For files with lots of holes, quite a lot of time is spent investigating those. It might help to be able to pre-load raw offsets found previously.
(Partially inspired by Andrew Kintz mentioning sending around pickled versions of raw offsets in multiprocessing.)
EDIT: if one allows passing this on in the initializer, one might also consider passing on other parameters that are usually inferred from the file itself, like header0 (see #469)
For files with lots of holes, quite a lot of time is spent investigating those. It might help to be able to pre-load raw offsets found previously.
(Partially inspired by Andrew Kintz mentioning sending around pickled versions of raw offsets in multiprocessing.)
EDIT: if one allows passing this on in the initializer, one might also consider passing on other parameters that are usually inferred from the file itself, like
header0(see #469)