West Health Data Science Blog

Applied Data Science to Lower Healthcare Costs for Successful Aging

Mon 01 March 2021

Exploring fillable forms with PDFrw

Posted by Haw-minn Lu in Algorithms   

Introduction and Background

With the goal of providing a senior friendly vaccine experience, Gary and Mary West PACE which stood up a small senior oriented covid vaccine clinic desires to mitigate the amount of paperwork a frail senior is subjected to. Quite a lot of data is repeatedly asked for to make appointments, on consent forms and in reminder cards. While the idea of using pre-populated fillable PDF forms is a simple one, implementation is full of challenges as many common programmatic PDF tools do not properly work with filled forms. To meet the challenges, PDF forms have repeated fields with same name, checkboxes and radio buttons are used. Furthermore, to make life easier for the staff, PDF forms for multiple patients needed to be consolidated into a single PDF.

Programatically filling in PDF forms can be a quick and accurate way to disseminate forms. Bits and pieces can be found throughout the Internet and places like Stack Overflow. No single source provides a complete answer, however, this blog post in Medium by Vivsvaan Sharma is a good starting place. The blog post is long on python practices and a bit short on PDF details. Another useful resource is the PDF 1.7 specification, but it is well over 750 pages!.

The scope of this series applies to PDF 1.7 at this time no investigation as to whether techniques apply to PDF 2.0 or forms designed with Adobe LifeCycle. The forms are confirmed to work with Acrobat Reader, but not tested across the board with different readers.

We will cover the basics of using PDFrw to explore fillable forms and look at basic data types within PDFrw. We will first look at text form fields and discuss how to access fields that are repeated. We will then look at checkboxes, radio buttons, combo boxes and lists. Finally, we'll offer a solution for consolidating multiple filled forms into a single PDF while preserving the filled forms (and maintain editability).

Finding Your Way Around PDFrw and fillable forms

If you search the internet, including the above mentioned Medium blog post, you will find a snippet of code which might look like the following:

pdf = pdfrw.PdfReader(file_path)
for page in pdf.pages:
    annotations = page['/Annots']
    if annotations is None:
        continue

    for annotation in annotations:
        if annotation['/Subtype']=='/Widget':
            if annotation['/T']:
                key = annotation['/T'].to_unicode()
                print (key)

The type of annotation['/T'] is pdfString while some sources use [1:-1] to extract the string from pdfString the .to_unicode() method is the proper way to extract the string. According to the PDF 1.7 specification § 12.5.6.19 all fillable forms use are widget annotation, so the check for the annotation['/SubType'] filters the annotation to only widget annotations.

As an example take the file sample_form1.pdf located at here at github. It has a fields for a person to enter their name and address information, plus we added a checkbox for an called opt in. If we run the segment of code on this file, the output looks like:

Name
Address
City
State
Zip
Opt in

So in this way we can figure out which annotation contains the fields All we have to do is set the value. Well almost as we'll explain below. To set the value, first we need to create a PDFString with our value with the encode method then update the annotation as shown in this code snippet.

annotation.update(pdfrw.PdfDict(V=pdfrw.objects.pdfstring.PdfString.encode(value)))

This converts your value into a PdfString and updates the annotation creating a value for. annotation['/V'].

As mentioned above, this won't quite do it. At the top level of your PdfReader object pdf you also need to set the NeedAppearances property in the interactive from dictionary, AcroForm (See § 12.7,2). Without this, the fields are updated but will not necessarily display. In our example, the corresponding snippet of code is

pdf.Root.AcroForm.update(pdfrw.PdfDict(NeedAppearances=pdfrw.PdfObject('true')))

To recap, we can write a simple form filler with the following block of code for text fields. We will discuss checkboxes below.

def form_filler(in_path, data, out_path):
    pdf = pdfrw.PdfReader(in_path)
    for page in pdf.pages:
        annotations = page['/Annots']
        if annotations is None:
            continue

        for annotation in annotations:
            if annotation['/Subtype'] == '/Widget':
                key = annotation['/T'].to_unicode()
                if key in data:
                    pdfstr = pdfrw.objects.pdfstring.PdfString.encode(data[key])
                    annotation.update(pdfrw.PdfDict(V=pdfstr))
        pdf.Root.AcroForm.update(
            pdfrw.PdfDict(NeedAppearances=pdfrw.PdfObject('true')))
        pdfrw.PdfWriter().write(out_path, pdf)

Multiple Fields with Same Name

For text fields: Simple? Well not quite. Take a look at sample_form2.pdf. This form differs from sample_form1.pdf only in that the Name field appears twice. Once in the information block at the top where name and address information is filled out, but also at the bottom in the signature block. If we were to run the first code block to examine the annotations, we get the following:

Address
City
State
Zip
Opt in

So what happened to the Name field. Turns out whenever the multiple fields occur with the same name the situation is more complicated. One way to deal with this is to simply rename the fields to be different such as Name-1 and Name-2, which is fine if the sole use of the form is for automated form filling. However, if the form is also to be used for manual filling, this would require the user to enter the Name multiple times.

