Edit A
    field: a
    button: |
      A is ${ a }.
  - label: Edit B
    field: b
    button: |
      B is ${ b }.
  - note: |
      C is ${ c }.
---
mandatory: True
code: |
  need(a, b, c)
  review_screen

In this example, if the user goes through the interview to the end, but then edits a, then if and when a is changed to a different value, c and b will be undefined. The original value of b will be remembered, so that when the interview logic asks the question to define b, the original value will be presented as a default. When a is set, c is also undefined, so that when the interview logic requires a definition of c, the code block will be run to recompute the value of c.

If the user goes through the interview and then edits b, a change in b will trigger the invalidation of c.

The depends on specifier will also cause variables to be invalidated when they are changed by a code block.

question: |
  What is A?
fields:
  - A: a
    datatype: number
---
code: |
  b = a * a
depends on:
  - a
---
code: |
  c = a + b
depends on:
  - a
  - b
---
event: review_screen
skip undefined: False
question: |
  Review your answers.
review:
  - label: Edit A
    field: a
    button: |
      A is ${ a }.
  - note: |
      B is ${ b }.
  - note: |
      C is ${ c }.
---
mandatory: True
code: |
  need(a, b, c)
  review_screen

In this interview, the variable b is set by a code block. If the user edits a to a different value, the depends on specifier on the code block causes the code block to be re-run. The change in b causes the value of c to be invalidated. As a result, c is automatically updated when a changes.

Note that the depends on specifier results in invalidation when a variable is changed, not when it is defined. If a variable is undefined and is then defined, this is not considered a change for purposes of the depends on specifier. If a user presses Continue on a screen but does not change the value of a variable, the depends on logic is not triggered.

The depends on specifier can be used with iterator variables.

objects:
  - people: DAList.using(object_type=Individual, there_are_any=True, complete_attribute='complete')
---
question: |
  How much does ${ people[i] }
  get paid per pay period?
fields:
  - Amount: people[i].income
    datatype: currency
depends on:
  - people[i].pay_period
---
question: |
  How frequently does
  ${ people[i] } get paid?
fields:
  - no label: people[i].pay_period
    datatype: number
    choices:
      - Monthly: 12.0
      - Biweekly: 26.0
      - Semi-monthly: 24.0
      - Weekly: 52.0
---
code: |
  people[i].annual_income = people[i].income * people[i].pay_period
depends on:
  - people[i].pay_period
  - people[i].income
---
question: |
  What is the name of the
  ${ ordinal(i) } person?
fields:
  - First name: people[i].name.first
  - Last name: people[i].name.last
---
code: |
  people[i].name.first
  people[i].annual_income
  people[i].complete = True
---
question: |
  Is there another person?
yesno: people.there_is_another
---
field: review_screen
question: Review
subquestion: |
  ${ people.table }

  ${ people.add_action() }
---
table: people.table
rows: people
columns:
  - Name: row_item.name
  - Annual income: currency(row_item.annual_income)
edit:
  - name.first
  - pay_period
---
event: final_screen
question: Summary
subquestion: |
  % for person in people:
  ${ person }
  makes
  ${ currency(person.income) }
  ${ nice_number(person.pay_period) }
  times per year, for an annual income of
  ${ currency(person.annual_income) }.

  % endfor

  [Review](${ url_action('review_screen') })
---
mandatory: True
code: |
  people.gather()
  people.table
  final_screen

Screenshot of depends-on-iterator example

In this interview, the “Edit” button on the table only triggers the asking of the .pay_period question, but the depends on logic will cause the .income question to be re-asked, and the .annual_income amount to be re-calculated, if the .pay_period answer changes.

In situations where variables that are part of a list need to be invalidated when a variable that is not part of the same list item changes, the on change block can be used.

`reconsider`

The reconsider modifier can be used in two ways: it can be set to a list of variables, or it can be set to True.

Effect when set to a list of variable names

When you set reconsider to a list of variable names, then before the question is asked, the variables will be undefined (if they are defined at all), and then the definition of each variable will be sought again.

reconsider:
  - minutes_since_world_series
question: |
  It has been ${ minutes_since_world_series } minutes since
  your team won the world series.  Have you gotten over
  your excitement yet?
yesno: gotten_over_excitement

This can be useful when your question refers to a computed variable that might have become out-of-date since the last time it was computed.

