Re: [PATCH] c++/modules: ICE with GMF variable later defined inline [PR126783]

[Patrick Palka]

On Wed, 12 Aug 2026, Patrick Palka wrote:

> Tested on x86_64-pc-linux-gnu, does this look OK for trunk/16?
> 
> -- >8 --
> 
> Here transfer_defining_module (added in r16-5213), assumes that if
> a redeclaration has DECL_LANG_SPECIFIC allocated, then so must the
> original declaration, but this isn't true for q below.  Instead we
> should call retrofit_lang_decl to allocate DECL_LANG_SPECIFIC on
> demand, as done in other parts of duplicate_decls.
> 
>       PR c++/126783
> 
> gcc/cp/ChangeLog:
> 
>       * module.cc (transfer_defining_module): Call retrofit_lang_decl
>       instead of assuming the old declaration already has
>       DECL_LANG_SPECIFIC allocated.
> 
> gcc/testsuite/ChangeLog:
> 
>       * g++.dg/modules/gmf-6.C: New test.
> ---
>  gcc/cp/module.cc                     | 2 +-
>  gcc/testsuite/g++.dg/modules/gmf-6.C | 7 +++++++
>  2 files changed, 8 insertions(+), 1 deletion(-)
>  create mode 100644 gcc/testsuite/g++.dg/modules/gmf-6.C
> 
> diff --git a/gcc/cp/module.cc b/gcc/cp/module.cc
> index d6bd4c78d804..09c6c4467347 100644
> --- a/gcc/cp/module.cc
> +++ b/gcc/cp/module.cc
> @@ -22415,7 +22415,7 @@ transfer_defining_module (tree olddecl, tree newdecl)
>  
>    if (DECL_LANG_SPECIFIC (new_inner))
>      {
> -      gcc_checking_assert (DECL_LANG_SPECIFIC (old_inner));
> +      retrofit_lang_decl (old_inner);
>        if (DECL_MODULE_PURVIEW_P (new_inner))
>       DECL_MODULE_PURVIEW_P (old_inner) = true;
>        if (!DECL_MODULE_IMPORT_P (new_inner))

We can be lazier about allocating DECL_LANG_SPECIFIC here, and only do
it if DECL_MODULE_PURVIEW_P is set:

gcc/cp/module.cc                     | 9 ++++++---
 gcc/testsuite/g++.dg/modules/gmf-6.C | 7 +++++++
 2 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/gcc/cp/module.cc b/gcc/cp/module.cc
index d6bd4c78d804..0a1da706f14a 100644
--- a/gcc/cp/module.cc
+++ b/gcc/cp/module.cc
@@ -22415,10 +22415,13 @@ transfer_defining_module (tree olddecl, tree newdecl)

   if (DECL_LANG_SPECIFIC (new_inner))
     {
-      gcc_checking_assert (DECL_LANG_SPECIFIC (old_inner));
       if (DECL_MODULE_PURVIEW_P (new_inner))
-       DECL_MODULE_PURVIEW_P (old_inner) = true;
-      if (!DECL_MODULE_IMPORT_P (new_inner))
+       {
+         retrofit_lang_decl (old_inner);
+         DECL_MODULE_PURVIEW_P (old_inner) = true;
+       }
+      if (!DECL_MODULE_IMPORT_P (new_inner)
+         && DECL_LANG_SPECIFIC (old_inner))
        DECL_MODULE_IMPORT_P (old_inner) = false;
     }

But I don't think this micro-optimization is worth it, since apparently
it's quite rare that DECL_LANG_SPECIFIC isn't already allocated on the
old decl.

> diff --git a/gcc/testsuite/g++.dg/modules/gmf-6.C 
> b/gcc/testsuite/g++.dg/modules/gmf-6.C
> new file mode 100644
> index 000000000000..201650917a32
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/modules/gmf-6.C
> @@ -0,0 +1,7 @@
> +// PR c++/126783
> +// { dg-additional-options "-fmodules -Wno-global-module" }
> +// { dg-module-cmi M }
> +module;
> +extern int const q;
> +inline constexpr int q = 1;
> +export module M;
> -- 
> 2.55.0.559.g11c6700f10
> 
>