The custom stream types were almost always slower than just reading the
entire data into memory, and there's no reason not to do that -
resources are small enough that memory usage and disk IO speed aren't a
concern (at least not for any machine that's modern enough to run
Python 3...).
Perhaps the only performance advantage was when reading a small amount
of data from the start of a compressed resource. In that case the
custom stream could incrementally decompress only the part of the data
that's actually needed, which was a bit faster than decompressing the
entire resource and then throwing away most of the data. But this
situation is rare enough that it's not worth handling in the rsrcfork
library. If this is a real performance issue for someone, they can
manually call the incremental decompression functions from
rsrcfork.compress where needed.
Although the underlying stream is already buffered, the extra
BufferedReader wrapper around the SubStream results in a noticeable
performance improvement.
mypy does not like it when the same variable name is used for different
types, even if it is in unrelated if branches and could never actually
cause any problems.
This way all reads performed on a resource data stream are forwarded
to the underlying resource file stream, with the read offsets and
lengths adjusted appropriately.
The old functions/methods still exist, so that they continue to raise
the same exceptions as before (which are different depending on
context), but they now use the same implementation internally.
As of setuptools 46.4.0, this extracts the attribute value statically
using the ast module, if possible. This allows it to work properly even
if the attribute is stored in a file that cannot be imported at setup
time (e. g. because of dependencies that might not be installed yet).
Reading the version number using attr: rsrcfork.__version__ will no
longer work properly if rsrcfork has non-stdlib dependencies at import
time, because setuptools needs to be able to import rsrcfork and read
the version number before the dependencies are installed.
As a workaround, our setup.py now manually parses the version number
from rsrcfork/__init__.py using the ast module.
The {envpython} substitution is not quoted, so spaces in the path are
treated as argument separators and cause the test runs to fail.
To work around this, we now always use an unqualified python command
instead of the {envpython} substitution. This is safe because the tox
commands are always run in a virtual environment, so the python command
is guaranteed to point to the environment's Python and not the system
default.
Because Python's zip terminates once *any* of the input iterables
terminates, the previous code would not detect if the file was missing
resources or contained extra ones.
