Previously, for segmented layers, no chunk larger than a full segment
was scanned. Now, we combine contiguous segments up to
scanner.chunk_size (or until a non-present segment) in order to scan it.
This may be slightly slower (by concatting the data together) but gives
results of hits across page boundaries (the point of the scanning
capability), whereas previously each page was scanned individually.
Turns out that #133 exposed a mistake in how we were using the end_bit
field (and it should have been picked up in review, my bad).
Essentially we were masking based on the length of the end_bit after
*already* shifting by the start_bit. We should mask then shift, not the
other way around.
May well fix issue #135 (pdb generation will have been affected by
this).
Previously if submodules were imported after the plugin/symbol __path__s had
been set, it would only use the __path__s from the constants module.
We now use the parent module's __path__s to ensure we get the right
sub__path__s.
This seems better/more flexible than changing the original constants
(since they'd no longer be constant) and allows people more chance to
override things. It brings with it potential confusion as a local
plugin directory will need __init__ files with this magic to allow the
core modules to run (since they come first in the path list).
There were a number of issues with commit 3df5e995 that was applied in
haste (notably, that exceptions wasn't imported in several cases, which
would break the code if it were ever run).
We now give debugging output when a process can't be constructed and
provide as much available information as possible.
Two unused lines were also removed from verinfo.