With a little sleuthing and experimentation, you will see that there are some widget annotations without the /T field but with a /Parent field. As it turns out this /Parent contains the field name \T as well as the default value \V. So for our examples there is one \Parent and two \Kids. With a simple modification to our code by inserting the lines:

if not annotation['/T']:
    annotation=annotation['/Parent']

That can allow us to inspect and modify annotations that appear more than once. With this modification, the result of our inspection code yields:

pdf = pdfrw.PdfReader(file_path)
for page in pdf.pages:
    annotations = page['/Annots']
    if annotations is None:
        continue

    for annotation in annotations:
        if annotation['/Subtype']=='/Widget':
            if not annotation['/T']:
                annotation=annotation['/Parent']
            if annotation['/T']:
                key = annotation['/T'].to_unicode()
                print (key)

It should be noted that Name now appears twice, once for each instance, but they both point to the same /Parent. With this modification, the form filler will actually fill the /Parent value twice, but this has no impact since it is overwriting the default value with the same value while keeping the code simple.

Checkboxes

In accordance to §12.7.4.2.3 the you can set the checkbox state as follows:

def checkbox(annotation, value):
    if value:
        val_str = pdfrw.objects.pdfname.BasePdfName('/Yes')
    else:
        val_str = pdfrw.objects.pdfname.BasePdfName('/Off')
    annotation.update(pdfrw.PdfDict(V=val_str))

This will work especially when the export value of the checkbox is Yes, but doesn't need to be. The easiest solution if you designed the form or can use acrobat to edit the form is to ensure that the export value of the checkbox is Yes and the default state of the box is unchecked. In fact the recommendation in the specification is that it be set to Yes. However, you may not have the luxury and upon closer inspection of a form where the export value is not set to Yes. You will see that the /V and /AS fields are set to the export value not Yes.

If you are using the form not only for automatic filling but also for manual filling you may wish the box to be checked as a default. In that case, while the code does work, we feel the the best solution is to delete the /V as well as the /ASfield from the dictionary. If you do not have acrobat and can not find the export value, you can discover it by looking at appearance dictionary /AP and specifically at the /N field. Each annotation has up to 3 appearances in it's appearance dictionary /N, /R and /D, standing for normal, rollover, and down (§12.5.5). The latter two has to do with appearance in interacting with the mouse, the normal appearance has to do with how the form is printed. You can find the export value by examining the normal appearance keys are set to the export value and /Off. For example if you run,

print (annotation['/AP']['/N'].keys())

you will get annotations something like

['/Export', '/Off']

where Export is the export value. The resulting sophisticated version of function should be more this.

def checkbox(annotation, value, export=None):
    if export:
        export = '/' + export
    else:
        keys = annotation['/AP']['/N'].keys()
        keys.remove('/Off')
        export = keys[0]
    if value:
        annotation.update(pdfrw.PdfDict(V=export, AS=export))
    else:
        del annotation['/V']
        del annotation['/AS']

According to the PDF specification for checkboxes, the appearance stream /AS should be set to the same value /V. Failure to do so may mean in some circumstances the checkboxes do not appear. It should be noted that there isn't really strict enforcement within PDF readers, so it is best not to tempt fate and enter a value other than the export value for a checked value. Additionally, all these complicated machinations with the appearance dictionary come into play when dealing with more complex form elements.

More Complex Forms

For the purpose of the vaccine clinic application, filling text fields and checkboxes along with the discussion of consolidation files below are sufficient. However, in the interest of not leaving a partial solution. We'll take this topic further and address filling in all other form fields.

Radio Buttons

Radio buttons are by far the most complex of the form entries types. Each widget links to /Kids which represent the other buttons in the radio group. But each widget in a radio group will link to the same 'kids'. Much like the 'parents' for the repeated forms fields with the same name, you need only update each once, but it can't hurt to apply the same update multiple times if it simplifies your code.

In a nutshell, the value /V of each widget in a radio group needs to be set to the export value of the button selected. In each kid, the appearance stream /AS should be set to /Off except for the kid corresponding to the export value. In order to identify the kid with its corresponding export value, we need to look again to the /N field of the appearance dictionary /AP just as was done with the checkboxes.

