Release notes from redex

v2017.10.31: regalloc: Always rebuild CFG after inserting instructions

2017-10-31T08:48:48Z

Summary:
The lack of rebuilding caused a bug in the code that inserted loads for param
register in their immediate dominators. Roughly, we started with this code:

B0:
load-param v16
load-param v17
load-param v18
load-param v19
if-eqz v16 :foo
B1:
if-eqz v17 :bar

In the first iteration of the regalloc loop, we inserted a load for v18 in B1.
Without rebuilding the CFG, we end up with:

B0:
load-param v16
load-param v17
load-param v18
load-param v19
if-eqz v16 :foo
move v0 v18
B1:
if-eqz v17 :bar

Now in the second iteration of the regalloc loop, we wanted to insert a load
for v19 in B0. We looked to see if the last instruction was a branch opcode --
that would necessitate inserting the move instruction before the branch -- but
since the last opcode in B0 is now a move, we inserted our load in the wrong
place.

I'm not sure why the ART verifier doesn't choke on this, but I'm glad arnaudvenet's
IRTypeChecker did...

Reviewed By: kgsharma

Differential Revision: D6194764

fbshipit-source-id: f7945af4771a82afc6c65f9839b84d34dfc18f15

v1.1.0: Produce a single-file self-extracting script for OSS

2017-01-11T17:58:17Z

Summary:
For the benefit of Buck, homebrew, and all the users who have no idea
what redex-all is, we'll now build redex as a magic self-extracting script. The
script contains a bash extraction preamble followed by a tar file containing
redex-all and the python wrapper (and libraries). When redex is invoked it
untars this payload to a temp location, executes redex with the specified
parameters, and then deletes the untarred payload.

Differential Revision: D4401602

fbshipit-source-id: 94fa04d7a966be55ab86720c4d426ea566b562bb

v1.0.0: Update stringly-typed method refs to account for colon

2016-10-11T22:53:00Z

Summary:
We added a colon to the method descriptor string used by ProguardMap
and the canonical-name form of DexMethod::get_method. This is why using strings
to represent complex data is bad.

This was causing remove-unreachable to go haywire, since it's driven by seeds
information, which is in turn parsed using these assumptions.

Reviewed By: satnam6502

Differential Revision: D4000864

fbshipit-source-id: 561b67794f518d534cce16dab6ee9580bfb44ff6