GitLab

This adds -Xcheck:jni to the core test command line.  The tests will run
a bit more slowly when native code is involved, but I don't think we're
doing any performance testing in here, and the extra JNI testing is
useful.

The "make clean" note was out of date -- it only cleaned one of our four
targets.

* changes:
  DPQ Comment cleanup (and a few very minor code changes)

For float[] and double[], the new versions should run significantly faster than the old.
Removed some obsolete helper methods.

* changes:
  Update docs for debugger change.

Update notes re: breakpoint implementation (changed Oct 28, 96516932).

Three themes to this change: not making unnecessary native calls (like the
pointless "exists" check in canWrite), being consistent in our
treatment of the empty path, and removing unnecessary cruft from the native
code.

I'm sure there must be a better implementation for handling the empty
path (the few methods for which it *isn't* invalid should be the special
cases, not every single method), but in this patch I'm just interested
in correctness.

With this patch, we pass the jtreg empty path test (which we previously
failed) and we pass my more complete test (added to FileTest.java in this
patch).

Bug: 2281992

Not only were the old implementations of these methods over-complicated, they
were both incorrect. We now pass all our existing tests plus the jtreg tests.

Bug: 2281992

* changes:
  Changing JtregRunner to support running tests off device.

* changes:
  Fix an assertion.

The assertion probably dated from before the binary-search path existed.

Merge commit 'e732bb040be84bd2ac652ca53330b80367a65725'

* commit 'e732bb040be84bd2ac652ca53330b80367a65725':
  Fix browser crashes if server certificates have > 32 subjectAltNames.

Merge commit '1c7705b6066ea57109557fc1cb0609991ac0af24' into eclair-mr2-plus-aosp

* commit '1c7705b6066ea57109557fc1cb0609991ac0af24':
  Fix browser crashes if server certificates have > 32 subjectAltNames.

* changes:
  Fix browser crashes if server certificates have > 32 subjectAltNames.

Android-only changes added an arbitrary limit that upstream doesn't have.
This limit-free implementation is still more efficient than upstream (which
always allocates three Vectors).

Dr No: danfuzz
Bug: 2281869

* changes:
  Fix windows SDK build.

Merge commit '683daacf21830a0b0510320c735485c75749c4f1'

* commit '683daacf21830a0b0510320c735485c75749c4f1':
  DO NOT MERGE: Removing the use of FileCanonPathCache.

Merge commit 'd1aead768d5c6e4aa18c4f43e6be97f8a40fbf84' into eclair-mr2-plus-aosp

* commit 'd1aead768d5c6e4aa18c4f43e6be97f8a40fbf84':
  DO NOT MERGE: Removing the use of FileCanonPathCache.

* changes:
  DO NOT MERGE: Removing the use of FileCanonPathCache.

Aside from being an unjustified optimization, users have reported
problems with this in the wild. This cache has already been removed
in master.

* changes:
  Fix java.io.File's JNI's fixed-length buffers.

I've also removed most of the duplication, simplified a lot of the
implementation, and added loads of TODOs now it's possible to see
what's going on under all the obfuscation. (The native code is
roughly half its previous size, but more functional.)

I want to stop here rather than start fixing any of the TODOs
because this change is already large enough and the history will
be clearer if unrelated changes are kept separate (easy though many
of them are).

Strictly speaking, I haven't removed all the fixed-length buffers:
the File.list implementation still uses fixed-length buffers, but
as the new TODOs point out, I think we want to rewrite that code
to better match its callers, and doing so will make the fixed-length
buffers go away. There's no point polishing code that's already
scheduled for deletion.

Add a couple of basic tests, one that assumes that if Path copes
with long paths in a couple of File's methods, it works in all of
them; another that singles out our readlink(2) wrapper because
that's the only place so far where we cope with arbitrary-length
paths moving in the opposite direction (from kernel to JNI to Java).

* changes:
  Explicitly ignore non-IPv4/IPv6 addresses from IFA_ADDRESS.

* changes:
  Add a comment.

Broken by 96516932f1557d8f48a8b2dbbb885af01a11ef6e.

* changes:
  Replaced primitive sorts with Iaroslavski, Bentley, and Bloch's Dual Pivot Quicksort. The originals were based on Bentley and McIlroy's "Engineering a Sort Function." The original floating point sorts suffered from poor performance due to the use of a naive comparison function.  In round numbers, the new version is 1.5x as fast as the old one on integers and twice as fast on floating point numbers (on the latest Android build running on Sholes).  On some data sets (e.g., nearly sorted data, the new version is substantially faster.

The originals were based on Bentley and McIlroy's "Engineering a Sort Function."
The original floating point sorts suffered from poor performance due to the use
of a naive comparison function.  In round numbers, the new version is 1.5x as fast
as the old one on integers and twice as fast on floating point numbers (on the
latest Android build running on Sholes).  On some data sets (e.g., nearly sorted data,
the new version is substantially faster.

Now, with added performance tweaks from Jesse and Bob!! With these tweaks, the sort
is 70% faster than the original sort on integers and over twice as fast on doubles.
None of the optimizations are Dalvik-specific, and we may be able to make it even
faster by adding Dalvik-specific optimizations.

Also added beefier tests.

I don't think this can happen (some other calls return hardware addresses too,
but I've never seen a hardware address in an IFA_ADDRESS message), but netlink
is badly documented, and it's better to be safe than sorry.

OP_MOVE_EXCEPTION handler was neglecting to reset.
Blocking mode was failing to signal empty queue in some cases
Self-cosim was including operations in traces that can't be done twice
Added OP_MOVE_EXCEPTION to self cosim's no-replay ops (it has side effects)
Restored threshold of 1 to self-cosim (now able to boot device with self-cosim)
When threshold < 6, disable 2nd-level translation filter

The original Codegen.c is broken into three components:

- CodegenCommon.c (arch-independend)
- CodegenFactory.c (Thumb1/2 dependent)
- CodegenDriver.c (Dalvik dependent)

For the Thumb/Thumb2 directories, each contain the followin three files:

- Factory.c (low-level routines for instruction selections)
- Gen.c     (invoke the ISA-specific instruction selection routines)
- Ralloc.c  (arch-dependent register pools)

The FP directory contains FP-specific codegen routines depending on
Thumb/Thumb2/VFP/PortableFP:

- Thumb2VFP.c
- ThumbVFP.c
- ThumbPortableFP.c

Then the hierarchy is formed by stacking these files in the following top-down
order:

1 CodegenCommon.c
2 Thumb[2]/Factory.c
3 CodegenFactory.c
4 Thumb[2]/Gen.c
5 FP stuff
6 Thumb[2]/Ralloc.c
7 CodegenDriver.c

* changes:
  Rewrite NetworkInterface's JNI for IPv6.

* changes:
  Switch java_io_File.c over to C++.

Merge commit '070dbac882856834111bc0e766224ee53c8d5471'

* commit '070dbac882856834111bc0e766224ee53c8d5471':
  Fix bugs created by caching NumberFormat & DecimalFormatSymbols objects

Merge commit '0dfef7bc223f18a1372b4c44c822b9cdc5d5f741' into eclair-mr2-plus-aosp

* commit '0dfef7bc223f18a1372b4c44c822b9cdc5d5f741':
  Fix bugs created by caching NumberFormat & DecimalFormatSymbols objects