AU: fix delta generation when running out of scratch.

AU: fix delta generation when running out of scratch.

When running out of scratch, deltas were incorrectly generated. The
issue is that cuts used to be processed in arbitrary order, so
sometimes a cut would be processed (thus having temp blocks
allocated), and later another cut would be impossible to process, and
then the node in question would be converted to a REPLACE
operation. That's fine, but we weren't deleting the graph nodes that
moved data aside. Thus, we would generate a delta w/ extra ops that
were moving data around(!!).

This change causes us to prcocess all cuts for a node together, so
they all succeed or fail.

BUG=8063
TEST=unittests, generated/applied delta locally

Committed: http://chrome-svn/viewvc/chromeos?view=rev&revision=4ba850d

[petkov, 2010-10-23 04:52:52]

LGTM w/ a few comments and suggestions.

http://codereview.chromium.org/3997007/diff/1/2
File delta_diff_generator.cc (right):

http://codereview.chromium.org/3997007/diff/1/2#newcode927
delta_diff_generator.cc:927: bool ConvertCutsToFull(
Might be useful to add a docstring.

http://codereview.chromium.org/3997007/diff/1/2#newcode965
delta_diff_generator.cc:965: bool AssignBlockForAdjoiningCuts(
A docstring might be useful. Also, it might make sense to split this routine in
3 pieces for readability purposes?

http://codereview.chromium.org/3997007/diff/1/2#newcode1520
delta_diff_generator.cc:1520: LOG(WARNING) << "bad blob offset! " <<
op->data_offset() << " != "
why is this not FATAL any more?

[petkov, 2010-10-25 16:53:44]

Btw, still LGTM but we should consider breaking up delta_diff_generator into
smaller classes -- for testability and readability purposes.