[Bug]: deskew results in "empty" output file

[hatl]

Describe the bug

When running ocrmypdf on some files (all cropped in PDF Arranger), the output is broken.
The issue only occurs when adding the deskew flag.

I ran into this issue using the latest version of paperless-ngx.
see also: paperless-ngx/paperless-ngx#8375

Steps to reproduce

1. Run `ocrmypdf --jobs 6 -l deu --output-type pdf --rotate-pages --rotate-pages-threshold 12.0 --skip-text --clean --deskew  scan0003.pdf output.pdf`
2. output is broken

remove deskew and the output is fine

Files

scan0003.pdf
output.pdf

How did you download and install the software?

Ubuntu snap

OCRmyPDF version

16.6.3.dev8+gfe89be5d

Relevant log output

No response

[jbarlow83]

This appears to be related to a general issue with cropped page boxes in ocrmypdf.
The rather unusual mediabox for the input file has these dimensions
0 449.835616 249.069767 842

where typically the third number is 0.

[dennyh]

Sharing my workaround:
Input pdf (problematic.pdf) is rotated with pdftk, cropped with BRISS and joined with pdftk.

$ pdfinfo -box problematic.pdf

Creator:         pdftk-java 3.3.3
Custom Metadata: no
Metadata Stream: no
Tagged:          no
UserProperties:  no
Suspects:        no
Form:            none
JavaScript:      no
Pages:           41
Encrypted:       no
Page size:       624 x 419 pts
Page rot:        90
MediaBox:           19.00   584.00   643.00  1003.00
CropBox:            19.00   584.00   643.00  1003.00
BleedBox:           19.00   584.00   643.00  1003.00
TrimBox:            19.00   584.00   643.00  1003.00
ArtBox:             19.00   584.00   643.00  1003.00
File size:       8419541 bytes
Optimized:       no
PDF version:     1.4

Deskewering it like so:
$ ocrmypdf -d problematic.pdf bad_deskewed.pdf
produces split/half or sometimes empty pages. When the text is partially visible, it looks deskewed.

Solution: Convert problematic.pdf to pdfa:
$ gs -dPDFA=1 -dBATCH -dNOPAUSE -sDEVICE=pdfwrite -sOutputFile=not_problematic.pdfa -dPDFACompatibilityPolicy=1 problematic.pdf

Then feed it to ocrmypdf:
$ ocrmypdf -d not_problematic.pdfa good_deskewed.pdf
This produces a good result.