https://bugs.gentoo.org/935754
https://lists.gnu.org/archive/html/bison-patches/2024-07/msg00000.html

* data/skeletons/glr2.cc (yyundeleteLastStack): Recover
yylookaheadNeeds stack entry also, to prevent desyncing the sizes
of yystates and yylookaheadNeeds.
(class glr_state_set): New field yylookaheadNeedLastDeleted,
tracks lookahead need of last deleted state.
(yymarkStackDeleted): Save yylookaheadNeeds also.
---
Hi!

GCC 15 has added bounds-checks to vector<bool> when assertions are
enabled.  This has caught a bug in Bison.  See bug referenced above.

Now, WRT the fix: I am not sure of its correctness, but it appears to
pass tests and everything seems to indicate that the presumption it is
based on is correct.  That presumption is that the two stacks mentioned
above (yylookaheadNeeds and yystates) ought to be the same size.  I
inferred this because all other locations that alter the size of the two
vectors appear to do so in lock-step.

TIA, have a lovely day.

 data/skeletons/glr2.cc | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/data/skeletons/glr2.cc b/data/skeletons/glr2.cc
index 970ccfdf..8b962d6f 100644
--- a/data/skeletons/glr2.cc
+++ b/data/skeletons/glr2.cc
@@ -1080,7 +1080,10 @@ namespace
     {
       size_t k = yyk.uget ();
       if (yystates[k] != YY_NULLPTR)
-        yylastDeleted = yystates[k];
+        {
+          yylastDeleted = yystates[k];
+          yylookaheadNeedLastDeleted = yylookaheadNeeds[k];
+        }
       yystates[k] = YY_NULLPTR;
     }
 
@@ -1093,6 +1096,7 @@ namespace
       if (yylastDeleted == YY_NULLPTR || !yystates.empty ())
         return;
       yystates.push_back (yylastDeleted);
+      yylookaheadNeeds.push_back (yylookaheadNeedLastDeleted);
       YYCDEBUG << "Restoring last deleted stack as stack #0.\n";
       clearLastDeleted ();
     }
@@ -1102,6 +1106,7 @@ namespace
     void
     yyremoveDeletes ()
     {
+      YYASSERT(yystates.size () == yylookaheadNeeds.size ());
       size_t newsize = yystates.size ();
       /* j is the number of live stacks we have seen.  */
       for (size_t i = 0, j = 0; j < newsize; ++i)
@@ -1160,6 +1165,7 @@ namespace
 
     /** The last stack we invalidated.  */
     glr_state* yylastDeleted;
+    bool yylookaheadNeedLastDeleted;
   }; // class glr_state_set
 } // namespace
 
-- 
2.45.2