https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124225
--- Comment #4 from Liam Powell <liam at liampwll dot com> ---
To save a couple of minutes for anyone who does attempt to fix this one day:
1. Write_Lock obviously needs to go below the null check.
2. An atomic won't work on all platforms due to protected pointers being 2
pointers internally.
3. If the ability to make protected pointers atomic on all platforms is ever
restored then obviously that solves the memory corruption issues and there's a
way to make this much more robust without locks:
@@ -1055,8 +1055,8 @@ package body System.Tasking.Stages is
-- If there is a fall back handler, store its pointer for later
-- execution.
- if ID.Common.Fall_Back_Handler /= null then
- TH := ID.Common.Fall_Back_Handler;
+ TH := ID.Common.Fall_Back_Handler;
+ if TH /= null then
There's still some weird edge cases there if the user is moving the fallback
handler between levels but it is strictly more robust then the old version.
