On Thu, Apr 01, 2021 at 03:09:44PM -0400, Jason Merrill wrote:
> On 4/1/21 2:15 PM, Marek Polacek wrote:
> > Here we crash in reshape_init because we're accessing ggc_freed
> > & poisoned data: since r277865 in defaulted_late_check we call
> > synthesize_method here:
> > 
> >    if (kind == sfk_comparison)
> >      {
> >        /* If the function was declared constexpr, check that the definition
> >           qualifies.  Otherwise we can define the function lazily.  */
> >        if (DECL_DECLARED_CONSTEXPR_P (fn) && !DECL_INITIAL (fn))
> >          synthesize_method (fn);
> >        return;
> >      }
> > 
> > which in this test triggers when we're processing the string<"a">{} in
> > the static_assert.  First, we create a CONSTRUCTOR for the "{}" in
> > cp_parser_functional_cast, then we call finish_compound_literal which
> > calls complete_type and that results in garbage collection, which then
> > frees the CONSTRUCTOR {} we created when parsing the braced-list in
> > string<"a">{} -- at this point, it's not referenced by anything.
> > (That's not the case for 'type' in finish_compound_literal: the symbol
> > table contains a node for operator==, so ggc_mark_roots goes and marks
> > the fn decl, its type, its arguments etc., as used, so we don't collect
> > it.)
> > 
> > We could just bump function_depth around the new call to
> > synthesize_method.  Alternatively, as in this patch, we can create a new
> > GC root to hold the expression list.  Since I'm adding a new global, I
> > felt compelled to get rid of one, and I actually did find an unused one!
> > So, 86 it.  But I'm also happy to just go with bumping function_depth.
> 
> I think bumping function_depth is the right solution; this patch only deals
> with one source of unreachable expressions that could be collected
> prematurely if complete_type results in GC running.

Fair enough, let's go back to my original patch.

Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk/10?

And now I should get back to
<https://gcc.gnu.org/pipermail/gcc-patches/2019-August/527699.html>
where we actually want a new GC root...

-- >8 --
Here we crash in reshape_init because we're accessing ggc_freed
& poisoned data: since r277865 in defaulted_late_check we call
synthesize_method here:

  if (kind == sfk_comparison)
    {
      /* If the function was declared constexpr, check that the definition
         qualifies.  Otherwise we can define the function lazily.  */
      if (DECL_DECLARED_CONSTEXPR_P (fn) && !DECL_INITIAL (fn))
        synthesize_method (fn);
      return;
    }

which in this test triggers when we're processing the string<"a">{} in
the static_assert.  First, we create a CONSTRUCTOR for the "{}" in
cp_parser_functional_cast, then we call finish_compound_literal which
calls complete_type and that results in garbage collection, which then
frees the CONSTRUCTOR {} we created when parsing the braced-list in
string<"a">{} -- at this point, it's not referenced by anything.
(That's not the case for 'type' in finish_compound_literal: the symbol
table contains a node for operator==, so ggc_mark_roots goes and marks
the fn decl, its type, its arguments etc., as used, so we don't collect
it.)

We could just bump function_depth around the new call to synthesize_method
to prevent GC.

gcc/cp/ChangeLog:

        PR c++/99831
        * method.c (defaulted_late_check): ++ and -- function_depth around
        the call to synthesize_method.
        * pt.c: Remove the saved_trees global.

gcc/testsuite/ChangeLog:

        PR c++/99831
        * g++.dg/other/gc6.C: New test.
---
 gcc/cp/method.c                  |  7 ++++++-
 gcc/cp/pt.c                      |  1 -
 gcc/testsuite/g++.dg/other/gc6.C | 16 ++++++++++++++++
 3 files changed, 22 insertions(+), 2 deletions(-)
 create mode 100644 gcc/testsuite/g++.dg/other/gc6.C

diff --git a/gcc/cp/method.c b/gcc/cp/method.c
index 8ae7496f023..0f416bec35b 100644
--- a/gcc/cp/method.c
+++ b/gcc/cp/method.c
@@ -3131,7 +3131,12 @@ defaulted_late_check (tree fn)
       /* If the function was declared constexpr, check that the definition
         qualifies.  Otherwise we can define the function lazily.  */
       if (DECL_DECLARED_CONSTEXPR_P (fn) && !DECL_INITIAL (fn))
-       synthesize_method (fn);
+       {
+         /* Prevent GC.  */
+         function_depth++;
+         synthesize_method (fn);
+         function_depth--;
+       }
       return;
     }
 
diff --git a/gcc/cp/pt.c b/gcc/cp/pt.c
index dc6f2f37f9b..7956e83c1de 100644
--- a/gcc/cp/pt.c
+++ b/gcc/cp/pt.c
@@ -65,7 +65,6 @@ static GTY(()) struct pending_template *last_pending_template;
 int processing_template_parmlist;
 static int template_header_count;
 
-static GTY(()) tree saved_trees;
 static vec<int> inline_parm_levels;
 
 static GTY(()) struct tinst_level *current_tinst_level;
diff --git a/gcc/testsuite/g++.dg/other/gc6.C b/gcc/testsuite/g++.dg/other/gc6.C
new file mode 100644
index 00000000000..ff45dd313d6
--- /dev/null
+++ b/gcc/testsuite/g++.dg/other/gc6.C
@@ -0,0 +1,16 @@
+// PR c++/99831
+// { dg-do compile { target c++20 } }
+// { dg-options "--param ggc-min-heapsize=0 --param ggc-min-expand=0" }
+
+template <int N> struct S {
+  constexpr S(const char (&str)[N]) : value{} { }
+  char value[N];
+};
+template <S> struct string {
+  constexpr bool operator==(const string &) const = default;
+};
+template <S L2> void operator+(string<L2>) {
+  char value[1];
+  S{value};
+}
+static_assert(string<"a">{} == string<"a">{});

base-commit: 0cf4813202f19768e31666f6aa82bde4dce4065a
-- 
2.30.2

Reply via email to