c++/modules: Prevent emission of really-extern vtables in importers [PR114229]
Checks
Context |
Check |
Description |
linaro-tcwg-bot/tcwg_gcc_build--master-aarch64 |
success
|
Testing passed
|
linaro-tcwg-bot/tcwg_gcc_build--master-arm |
success
|
Testing passed
|
linaro-tcwg-bot/tcwg_gcc_check--master-aarch64 |
success
|
Testing passed
|
linaro-tcwg-bot/tcwg_gcc_check--master-arm |
fail
|
Testing failed
|
Commit Message
Bootstrapped and regtested on x86_64-pc-linux-gnu, OK for trunk?
-- >8 --
Currently, reading a variable definition always marks that decl as
DECL_NOT_REALLY_EXTERN, with anything else imported still being
considered external. This is not sufficient for vtables, however; for an
extern template, a vtable may be generated (and its definition emitted)
but nonetheless the vtable should only be emitted in the TU where that
template is actually instantiated.
While playing around with various settings of DECL_EXTERNAL I also
noticed that we mark variables not declared 'inline' as external, which
makes sense (there's a definition in another TU, while for vague linkage
variables we'll be importing the definition too), but we also do so for
'static constexpr' members that aren't directly marked 'inline'. This
fixes that for consistency, if nothing else (I wasn't able to cause an
actual issue exploiting this currently), but may become relevant with
https://github.com/itanium-cxx-abi/cxx-abi/issues/170.
PR c++/114229
gcc/cp/ChangeLog:
* module.cc (trees_out::core_bools): Count 'static constexpr'
vars also as inline.
(trees_out::write_var_def): Stream DECL_NOT_REALLY_EXTERN for
vtables.
(trees_in::read_var_def): Read it.
gcc/testsuite/ChangeLog:
* g++.dg/modules/virt-2_c.C:
* g++.dg/modules/virt-3_a.C: New test.
* g++.dg/modules/virt-3_b.C: New test.
* g++.dg/modules/virt-3_c.C: New test.
* g++.dg/modules/virt-3_d.C: New test.
Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
---
gcc/cp/module.cc | 15 +++++++++++++--
gcc/testsuite/g++.dg/modules/virt-2_c.C | 14 +++++++++-----
gcc/testsuite/g++.dg/modules/virt-3_a.C | 9 +++++++++
gcc/testsuite/g++.dg/modules/virt-3_b.C | 6 ++++++
gcc/testsuite/g++.dg/modules/virt-3_c.C | 3 +++
gcc/testsuite/g++.dg/modules/virt-3_d.C | 7 +++++++
6 files changed, 47 insertions(+), 7 deletions(-)
create mode 100644 gcc/testsuite/g++.dg/modules/virt-3_a.C
create mode 100644 gcc/testsuite/g++.dg/modules/virt-3_b.C
create mode 100644 gcc/testsuite/g++.dg/modules/virt-3_c.C
create mode 100644 gcc/testsuite/g++.dg/modules/virt-3_d.C
Comments
On 3/5/24 22:06, Nathaniel Shead wrote:
> Bootstrapped and regtested on x86_64-pc-linux-gnu, OK for trunk?
>
> -- >8 --
>
> Currently, reading a variable definition always marks that decl as
> DECL_NOT_REALLY_EXTERN, with anything else imported still being
> considered external. This is not sufficient for vtables, however; for an
> extern template, a vtable may be generated (and its definition emitted)
> but nonetheless the vtable should only be emitted in the TU where that
> template is actually instantiated.
Does the vtable go through import_export_decl? I've been thinking that
that function (and import_export_class) need to be more module-aware.
Would it make sense to do that rather than stream DECL_NOT_REALLY_EXTERN?
Jason
@@ -5418,7 +5418,7 @@ trees_out::core_bools (tree t)
&& !(TREE_STATIC (t)
&& DECL_FUNCTION_SCOPE_P (t)
&& DECL_DECLARED_INLINE_P (DECL_CONTEXT (t)))
- && !DECL_VAR_DECLARED_INLINE_P (t))
+ && !DECL_INLINE_VAR_P (t))
is_external = true;
break;
@@ -11799,6 +11799,12 @@ trees_out::write_var_def (tree decl)
}
tree_node (dyn_init);
}
+
+ /* For vtables we need to know if they're actually extern or not,
+ even if we get a definition; for other kinds of variables we
+ can assume that if we have a definition they can be emitted. */
+ if (streaming_p () && VAR_P (decl) && DECL_VTABLE_OR_VTT_P (decl))
+ u (DECL_NOT_REALLY_EXTERN (decl));
}
void
@@ -11816,6 +11822,11 @@ trees_in::read_var_def (tree decl, tree maybe_template)
tree dyn_init = init ? NULL_TREE : tree_node ();
unused -= vtable;
+ /* Assume for most vars that if we have a definition it's not extern. */
+ bool not_really_extern = true;
+ if (vtable)
+ not_really_extern = u ();
+
if (get_overrun ())
return false;
@@ -11826,7 +11837,7 @@ trees_in::read_var_def (tree decl, tree maybe_template)
if (installing)
{
if (DECL_EXTERNAL (decl))
- DECL_NOT_REALLY_EXTERN (decl) = true;
+ DECL_NOT_REALLY_EXTERN (decl) = not_really_extern;
if (VAR_P (decl))
{
DECL_INITIALIZED_P (decl) = true;
@@ -9,8 +9,12 @@ int Foo ()
return !(Visit (&v) == 0);
}
-// We do emit Visitor vtable
-// andl also we do emit rtti here
-// { dg-final { scan-assembler {_ZTVW3foo7Visitor:} } }
-// { dg-final { scan-assembler {_ZTIW3foo7Visitor:} } }
-// { dg-final { scan-assembler {_ZTSW3foo7Visitor:} } }
+// Again, we do not emit Visitor vtable
+// but we do emit rtti here
+
+// but see https://github.com/itanium-cxx-abi/cxx-abi/issues/170:
+// we should only emit RTTI in virt-2_a.C, alongside the vtable
+
+// { dg-final { scan-assembler-not {_ZTVW3foo7Visitor:} } }
+// { dg-final { scan-assembler-not {_ZTIW3foo7Visitor:} { xfail *-*-* } } }
+// { dg-final { scan-assembler-not {_ZTSW3foo7Visitor:} { xfail *-*-* } } }
new file mode 100644
@@ -0,0 +1,9 @@
+// PR c++/114229
+// { dg-additional-options "-fmodules-ts -Wno-global-module" }
+// { dg-module-cmi modA }
+
+module;
+template<class> struct basic_streambuf { virtual void overflow() { } };
+extern template struct basic_streambuf<long>;
+export module modA;
+export basic_streambuf<long> *p;
new file mode 100644
@@ -0,0 +1,6 @@
+// PR c++/114229
+// { dg-additional-options "-fmodules-ts -fno-module-lazy" }
+// { dg-module-cmi modB }
+
+export module modB;
+import modA;
new file mode 100644
@@ -0,0 +1,3 @@
+// PR c++/114229
+template<class> struct basic_streambuf { virtual void overflow() { } };
+template struct basic_streambuf<long>;
new file mode 100644
@@ -0,0 +1,7 @@
+// PR c++/114229
+// { dg-module-do link }
+// { dg-additional-options "-fmodules-ts -fno-module-lazy" }
+
+import modA;
+import modB;
+int main() { }