Effect when set to `True`

If reconsider is set to True on a code block, then docassemble will always “reconsider” the values of any of the variables set by the block. That is, every time the interview is assembled (every time the screen loads) docassemble will forget about the value of any of the variables set by the code block.

You will want to set reconsider to True if your interview flow is such that you want docassemble to reconsider its definition of a variable based on information that might be gathered in the future.

For example, see if you can find the problem with the interview below.

---
code: |
  cat_food_cans_needed = number_of_cats * 4
---
question: |
  Does your neighbor's cat sometimes eat at your house?
subquestion: |
  To feed your own cat, you will need ${ cat_food_cans_needed } cans
  of cat food, but you might need more for your neighbor's cat.
buttons:
  - "Yes":
      code: |
        number_of_cats = number_of_cats + 1
        has_neighboring_cat = True
  - "No":
      code: |
        has_neighboring_cat = False
---
question: How many cats do you have?
fields:
  - Cats: number_of_cats
    datatype: integer
---
question: |
  To feed your cat
  % if has_neighboring_cat:
  and your neighbor's cat
  % endif
  you will need to buy ${ cat_food_cans_needed } cans of cat food.
sets: all_done
---
mandatory: True
code: all_done

The problem with this interview is that it will compute the number of cans of cat food needed when it says “To feed your own cat, you will need . . . cans of cat food,” but it will not increase the number of cans of cat food to account for later-acquired information (i.e. the fact that the neighbor’s cat comes over). Once cat_food_cans_needed has been defined once, docassemble will continue to use that definition whenever the interview calls for the definition of cat_food_cans_needed.

This problem can be fixed by adding reconsider: True to the code block:

---
code: |
  cat_food_cans_needed = number_of_cats * 4
reconsider: True
---

The reconsider modifier tells docassemble to always reconsider the variables in the code block. When the final screen comes up, docassemble will have forgotten about the earlier-defined value of cat_food_cans_needed and will therefore re-define the value by re-running the code block.

code: |
  cat_food_cans_needed = number_of_cats * 4
reconsider: True

The reconsider modifier is particularly important to use when you allow interviewees to go back and modify past answers using a review block. For more information about how to implement such features, see review, event, url_action(), process_action(), action_menu_item(), and menu_items.

docassemble also offers the reset initial block, which has the same effect as the reconsider modifier, but using a different way of specifying which variables should be reconsidered. Whether you use the reset initial block or the reconsider modifier is a question of what you consider to be more convenient and/or readable.

`undefine`

When you set undefine to a list of variable names, then before the question is asked, the variables will be undefined.

undefine:
  - favorite_foods
question: |
  What is your favorite fruit?
fields:
  - Favorite fruit: favorite_fruit
---
code: |
  favorite_foods = [favorite_vegetable, favorite_fruit]
---
mandatory: True
question: |
  Your favorite foods are
  ${ comma_and_list(favorite_foods) }.

This can be useful when you allow users to change their answers using review screens. Sometimes a change to one variable will invalidate answers to other questions, or to computations made by code blocks.

Combining multiple interviews into one

Using an umbrella YAML file

If you have multiple interviews and you want the user to choose which interview to run, you could offer the multiple interviews as a single interview, where there is an “umbrella” YAML file that includes the others.

For example:

include:
  - interview-fruit.yml
  - interview-vegetables.yml
  - interview-flowers.yml
---
question: |
  What interview would you like to do?
field: interview_choice
choices:
  - Fruit
  - Vegetables
  - Flowers
---
mandatory: true
code: |
  if interview_choice == 'Fruit':
    interview_fruit_done
  elif interview_choice == 'Vegetables':
    interview_vegetables_done
  elif interview_choice == 'Flowers':
    interview_flowers_done

Screenshot of umbrella-interview example

Note that this interview includes three separate YAML files. The controlling logic is the code block in the “umbrella” interview that pursues a different endpoint depending on the value of interview_choice.

