Re: [PATCH] c++/modules: Setup aliases imported from modules [PR106820]
On Fri, Apr 12, 2024 at 03:54:21PM -0400, Jason Merrill wrote:
> On 3/26/24 09:24, Nathaniel Shead wrote:
> >
> > I wonder if more generally we need to be doing more work when importing
> > definitions from header units especially to handle all the work that
> > 'make_rtl_for_nonlocal_decl' and 'rest_of_decl_compilation' would have
> > been performing.
>
> Can we just call those functions?
>
Probably at least 'rest_of_decl_compilation' can (should?) be called
from modules, but I haven't worked out exactly in which circumstances
yet and whether there's any other work it does that doesn't make sense
in a modules context. I don't think 'make_rtl_for_nonlocal_decl' can
easily be called directly though.
At the very least it'll probably need to get called in
`read_definition` which doesn't clash with aliases (which are never
definitions, as far as I can tell), so this will still be necessary.
> > But this patch fixes at least one missing step.
>
> OK.
>
Thanks, pushed as r14-9959-g62a0ef0d02cbb7.
> > PR c++/106820
> >
> > gcc/cp/ChangeLog:
> >
> > * module.cc (trees_in::decl_value): Assemble alias when needed.
> >
> > gcc/testsuite/ChangeLog:
> >
> > * g++.dg/modules/pr106820_a.H: New test.
> > * g++.dg/modules/pr106820_b.C: New test.
> >
> > Signed-off-by: Nathaniel Shead
> > ---
> > gcc/cp/module.cc | 9 +
> > gcc/testsuite/g++.dg/modules/pr106820_a.H | 5 +
> > gcc/testsuite/g++.dg/modules/pr106820_b.C | 8
> > 3 files changed, 22 insertions(+)
> > create mode 100644 gcc/testsuite/g++.dg/modules/pr106820_a.H
> > create mode 100644 gcc/testsuite/g++.dg/modules/pr106820_b.C
> >
> > diff --git a/gcc/cp/module.cc b/gcc/cp/module.cc
> > index 8aab9ea0bae..b4e3b38c6fe 100644
> > --- a/gcc/cp/module.cc
> > +++ b/gcc/cp/module.cc
> > @@ -217,6 +217,7 @@ Classes used:
> > #include "dumpfile.h"
> > #include "bitmap.h"
> > #include "cgraph.h"
> > +#include "varasm.h"
> > #include "tree-iterator.h"
> > #include "cpplib.h"
> > #include "mkdeps.h"
> > @@ -8302,6 +8303,14 @@ trees_in::decl_value ()
> > if (state->is_header ()
> > && decl_tls_wrapper_p (decl))
> > note_vague_linkage_fn (decl);
> > +
> > + /* Setup aliases for the declaration. */
> > + if (tree alias = lookup_attribute ("alias", DECL_ATTRIBUTES (decl)))
> > + {
> > + alias = TREE_VALUE (TREE_VALUE (alias));
> > + alias = get_identifier (TREE_STRING_POINTER (alias));
> > + assemble_alias (decl, alias);
> > + }
> > }
> > else
> > {
> > diff --git a/gcc/testsuite/g++.dg/modules/pr106820_a.H
> > b/gcc/testsuite/g++.dg/modules/pr106820_a.H
> > new file mode 100644
> > index 000..7d32d4e5fc3
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/modules/pr106820_a.H
> > @@ -0,0 +1,5 @@
> > +// PR c++/106820
> > +// { dg-additional-options "-fmodules-ts" }
> > +// { dg-module-cmi {} }
> > +
> > +static int __gthrw___pthread_key_create()
> > __attribute__((__weakref__("foo")));
> > diff --git a/gcc/testsuite/g++.dg/modules/pr106820_b.C
> > b/gcc/testsuite/g++.dg/modules/pr106820_b.C
> > new file mode 100644
> > index 000..247fe26e778
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/modules/pr106820_b.C
> > @@ -0,0 +1,8 @@
> > +// PR c++/106820
> > +// { dg-additional-options "-fmodules-ts" }
> > +
> > +import "pr106820_a.H";
> > +
> > +int main() {
> > + __gthrw___pthread_key_create();
> > +}
>
Re: [PATCH] c++/modules: Setup aliases imported from modules [PR106820]
On 3/26/24 09:24, Nathaniel Shead wrote:
I wonder if more generally we need to be doing more work when importing
definitions from header units especially to handle all the work that
'make_rtl_for_nonlocal_decl' and 'rest_of_decl_compilation' would have
been performing.
Can we just call those functions?
But this patch fixes at least one missing step.
OK.
PR c++/106820
gcc/cp/ChangeLog:
* module.cc (trees_in::decl_value): Assemble alias when needed.
gcc/testsuite/ChangeLog:
* g++.dg/modules/pr106820_a.H: New test.
* g++.dg/modules/pr106820_b.C: New test.