The resulting code could look like the following:

def radio_button(annotation, value):
    for each in annotation['/Kids']:
        # determine the export value of each kid
        keys = each['/AP']['/N'].keys()
        keys.remove('/Off')
        export = keys[0]

        if f'/{value}' == export:
            val_str = pdfrw.objects.pdfname.BasePdfName(f'/{value}')
        else:
            val_str = pdfrw.objects.pdfname.BasePdfName(f'/Off')
        each.update(pdfrw.PdfDict(AS=val_str))

    annotation.update(pdfrw.PdfDict(V=pdfrw.objects.pdfname.BasePdfName(f'/{value}')))

Combo Boxes and Lists

Both combo boxes and lists are forms of the choice form type. The combo boxes resemble drop down menus and lists are similar to list pickers in HTML. Functionally they are very similar as for form filling. The value /V and appearance stream /AS need to be set to their exported values. The /Op yields a list of lists associating the exported value with the value that appears in the widget. For example if you examine the /Opt field of the gender form field in sample-form3.pdf you'll see:

[['(Male-Value)', '(Male)'], ['(Female-Value)', '(Female)']]

where the exported values Male-Value and Female-Value are associated with the text Male and Female.

To set the combo box, you simply need to set the value to the export value.

def combobox(annotation, value):
    export=None
    for each in annotation['/Opt']:
        if each[1].to_unicode()==value:
            export = each[0].to_unicode()
    if export is None:
        raise KeyError(f"Export Value: {value} Not Found")
    pdfstr = pdfrw.objects.pdfstring.PdfString.encode(export)
    annotation.update(pdfrw.PdfDict(V=pdfstr, AS=pdfstr))

Lists are structurally very similar. The list of exported values can be found in the /Opt field. The main difference is that lists based on their configuration can take multiple values. Multiple values can be set with Pdfrw by setting \V and \AS to a list of PdfStrings. We write it as separate helpers, but of course, you could combine the code into one function.

def listbox(annotation, values):
    pdfstrs=[]
    for value in values:
        export=None
        for each in annotation['/Opt']:
            if each[1].to_unicode()==value:
                export = each[0].to_unicode()
        if export is None:
            raise KeyError(f"Export Value: {value} Not Found")
        pdfstrs.append(pdfrw.objects.pdfstring.PdfString.encode(export))
    annotation.update(pdfrw.PdfDict(V=pdfstrs, AS=pdfstrs))

Putting it all together

Now that we have shown how to fill in all the specific types of form elements in a PDF field. (With the exception of the signature form, which probably should not be filled programatically). Let's put this all together. If you have access to the forms themselves, you will know what type of form field each corresponds to each label. However, it would be nice to be able to determine the field type and work appropriately.

Determining Form Field Types Programmatically

To address the missing ingredient, it is important to understand that fillable forms fall into 4 form types, button (push button, checkboxes and radio buttons), text, choice (combo box and list box) and signature. They correspond to following values of the /FT form type field of our annotation, /Btn, /Tx, /Ch and /Sig, respectively. We will omit the signature type as we do not support filling in signature. Furthermore, the push button is a widget which can cause an action but is not fillable.

To distinguish the types of buttons and choices, we can examine the form flags /Ff field. For radio buttons, the 16th bit is set. For combo box the 18th bit is set. Please note that annotation['/Ff'] returns a PdfObject when returned and must be coerced into an int for bit testing.

def field_type(annotation):
    ft = annotation['/FT']
    ff = annotation['/Ff']

    if ft == '/Tx':
        return 'text'
    if ft == '/Ch':
        if ff and int(ff) & 1 << 17:  # test 18th bit
            return 'combo'
        else:
            return 'list'
    if ft == '/Btn':
        if ff and int(ff) & 1 << 15:  # test 16th bit
            return 'radio'
        else:
            return 'checkbox'

For completeness, we should present a text_form filler helper.

def text_form(annotation, value):
    pdfstr = pdfrw.objects.pdfstring.PdfString.encode(value)
    annotation.update(pdfrw.PdfDict(V=pdfstr, AS=pdfstr))

So now we have all the building blocks to put an automatic form filler together. The finished form filler can be found in our github repository at.

Consolidating Multiple Filled Forms

There are two problems with consolidating multiple filled forms. The first problem is that when two PDF files are merged matching names are associated with each other. For instance, if I had John Doe entered in one form and Jane Doe in the second, when I combine them John Doe will override the second form's name field and John Doe would appear in both forms. The second problem is that most simple command line or programmatic methods of combining two or more PDF files lose form data. One solution is to "flatten" the each PDF file. This is equivalent to printing the file to PDF. In effect, this bakes in the filled form values and does not permit the editing the fields. Going even further, one could render the PDFs as images if the only requirement is that the combined files be printable. However, at the surface tools like ghostscript and imagemagick don't do a good job of preserving form data. Other tools like PDFUnite don't solve any of these problems.

