Pages and Layout

.text is flattened; .pages is not

result.text (Lesson 2) is every page's text joined into one string, in reading order, with page boundaries thrown away. That's fine when you don't care which page a sentence came from. The moment you do, page numbers for citations, per-page chunking for a vector index, skipping a known-bad page, you need result.pages instead.

ParsedPage, field by field

Each entry in result.pages is a ParsedPage:

text_items is the layer underneath .text: instead of one flat string, you get where each piece of text sits on the page (x, y, width, height, in the same point units as the page itself) and what font/size it was set in. This is the foundation later lessons build on: form fields and annotations (Lesson 6, 9) come with their own rects in this same coordinate space.

The code, piece by piece

for page in result.pages:
    print(page.page_num, page.width, page.height, len(page.text))

Straightforward iteration. Note page.text here is that page's own text, not the whole document's, this document only has one page so they happen to match, but on a multi-page PDF they wouldn't.

first_item = page.text_items[0]
print(first_item.text, first_item.x, first_item.y, first_item.font_name, first_item.font_size)

The document's title, as a single TextItem, positioned near the top of the page (y=56.7, close to 0 which is the page's top edge in LiteParse's top-left-origin coordinate system) in a 10pt monospaced font.

page_one = result.get_page(1)
missing = result.get_page(99)

get_page(n) is a convenience lookup by 1-indexed page number, an alternative to result.pages[n - 1]. It returns None instead of raising when the page doesn't exist, worth checking for if page numbers come from somewhere outside your control (like user input).