[Bug lto/78211] [7 Regression] -fcompare-debug failure with -flto -fno-use-linker-plugin

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78211

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |marxin at gcc dot gnu.org

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
This seems to be IPA-ICF bug.
With -fdump-ipa-icf-all
I'm seeing differences like:
   group: with 1 classes:
-    class with id: 1, hash: 3010449829, items: 2
-     
_ZNK8VwViewer13FindViewPlaneERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE(0x7fdb1be7b900/2)
_ZN8VwViewer13FindViewPlaneERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE(0x7fdb1be7b800/1) 
+    class with id: 0, hash: 1515816170, items: 2
+      _ZN11VwViewer_2D15undrawStickyBoxEv(0x7f12887d2c00/4)
_ZN11VwViewer_2D13drawStickyBoxEv(0x7f12887d2b00/3) 
   group: with 1 classes:
-    class with id: 0, hash: 698082993, items: 2
-      _ZN11VwViewer_2D15undrawStickyBoxEv(0x7fdb1be7bb00/4)
_ZN11VwViewer_2D13drawStickyBoxEv(0x7fdb1be7ba00/3) 
+    class with id: 1, hash: 3232079217, items: 2
+     
_ZNK8VwViewer13FindViewPlaneERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE(0x7f12887d2a00/2)
_ZN8VwViewer13FindViewPlaneERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE(0x7f12887d2900/1) 
 Dump after WPA based types groups
 Congruence classes: 2 (unique hash values: 2), with total: 4 items
 Class size histogram [num of members]: number of classe number of classess

Remember the basic rules for -fcompare-debug - DECL_UID can be different, but
their order must be the same (essentially -g can create bigger gaps in between
them), SSA_NAME_VERSION must be identical, and cfun->funcdef_no must be
identical.
The last one is what breaks, IPA-ICF creates funcdef_no that are swapped.
Hashing to different hash values is fine, but care must be taken when actually
traversing the hash tables to generate stuff in the same order between -g and
-g0 if it could affect code generation in any way.