Bump to 0.3.3

This release incorporates the zlo library in favor of bincode, giving
hopper a significant reduction in on-disk storage space, about 50%
for the cernan project.

Signed-off-by: Brian L. Troutwine <blt@postmates.com>

Move senders' file descriptor into FsSync (postmates#17)

This commit moves the Sender file into FsSync, reducing the total
number of file descriptors hopper requires and freeing up disk
space in the event of a slow Sender not replacing its private
file descriptor.

This resolves postmates#16

Signed-off-by: Brian L. Troutwine <blt@postmates.com>

fixes unbounded memory growth issues

tagging 0.3.0 release for parity with crates.io

Bump to 0.2.2

This commit updates the bincode dependency of hopper and does
little more than that. Nothing is changed from the end user
perspective.

Signed-off-by: Brian L. Troutwine <blt@postmates.com>

Bump to 0.2.1

This release update the bincode internal dependency to alpha2 from
alpha1. This has been done to resolve build failures reported
against alpha1.

Signed-off-by: Brian L. Troutwine <blt@postmates.com>

Bump to 0.2.0

This release includes improvement to iteration implementation of
receiver which, at the same time, changes the API of hopper. Our
use of serde has been updated to 0.9 and we no longer have an
extra build step for serde codegen, restricting hopper 0.2.0 to
rust 1.15 and above.

Signed-off-by: Brian L. Troutwine <blt@postmates.com>

Bump to 0.1.2

This release corrects the out-of-order delivery issue that was
introduced in 0.1.1. Hopper now keeps an explicit list of write
indecies for the receiver to read from. This is potentially an
unbounded list--as we cannot toss any writes--but the overhead
is one two-tuple of usize per index bound and there should be
progress made between the receiver and senders to cause that
list to be practically bounded.

Signed-off-by: Brian L. Troutwine <blt@postmates.com>

Bump to 0.1.1

This release introduces an optimization for Sender writes. Previously
we would do a great many small WRITE/FLUSH pairs, one for each new
event. Now we write events into a secondary disk_buffer and flush
that once full. Performance has improved substantially.

Signed-off-by: Brian L. Troutwine <blt@postmates.com>