The three interview files included are:

Note that these interview files contain everything needed for the interview except for any mandatory blocks that would define an interview endpoint; that function is reserved for the “umbrella” interview.

Using hyperlinks

There are other ways to offer users a choice of interviews. For example, you can use the interview_url() function with the i optional keyword parameter to point users from one interview to another:

mandatory: true
question: |
  What interview would you
  like to use?
subquestion: |
  * [Fruit]
  * [Vegetables]
  * [Flowers]

  [Fruit]: ${ interview_url(i='docassemble.demo:data/questions/interview-about-fruit.yml') }
  [Vegetables]: ${ interview_url(i='docassemble.demo:data/questions/interview-about-vegetables.yml') }
  [Flowers]: ${ interview_url(i='docassemble.demo:data/questions/interview-about-flowers.yml') }

Screenshot of interview-url-refer example

You might also offer these hyperlinks in the menu, using the menu_items special variable:

mandatory: True
code: |
  menu_items = [ {'url': interview_url(i='docassemble.demo:data/questions/interview-about-fruit.yml'),
                  'label': 'Fruit'},
                 {'url': interview_url(i='docassemble.demo:data/questions/interview-about-vegetables.yml'),
                  'label': 'Vegetables'}, 
                 {'url': interview_url(i='docassemble.demo:data/questions/interview-about-flowers.yml'),
                  'label': 'Flowers'} ]
---
mandatory: True
question: |
  Select an option from the
  menu in the upper right.

You can also use the dispatch configuration directive in combination with show dispatch link to allow the user to access a list of interviews available on your server by selecting “Available Interviews” from the menu.

A/B testing with redirects

The hyperlinks described in the previous subsection can also be used with the command() function to automatically redirect the user to a particular interview, for example for the purposes of A/B testing.

The following interview seamlessly redirects the user to either the demo interview or the example interview for the redact() function, depending on a computational coin flip.

mandatory: true
code: |
  import random
  if random.random() < 0.5:
    url = interview_url(i="docassemble.demo:data/questions/questions.yml")
  else:
    url = interview_url(i="docassemble.demo:data/questions/examples/redact-docx.yml")
  command('exit', url=url)

The use of 'exit' in the command() function is important here because it will cause this brief interview session to be deleted from the user’s list of interview sessions, since its sole purpose is to redirect the user.

An interview like this might also log some data for purposes of collecting metrics, perhaps using Redis. In the interviews being A/B tested, metrics could be logged using Redis or the Google Analytics feature.

Using multiple endpoints in a single interview

Another way to offer an “interview inside an interview” is to populate variables and then delete them.

objects:
  - user_global: Individual
  - user: Individual
  - endpoint: DADict
---
mandatory: True
code: |
  while True:
    del endpoint[user.goal]
    del user
---
question: |
  What interview would you like to complete?
field: user.goal
choices:
  - Fruit quiz: fruit
  - Vegetable quiz: vegetable
  - Legume quiz: legume
---
code: |
  user.name.first = user_global.name.first
  user.name.last = user_global.name.last
---
code: |
  user.age_category = user_global.age_category
---
question: |
  What is your name?
fields:
  - First name: user_global.name.first
  - Last name: user_global.name.last
---
question: |
  How old are you, ${ user }?
field: user_global.age_category
choices:
  - "Spring chicken"
  - "The music I like isn't cool anymore"
  - "Wiser than most"
---
question: |
  Fruit quiz results
subquestion: |
  Your favorite fruit is
  ${ user.favorite_fruit }.
field: endpoint['fruit']
continue button label: Try again
---
question: |
  Vegetable quiz results
subquestion: |
  % if user.age_category == 'Spring chicken':
  You are still growing, ${ user }, so you should eat lots of vegetables.
  % endif
  
  Your favorite vegetable is
  ${ user.favorite_vegetable }.
field: endpoint['vegetable']
continue button label: Try again
---
question: |
  Legume quiz results
subquestion: |
  % if user.knows_legumes:
  ${ user } definitely knows what a legume is.
  % else:
  You don't know what a legume is.
  % endif
  