Form Field Name Collisions

In our use case of the vaccine clinic, we have the same form being filled out for multiple patients. So to combine a batch of these requires all form field names to be different. The solution is quite simple, in the process of filling out the form using the code above, we can also rename (set) the value of /T.

def form_filler(in_path, data, out_path, suffix):
    pdf = pdfrw.PdfReader(in_path)
    for page in pdf.pages:
        annotations = page['/Annots']
        if annotations is None:
            continue

        for annotation in annotations:
            if annotation['/SubType'] == '/Widget':
                key = annotation['/T'].to_unicode()
                if key in data:
                    pdfstr = pdfrw.objects.pdfstring.PdfString.encode(data[key])
                    new_key = key + suffix
                    annotation.update(pdfrw.PdfDict(V=pdfstr, T=new_key))
        pdf.Root.AcroForm.update(
            pdfrw.PdfDict(NeedAppearances=pdfrw.PdfObject('true')))
        pdfrw.PdfWriter().write(out_path, pdf)

So all you have to do is supply a unique suffix to each form. In our case, we simply number the batch so the suffix is just a sequential number.

Combining the files

If you search the internet for combine PDF files using pdfrw, you'll get a recipe like the following.

writer = PdfWriter()
for fname in files:
    r = PdfReader(fname)
    writer.addpages(r.pages)
writer.write("output.pdf")

While you don't lose the form data per se, you lose rendering information and hence the combined PDF fails to show the fields. The problem comes from the fact that the written PDF does not have an interactive form dictionary (see §12.7.2 of the PDF 1.7 specification). In particular the interactive forms dictionary contains the boolean NeedAppearances to be set in order for fields to be shown. If the forms being combined have different interactive form dictionaries, they will need to be merged. This is beyond the scope of this blog and the issues has not really been addressed though the stackoverflow article endeavors to make a first attempt. For our purposes since the source form is identical amongst the various copies, any AcroForm dictionary can be used.

After obtaining the dictionary, from pdf.Root.AcroForm (assuming the reader is stored in pdf), it is not clear how to add it to the PdfWriter object. The clue comes from a simple recipe for copying a pdf file.

pdf = PdfReader(in_file)
PdfWriter().write(out_file, pdf)

If one examines, these source code, the second parameter is set to the attribute trailer, so assuming acro_form contains the interactive forms PdfDict you can set it by writer.trailer.Root.AcroForm = acro_form.

Double Sided Printing

This issue is less about pdfrw than just making sure certain issues are taken into consideration. For example, in the clinic example, we wanted to pre-populate vaccination cards provided from a digital template from the CDC. Each card has a front and back but there are 6 cards per sheet. You must insure the information for the front of one card corresponds to the correct back of the card. In our case with 6 cards, if the front is arrayed like this:

1 3
3 4
5 6

Then the back needs to be arranged as follows provided you are printing with the "flip along the long side" option which you should use unless the back is intentionally printed upside down relative to the front.

2 1
4 3
6 5

Another consideration when printing double sided is that if you document has an odd number of pages you will need to insert a blank page somewhere to make the page count even. Otherwise, when you print a bundle, the first page of the second document will be on the back of last page of the first.

Blank Page

While the pdfrw library does not have a method for producing a blank page, but there is a simple recipe taken from this article.

def blank_page():
    blank = pdfrw.PageMerge()
    blank.mbox = [0, 0, 612, 792] # 8.5 x 11
    blank = blank.render()
    return blank

Please note that the mbox specifies the page dimensions in points. In our case, all documents are letter size. Of course, it is easy to turn blank_page into a function which takes page size as a parameter and generates a blank page. It should also be noted that the pages attribute of the PdfReader object returned by PdfReader is merely a list so it is easy to insert the blank page where desired. In our particular example, the last two pages were meant to be printed back to back so we needed to insert the blank page before the last two. The code ended up looking like the following

pages = PdfReader(doc_name).pages
if len(pages)%2==1 and double_sided: # Odd number of pages
    writer.addpages(pages[0:-2])
    writer.addpage(blank_page())
    writer.addpages(pages[-2:])

where double_sided was a flag indicating whether we wanted to go into "double sided mode."

Conclusion

A complete functional version of this PDF form filler can be found in our github repository. This process was able to produce large quantities of pre filled forms for seniors seeking Covid vaccinations relieving one of the bottlenecks that have plagued many other vaccine clinics.