invalidate phase info before starting the code selection; temporary disable freeing of old phases in dead code elimination to avoid problems

[r27309]

[r27308]

[r27307]

[r27306]

[r27305]

[r27304]

[r27303]

[r27302]

[r27301]

[r27300]

[r27297]

[r27295]

[r27293]

[r27291]

[r27290]

no need for obstack_printf from libc (we provide our own); no need to explicitely check for C89 function strerror

[r27289]

- deleted header file of already gone irdump_grgen.c

[r27287]

[r27286]

- rework ir_phase API
(sorry for mixing these 2 things into 1 commit)

[r27285]

[r27277]

[r27276]

[r27275]

[r27274]

[r27271]

[r27270]

[r27268]

[r27267]

the walking strategy in the dominance calculation doesn't always find all dead blocks - try to ignore them

[r27266]

- remove a bunch of code which unexpectedly dependent on current_ir_graph being
  set to something
- fix unnecessary double creation of backedge arrays in some cases

[r27265]

[r27264]

[r27259]

[r27257]

Correct a subtle bug in the ia32 backend: Sub(x, x) triggered that the Neg+Add trick was used, which resulted in incorrect code.

[r27256]

[r27255]

[r27254]

- cleanup ir_spec generation a bit

[r27251]

[r27250]

[r27249]

[r27248]

- kicked some superflous calls
- set svn properties

[r27247]