https://bugs.kde.org/show_bug.cgi?id=380456
Stefan Brüns <stefan.bru...@rwth-aachen.de> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |stefan.bruens@rwth-aachen.d
| |e
--- Comment #22 from Stefan Brüns <stefan.bru...@rwth-aachen.de> ---
(In reply to tagwerk19 from comment #21)
> Created attachment 143869 [details]
> pdftotext results from
> https://ipfs.io/ipfs/QmVqWhPuQkE7reTN5F9TiSeA75z62VNaZUSFZz3FdWTLbC
>
> (In reply to Adam Fontenot from comment #20)
> > ... The file, in their view, is pathological ...
> Applying a modicum of patience, running:
>
> nice -19 pdftotext QmVqWhPuQkE7reTN5F9TiSeA75z62VNaZUSFZz3FdWTLbC.pdf
>
> took 37 hours on a machine with 16GB memory 8-]
>
> The process gradually ate memory, reaching 10 GB. There wasn't an obvious
> impact on performance - but I would expect you'd see that bite when reaching
> the limits/starting to swap.
The long runtime is caused by some algorithmically bad implementation, i.e.
O(n^2) were e.g. O(n log n) is sufficient. The huge memory footprint is caused
by some problematic data arrangement and too greedy pre/overallocation.
I have filed two MRs [1],[2] for poppler, with both applied the extractions
runs in ~50 seconds on my 3 year old laptop, with a peak memory consumption of
1.8 GByte.
[1] https://gitlab.freedesktop.org/poppler/poppler/-/merge_requests/1514
[2] https://gitlab.freedesktop.org/poppler/poppler/-/merge_requests/1515
--
You are receiving this mail because:
You are watching all bug changes.
