Hey Aldy & Andrew,
I introduced a leak by calling enable_ranger() without pairing it
with one to disable_ranger() on the same function (PR 101984).
I didn't realize (or look to see) that enable_ranger() dynamically
allocates memory.
The patch below adds comments to make it clear that the calls need
to be paired. That seems obvious now but wasn't before from just
the function names. So I'm wondering if we might want to rename
them to make it more obvious that the former involves allocating
memory that must be explicitly deallocated.
If you agree, names along the following lines would make this
clearer (to me, anyway) but I'm open to others:
gimple_ranger *set_new_ranger (function *);
void release_ranger (function *);
Martin
diff --git a/gcc/gimple-range.cc b/gcc/gimple-range.cc
index 60b7d3a59cd..ef3afeacc90 100644
--- a/gcc/gimple-range.cc
+++ b/gcc/gimple-range.cc
@@ -381,6 +381,10 @@ gimple_ranger::dump (FILE *f)
m_cache.dump (f);
}
+/* Create a new ranger instance and associate it with function FUN.
+ Each call must be paired with a call to disable_ranger to release
+ resources. */
+
gimple_ranger *
enable_ranger (struct function *fun)
{
@@ -392,6 +396,9 @@ enable_ranger (struct function *fun)
return r;
}
+/* Destroy and release the ranger instance associated with function FUN
+ and replace it with the global ranger. */
+
void
disable_ranger (struct function *fun)
{
diff --git a/gcc/gimple-range.h b/gcc/gimple-range.h
index 41845b14fd6..eaebb9c5833 100644
--- a/gcc/gimple-range.h
+++ b/gcc/gimple-range.h
@@ -62,6 +62,9 @@ protected:
range_tracer tracer;
};
+/* Create a new ranger instance and associate it with a function.
+ Each call must be paired with a call to disable_ranger to release
+ resources. */
extern gimple_ranger *enable_ranger (struct function *);
extern void disable_ranger (struct function *);
