In tracking a surprise on the modules branch, I came across this code in
duplicate decls. Why would the newly minted decl already have
specializations? That smells wrong.
This chainon call was added by Mark Mitchell back in 1998, (rev 21876)
to deal with template friends of templates. Since that time we've
cleaned a lot of that code up, and in particular do not push the friend
template until instantiation time. I believe these changes make this
chainon unneeded.
This patch changes it to assert there aren't any such specializations to
concatenate. I'm putting this on trunk to give it wider exposure than I
can on the modules branch. It has survived bootstrap on x86_64-linux.
nathan
--
Nathan Sidwell
2019-05-28 Nathan Sidwell <nat...@acm.org>
* decl.c (duplicate_decls): Assert a template newdecl has no
specializations.
Index: cp/decl.c
===================================================================
--- cp/decl.c (revision 271712)
+++ cp/decl.c (working copy)
@@ -2026,7 +2026,8 @@ duplicate_decls (tree newdecl, tree oldd
tree new_result = DECL_TEMPLATE_RESULT (newdecl);
TREE_TYPE (olddecl) = TREE_TYPE (old_result);
- DECL_TEMPLATE_SPECIALIZATIONS (olddecl)
- = chainon (DECL_TEMPLATE_SPECIALIZATIONS (olddecl),
- DECL_TEMPLATE_SPECIALIZATIONS (newdecl));
+
+ /* The new decl should not already have gathered any
+ specializations. */
+ gcc_assert (!DECL_TEMPLATE_SPECIALIZATIONS (newdecl));
DECL_ATTRIBUTES (old_result)
