Use section contributions for initial parsing of .text

Use section contributions for initial parsing of .text

This CL modifies Decomposer to use section contributions for the first pass of parsing, followed by the use of function and data symbols for refinement. The hope was that this would free us from control-flow analysis and block merging, but unfortunately this is still required in order to properly parse NaCl blocks (whose section contributions are too fine grained).

As a side-effect, section contributions give us reliable information regarding padding blocks in .text. This CL also introduces padding block detection, and modifies the two relinkers to not output padding blocks when reordering a section.
Committed: http://code.google.com/p/sawbuck/source/browse/#svn/trunk384

[chrisha, 2011-07-20 16:19:19]

PTAL

[ericdingle, 2011-07-20 16:51:21]

LGTM

http://codereview.appspot.com/4805044/diff/2001/syzygy/pe/decomposer.cc
File syzygy/pe/decomposer.cc (right):

http://codereview.appspot.com/4805044/diff/2001/syzygy/pe/decomposer.cc#newco...
syzygy/pe/decomposer.cc:552: while (data < data_end) {
Handy for loop helps here

for (; data < data_end; ++data)

or even declare data_end in the initializer list!

http://codereview.appspot.com/4805044/diff/2001/syzygy/relink/order_relinker.cc
File syzygy/relink/order_relinker.cc (right):

http://codereview.appspot.com/4805044/diff/2001/syzygy/relink/order_relinker....
syzygy/relink/order_relinker.cc:152: // If this block is a padding block, and it
has no references or referrers,
Is there no way to extract this out to the base relinker so that these types of
changes don't have to be made in both relinker subclasses?

[chrisha, 2011-07-20 17:53:27]

Thanks, committing.

http://codereview.appspot.com/4805044/diff/2001/syzygy/pe/decomposer.cc
File syzygy/pe/decomposer.cc (right):

http://codereview.appspot.com/4805044/diff/2001/syzygy/pe/decomposer.cc#newco...
syzygy/pe/decomposer.cc:552: while (data < data_end) {
On 2011/07/20 16:51:21, ericdingle wrote:
> Handy for loop helps here
> 
> for (; data < data_end; ++data)
> 
> or even declare data_end in the initializer list!

What is this 'for' you speak of?

Done.

http://codereview.appspot.com/4805044/diff/2001/syzygy/relink/order_relinker.cc
File syzygy/relink/order_relinker.cc (right):

http://codereview.appspot.com/4805044/diff/2001/syzygy/relink/order_relinker....
syzygy/relink/order_relinker.cc:152: // If this block is a padding block, and it
has no references or referrers,
On 2011/07/20 16:51:21, ericdingle wrote:
> Is there no way to extract this out to the base relinker so that these types
of
> changes don't have to be made in both relinker subclasses?

The relinker is due for a bit of a refactor. There's currently a bunch of
duplicated code between the two relinkers. Maybe a helper function like
'OutputBlock'? Then the 'ReorderSection' overrides can call this...

For now, I'll put a TODO in there.

[Roger McFarlane, 2011-07-20 20:00:43]

lgtm