  % if user.age_category == 'Spring chicken':
  You are still growing, ${ user }, so you should eat lots of legumes.
  % endif
  
  Your favorite legume is
  ${ user.favorite_legume }.

field: endpoint['legume']
continue button label: Try again
---
question: |
  What is your favorite fruit?
fields:
  Favorite fruit: user.favorite_fruit
---
question: |
  What is your favorite vegetable?
fields:
  Favorite vegetable: user.favorite_vegetable
---
question: |
  What is your favorite legume?
fields:
  Favorite legume: user.favorite_legume
---
question: |
  Which of these are legumes?
fields:
  - no label: user.legume_picks
    datatype: checkboxes
    choices:
      - Peas
      - Lentils
      - Mango
      - String beans
      - Potato
      - Peanut
---
code: |
  if user.legume_picks['Peas'] \
     and user.legume_picks['Lentils'] \
     and user.legume_picks['String beans'] \
     and user.legume_picks['Peanut'] \
     and not user.legume_picks['Mango'] \
     and not user.legume_picks['Potato']:
    user.knows_legumes = True
  else:
    user.knows_legumes = False

Screenshot of interview_in_interview example

The central logic of this interview is in the following code block:

mandatory: True
code: |
  while True:
    del endpoint[user.goal]
    del user

This is concise but cryptic, so it may be easier to understand what the interview is doing by writing out the variables for which Python will seek definitions, in the order in which Python will seek them:

mandatory: True
code: |
  while True:
    user.goal
    endpoint[user.goal]
    del endpoint[user.goal]
    del user

First, the interview asks for the goal (user.goal) – whether the user wants do an interview about fruit, vegetables, or legumes.

Next, it seeks an endpoint for that goal – a variable like endpoint['vegetable']. This results in the “sub-interview” being conducted. Once that endpoint is reached (e.g., when endpoint['vegetable'] is set to True by the final question of the “sub-interview”), then the variables endpoint['vegetable'] and user are deleted (using the Python del statement). Then the logic loops back around to where it began. At this point, user.goal will be undefined, because the entire variable user had been deleted. So the user will be presented with the “fruit, vegetable, or legume” choice again, and can choose to repeat the same “sub-interview,” or start a different “sub-interview.”

Note that an interview like this is different from an interview that concludes with a restart button. While a restart button wipes out all of the user’s answers, this interview retains some of the information that was gathered. It does so by using two objects to track information about the user: information that is permanent is stored in the user_global object, and information that is temporary is stored in the user object.

Note that the interview developer only uses the object user when writing questions that refer to characteristics of the user. The following code blocks assert that information about the user’s name and age should by defined by reference to attributes of the user_global object:

code: |
  user.name.first = user_global.name.first
  user.name.last = user_global.name.last
---
code: |
  user.age_category = user_global.age_category

This means that whenever the interview needs the definition of user.name.first, it will actually seek out user_global.name.first. If the user has been asked for their name before, no question needs to be asked; the code will take care of defining user.name.first and user.name.last. But other attributes, like user.favorite_fruit, are lost when the interview logic does del user. As a result, the interview will remember some answers and forget others.

Best practices for interview logic and organization

Use only a single mandatory code block for each interview, and put it at the top of the file after the initial blocks.

Don’t reinvent the wheel; include other people’s questions.
Share your questions, code, and templates with others.
To that end, keep your question blocks in a separate YAML file from your mandatory interview logic, so that other people can incorporate your questions without having to edit your work. Your main interview file would consist only of:
- A metadata block saying who you are and what your interview is for;
- A block to include your file of questions;
- Any interview help blocks;
- A default role block, if you use roles;
- Any initial code;
- Your mandatory code or questions that set your interview in motion.
include other people’s question files directly from their docassemble packages, rather than by copying other people’s files into your package. That way, when the other developers make improvements to their questions, you can gain the benefit of those improvements automatically.
Don’t invent your own scheme for variable names; follow conventions and replicate what other people are doing.
If other people are including your questions and code, avoid changing your variable names unnecessarily, or else you will “break” other people’s interviews. This does limit your autonomy somewhat, but the benefits for the community of interview developers more than make up for the loss of autonomy.

Interview logic