Fix lhash implementation to avoid unwanted resizes during iteration.

Fix lhash implementation to avoid unwanted resizes during iteration.

This is a port of the following AOSP patch:
  https://android-review.googlesource.com/#/c/68853/

It fixes a bug that happens when trying to delete items from
a lhash table while it is being iterated over with a call
to lh_doall or lh_doall_arg.

It looks like the source tree is slightly out-of-sync from
the state of running ./import_from_android.sh, but the differences
are minor / not significant.

This patch tries to fix a P1 bug, so doesn't try to address this
(the differences have been removed for easier reviewing).

BUG=298606
R=agl@chromium.org,rsleevi@chromium.org,wtc@chromium.org
TBR=darin@chromium.org

[digit1, 2013-11-04 17:23:06]

I'll soon update this to the latest version of the upstream patch. However, I'd
like to know first if you're ok with getting a "reduced patch" that doesn't
address the fact that the source tree is out-of-sync compared with the result if
running import_from_android.sh. I can either include the corresponding minor
changes, or create another patch below / over this one. Thanks.

[agl, 2013-11-04 17:29:30]

LGTM.

What are the unrelated changes to the source tree? (I thought that I had synced
them up a couple of weeks ago.)

https://chromiumcodereview.appspot.com/57833003/diff/1/openssl/crypto/lhash/l...
File openssl/crypto/lhash/lhash.c (right):

https://chromiumcodereview.appspot.com/57833003/diff/1/openssl/crypto/lhash/l...
openssl/crypto/lhash/lhash.c:160: * the mathematical expression (where '//'
denotes the exact division operation):
(I've reviewed this code in https://android-review.googlesource.com/68853)

[Ryan Sleevi, 2013-11-04 19:46:04]

LGTM. Just TBR= darin.

Like agl said, what's the diff, so that we can keep a closer eye on things?

[digit1, 2013-11-05 10:12:28]

On my machine, if I run import_from_android.sh, I notice the following changes
in the current directory (these should not exist):

  - Various perl scripts have their mode changed from 100755 to 100644, i.e.
they are no longer executables.
  - openssl/crypto/bn/asm/s390x.S also changes from 100755 to 100644 (this file
should not be executable, so this is probably correct in a way).
  - openssl/include/openssl/ssl.h has slight formatting changes (comments, names
of macro parameters).
  - same for openssl/ssl/ssl.h which is a copy of the same file.
  - openssl/patches/channelidchromium.patch is added to the repository. This is
really minor, it should be part of the repository.

I'll upload a patch to fix this soon, and probably update to a more recent
version of the AOSP external/openssl sources, but this is lower priority.

[digit1, 2013-11-05 10:15:02]

darin@chromium.org: Please review changes in 

ben@chromium.org: Please review changes in

[digit1, 2013-11-05 10:16:05]

Looks like I can't push it, "git cl push" complains with:

  fatal: remote error: Git access forbidden

And "git cl dcommit" complains that this isn't a svn checkout. Any clue? Looks
like I have missed something in the svn -> git transition.

[digit1, 2013-11-05 12:46:32]

I've re-uploaded this patch through a svn server at
https://codereview.chromium.org/59793002/, please take a look. Thanks.

[darin (slow to review), 2013-11-05 17:11:12]

LGTM (rubber stamp, deferring to agl+rsleevi+wtc)

[digit1, 2013-11-05 17:21:16]

On 2013/11/05 17:11:12, darin wrote:
> LGTM (rubber stamp, deferring to agl+rsleevi+wtc)

Thanks Darin, can I ask you to put your LGTM on
https://codereview.chromium.org/59793002/ which was required because pushing
through git doesn't work at the moment?

[digit1, 2013-11-05 17:31:05]

Thank you, I'm closing this issue since the other patch has been submitted.