Detect article-like pages entirely in two columns when generating markdown #1269
Conversation
| elems.sort(key=generate_elem_prefer_left(left_col)) | ||
| else: |
There was a problem hiding this comment.
I'd just return here, making the special case self-contained and not needing the else and indentation.
| left_col = [] | ||
| right_col = [] | ||
| for elem in elems: | ||
| if elem.type == "Page-footer" or elem.type == "Page-header": |
There was a problem hiding this comment.
Should probably also exclude titles. Not sure about section headers.
| if x1 < 0.6 and x2 < 0.6: | ||
| left_col.append(elem) | ||
| elif x1 > 0.4 and x2 > 0.4: | ||
| right_col.append(elem) |
There was a problem hiding this comment.
Conceptually, this gets fooled by a 5-column newspaper-ish layout. It puts the first three columns in left. It also calls something two-column if it's one column on the side, but that may end up OK after processing.
The .4 and .6 constants are kind-of magic. I'm sure there will be false positives and false negatives.
| left_col = [] | ||
| right_col = [] |
There was a problem hiding this comment.
Nothing uses right_col. Might as well nuke it, unless you plan to use it. Perhaps if either array is empty, return false and sort by y.
| right_col.append(elem) | ||
| else: | ||
| return False, [], [] | ||
| return True, left_col, right_col |
There was a problem hiding this comment.
Instead of returning a list of "lefts", we could use the existing tagging approach used in this file. Then we might not need a fancy closure-based sort key.
| is_entirely_two_columns, left_col, _ = check_entirely_two_columns(elems) | ||
| if is_entirely_two_columns: | ||
| # if the entire page is two columns, sort the left column first | ||
| elems.sort(key=generate_elem_prefer_left(left_col)) |
There was a problem hiding this comment.
What would happen if we just used elem_left_top here?
|
While this solution improves robustness, we have chosen an even more robust solution here: #1273. |
No description provided.