Signed-off-by: Nathaniel Shead
---
gcc/cp/module.cc | 9 +
gcc/testsuite/g++.dg/modules/pr106820_a.H | 5 +
gcc/testsuite/g++.dg/modules/pr106820_b.C | 8
3 files changed, 22 insertions(+)
create mode 100644 gcc/testsuite/g++.dg/modules/pr106820_a.H
create mode 100644 gcc/testsuite/g++.dg/modules/pr106820_b.C
diff --git a/gcc/cp/module.cc b/gcc/cp/module.cc
index 8aab9ea0bae..b4e3b38c6fe 100644
--- a/gcc/cp/module.cc
+++ b/gcc/cp/module.cc
@@ -217,6 +217,7 @@ Classes used:
#include "dumpfile.h"
#include "bitmap.h"
#include "cgraph.h"
+#include "varasm.h"
#include "tree-iterator.h"
#include "cpplib.h"
#include "mkdeps.h"
@@ -8302,6 +8303,14 @@ trees_in::decl_value ()
if (state->is_header ()
&& decl_tls_wrapper_p (decl))
note_vague_linkage_fn (decl);
+
+ /* Setup aliases for the declaration. */
+ if (tree alias = lookup_attribute ("alias", DECL_ATTRIBUTES (decl)))
+ {
+ alias = TREE_VALUE (TREE_VALUE (alias));
+ alias = get_identifier (TREE_STRING_POINTER (alias));
+ assemble_alias (decl, alias);
+ }
}
else
{
diff --git a/gcc/testsuite/g++.dg/modules/pr106820_a.H
b/gcc/testsuite/g++.dg/modules/pr106820_a.H
new file mode 100644
index 000..7d32d4e5fc3
--- /dev/null
+++ b/gcc/testsuite/g++.dg/modules/pr106820_a.H
@@ -0,0 +1,5 @@
+// PR c++/106820
+// { dg-additional-options "-fmodules-ts" }
+// { dg-module-cmi {} }
+
+static int __gthrw___pthread_key_create() __attribute__((__weakref__("foo")));
diff --git a/gcc/testsuite/g++.dg/modules/pr106820_b.C
b/gcc/testsuite/g++.dg/modules/pr106820_b.C
new file mode 100644
index 000..247fe26e778
--- /dev/null
+++ b/gcc/testsuite/g++.dg/modules/pr106820_b.C
@@ -0,0 +1,8 @@
+// PR c++/106820
+// { dg-additional-options "-fmodules-ts" }
+
+import "pr106820_a.H";
+
+int main() {
+ __gthrw___pthread_key_create();
+}
[PATCH] c++/modules: Setup aliases imported from modules [PR106820]
Bootstrapped and regtested on x86_64-pc-linux-gnu, OK for trunk?
-- >8 --
I wonder if more generally we need to be doing more work when importing
definitions from header units especially to handle all the work that
'make_rtl_for_nonlocal_decl' and 'rest_of_decl_compilation' would have
been performing. But this patch fixes at least one missing step.
PR c++/106820
gcc/cp/ChangeLog:
* module.cc (trees_in::decl_value): Assemble alias when needed.
gcc/testsuite/ChangeLog:
* g++.dg/modules/pr106820_a.H: New test.
* g++.dg/modules/pr106820_b.C: New test.
Signed-off-by: Nathaniel Shead
---
gcc/cp/module.cc | 9 +
gcc/testsuite/g++.dg/modules/pr106820_a.H | 5 +
gcc/testsuite/g++.dg/modules/pr106820_b.C | 8
3 files changed, 22 insertions(+)
create mode 100644 gcc/testsuite/g++.dg/modules/pr106820_a.H
create mode 100644 gcc/testsuite/g++.dg/modules/pr106820_b.C
diff --git a/gcc/cp/module.cc b/gcc/cp/module.cc
index 8aab9ea0bae..b4e3b38c6fe 100644
--- a/gcc/cp/module.cc
+++ b/gcc/cp/module.cc
@@ -217,6 +217,7 @@ Classes used:
#include "dumpfile.h"
#include "bitmap.h"
#include "cgraph.h"
+#include "varasm.h"
#include "tree-iterator.h"
#include "cpplib.h"
#include "mkdeps.h"
@@ -8302,6 +8303,14 @@ trees_in::decl_value ()
if (state->is_header ()
&& decl_tls_wrapper_p (decl))
note_vague_linkage_fn (decl);
+
+ /* Setup aliases for the declaration. */
+ if (tree alias = lookup_attribute ("alias", DECL_ATTRIBUTES (decl)))
+ {
+ alias = TREE_VALUE (TREE_VALUE (alias));
+ alias = get_identifier (TREE_STRING_POINTER (alias));
+ assemble_alias (decl, alias);
+ }
}
else
{
diff --git a/gcc/testsuite/g++.dg/modules/pr106820_a.H
b/gcc/testsuite/g++.dg/modules/pr106820_a.H
new file mode 100644
index 000..7d32d4e5fc3
--- /dev/null
+++ b/gcc/testsuite/g++.dg/modules/pr106820_a.H
@@ -0,0 +1,5 @@
+// PR c++/106820
+// { dg-additional-options "-fmodules-ts" }
+// { dg-module-cmi {} }
+
+static int __gthrw___pthread_key_create() __attribute__((__weakref__("foo")));
diff --git a/gcc/testsuite/g++.dg/modules/pr106820_b.C
b/gcc/testsuite/g++.dg/modules/pr106820_b.C
new file mode 100644
index 000..247fe26e778
--- /dev/null
+++ b/gcc/testsuite/g++.dg/modules/pr106820_b.C
@@ -0,0 +1,8 @@
+// PR c++/106820
+// { dg-additional-options "-fmodules-ts" }
+
+import "pr106820_a.H";
+
+int main() {
+ __gthrw___pthread_key_create();
+}
--
2.43.2
