Create table instance not connected to a document with python-docx?

This is an X-Post related to Issue #1190 at python-docx upstream.

This is how I do create a table in a docx document currently.

doc = docx.Document()
tab = doc.add_table(rows=300, cols=5)

So the table object is "connected" to its parent document.

Is there a way to create a table object without having a connection to a parent object and add it to the document later? Somehow like this?

doc = docx.Document()
tab_one = docx.Table(rows=300, cols=5)
tab_two = docx.Table(rows=100, cols=3)

doc.add_table(tab_two)
doc.add_table(tab_one)

Or (as a workaround) can I move a table object from one document instance to another like this?

doc_temp = docx.Document()
tab = doc_temp.add_table(rows=300, cols=5)

doc_main = docx.Document()
doc_main.add_table(tab)

The background of my question is that I do create multiple tables with 100-300 rows and do formatting operations on each of its cells. So there is a lot of row and cell iterations going on which eat a lot of performance and time.

Doing this in multiprocessing where each worker has its own table object would speed up the process. I would like to create multiple tables in parallel and adding them to the document in a later step.

It is also clear that multiprocessing isn't the whole and best solution for a performance problem. Such a problem isn't solved just with adding more CPU resources into it. The algorithm itself should be optimized. For me the multiprocessing is just one step of the way to a better solution.

EDIT: As a real world example here you can see how I create docx-tables based on pandas.DataFrame objects.

No, not at the python-docx API level. The document-element objects you manipulate via the API maintain a reference into the overall connected graph of lxml objects representing a package part (like document, header, etc.) and allow you to edit that part, in situ. They are not free-standing components that you can compose and re-compose.

That said, lxml does allow what you're talking about, inserting a copy of a pre-formed XML subtree into an arbitrary position in an existing XML tree. So if you're willing to dig down, you could use python-docx to form the initial table subtree, then use lxml to copy it and insert it into other places in the XML.

python-docx can also help you locate where those go, at least getting you close. For example you could find a marker paragraph and get its <p> element using paragraph._p then use lxml to insert a <tbl> subtree as a sibling before or after the <p> element.

This is not for the faint of heart as it will require digging into the code and understanding lxml, but it's certainly a viable approach that's worked for others and a lot easier than accomplishing the same starting from a .docx file. A big part of the challenge is handling the unmarshalling of the .docx file into parts, each with their own XML tree and python-docx can handle all that for you, as well as re-marshalling the package (saving) once you've made your changes.