Review: restore doInitialization/doFinalization

LLVM tweaks: I had commented out the doInitialization/doFinalization calls to the pass manager because they didn't seem to do anything useful.  But on a hunch after some very rare crashes (maybe 1 in 20, not reliable) decided to restore them.  Now no crashes (for at least a couple hundred runs, so at the very least it's much rarer).

[ckulla, 2010-08-12 01:11:54]

http://codereview.appspot.com/1972042/diff/1/2
File src/liboslexec/llvm_instance.cpp (right):

http://codereview.appspot.com/1972042/diff/1/2#newcode3675
src/liboslexec/llvm_instance.cpp:3675: #if 0
You also commented out these opts - was this intentional?

[larrygritz, 2010-08-12 01:14:48]

http://codereview.appspot.com/1972042/diff/1/2
File src/liboslexec/llvm_instance.cpp (right):

http://codereview.appspot.com/1972042/diff/1/2#newcode3675
src/liboslexec/llvm_instance.cpp:3675: #if 0
On 2010/08/12 01:11:54, ckulla wrote:
> You also commented out these opts - was this intentional?

Yes, in side conversations with others the consensus appeared to be that it was
redundant to have to optimize the function AGAIN (after doing so earlier).

[ckulla, 2010-08-12 01:19:00]

LGTM then :)


So is it the doInitialization change that fixes the crashes? Or commenting out
the extra optimizations?

[lg_imageworks.com, 2010-08-12 03:17:53]

On Aug 11, 2010, at 6:19 PM, <ckulla@gmail.com> <ckulla@gmail.com> wrote:

> 
> So is it the doInitialization change that fixes the crashes? Or
> commenting out the extra optimizations?

Not sure, but I'll try to figure it out after I commit the change that seems to
fix things.

--
Larry Gritz
lg@imageworks.com