Unexpected behavior: get_rect() requires count_rects()

[guyrosin]

Hi,
It seems get_rect() returns the actual rectangle only if count_rects() was previously executed on the same PdfTextPage object.

Here's a code example to reproduce:

import urllib.request

import pypdfium2 as pdfium

pdf_path = "test.pdf"
urllib.request.urlretrieve(
    "https://www.africau.edu/images/default/sample.pdf", pdf_path
)
doc = pdfium.PdfDocument(pdf_path)
page = doc[0]
text_page = page.get_textpage()
print(text_page.get_rect(0))
print("----------")
text_page = page.get_textpage()
print(text_page.count_rects())
print(text_page.get_rect(0))

Output:

(0.0, 0.0, 0.0, 0.0)
----------
10
(57.375, 716.9340209960938, 292.9229736328125, 741.93603515625)

[mara004]

Thanks for reporting this!

PDFium docs say

// Function: FPDFText_CountRects
//          Counts number of rectangular areas occupied by a segment of text,
//          and caches the result for subsequent FPDFText_GetRect() calls.

so presumably this is a known (deliberate) requirement of pdfium that FPDFText_CountRects() needs to be called once before FPDFText_GetRect() (as should virtually always be the case in real-world code, but nvm).
However, I agree that our Python helper class should somehow handle this.

Apart from that, get_rect() also lacks a return code check. A failed get_rect() call should raise an exception rather than returning an all-zero rect (CC #208).

[guyrosin]

Yeah, I agree (and must say it's a weird design choice...)
Thanks for the quick response!

[mara004]

and must say it's a weird design choice...

Hmm, yeah, I guess GetRect() needs to know rect count for an OOB check of the caller's index (as this is C), but then one would think GetRect() could call CountRects() internally if no value was assigned to the cache yet.

[guyrosin]

Exactly!

[mara004]

Do you think a proper exception for get_rect() with hint to check if count_rects() was called would be sufficient?

I could also implement the implicit count_rects() call, but that would involve a state variable and an additional if check for all get_rect() calls on our side, just to handle this edge case. Hardly seems worth it TBH.

[guyrosin]

OK, that makes sense. An exception along with a note in get_rect()'s documentation will probably suffice :)

[guyrosin]

Just saw your commit. Looks great, thanks!

[mara004]

In fact this issue reveals a bigger problem than initially thought.
The test code that should have detected this missed assert statements, so the expectations were never confirmed (my bad, sorry). The problem is that count_rects() can be called with specific spans (as our tests do), but then the cached value is not suitable for an OOB check, so count_rects() needs to be called again with the full range. Our docs now clarify this, but it sounds like a design mistake of pdfium.

[guyrosin]

Oh :/ yeah that makes it sounds like a mistake rather than just weird

[mara004]

anyway, thanks a lot for reporting this