Memory consumption

[lc0]

I run nbdiff for relatively not big file(~1.7mb) and memory goes up to ~3.5GB

What is kinda surprising, given that number of cells, that have differences is not huge.

I wonder if it make sense to do some local diffs for non-exact cells? PS did not read much of code, so might sound not very smart 🙈

[lc0]

Potentially this library looks quite fast to resolve diff

https://github.com/google/diff-match-patch

[vidartf]

That sounds like a lot. Is this a notebook you are able to share? It could be very useful for profiling. Depending on where the issue is, likely improvements are:

• Finish the Myers algorithm for diffing instead of brute forcing it ( #402 ).
• Tweak comparison operators for certain output types.
• Ensure no dangling refs prevent GC of old objects.

[vidartf]

Note: The actual library used for doing text diffs are unlikely to affect this issue, but that should of course be considered as well.

[vidartf]

Closing due to missing repro.

[afeld]

I have a file where use of nbdiff seems to grow without bound. To reproduce:

1. Download https://github.com/afeld/python-public-policy/blob/129d5150e1796ecde2c947b4694c3430f388c8a1/lecture_3.ipynb
2. Run nbdiff lecture_3.ipynb lecture_3.ipynb