Slow table parsing for huge tables

[Trace2333]

`from docx.table import _Cell, Table
from docx.oxml.table import CT_Tbl

elif isinstance(child, CT_Tbl):
# table -> JSON
# DEBUG
table_obj = Table(child, parent)
list_table = [[k.text for k in j.cells] for j in table_obj.rows]
str_table = self.list_to_md_table(list_table)
yield str_table`

This is my current code reads Word tables and converts them to JSON, but performance degrades significantly when handling large tables — for example, a 9000-row × 10-column table takes too long to parse.

Is there a way to optimize or accelerate this process? Any suggestions for improving efficiency would be greatly appreciated!

[scanny]

I believe this problem stems from the table interface supporting mutation. That is, python-docx has no way of knowing whether you have mutated the table between calls to retrieve individual cells and takes a conservative stance which involves more "re-parsing" than one would want for performance purposes.

What would be nice is a ReadOnlyTable or TableSnapshot object which behaved like Table but allowed the user to "promise" they would not mutate the table in the middle of reading it, and got all cells as quickly as possible in a single parse.

That's not likely to happen anytime soon but you could create one of your own if you have the skills.

The general idea would be to use internal methods or an XPath expression to grab all the cells (or perhaps rows, and a separate XPath to grab its cells) and then construct the _Cell objects yourself from those element references.

I expect you would find the existing code here instructive for this purpose:
https://github.com/python-openxml/python-docx/blob/master/src/docx/table.py#L507

For example, you can get all the <w:tr> elements (rows) for a table with table._tbl.tr_lst. And you can get the <w:tc> elements in that row with tr.tc_lst. And you can create a _Cell with _Cell(tc, self.table).

There is the complication of handling merged cells, which depends on how much you want the structure of your output to resemble the underlying table. The code I linked shows how that's handled by default (repeat content in merged cells), and you can make up your own mind.