[Bug bootstrap/107950] New: partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[Bug bootstrap/107950] New: partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

            Bug ID: 107950
           Summary: partial LTO linking of libbackend.a:
                    gcc/gcc-rich-location.cc:207: undefined reference to
                    `range_label_for_type_mismatch::get_text(unsigned int)
                    const'
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
          Assignee: unassigned at gcc dot gnu.org
          Reporter: marxin at gcc dot gnu.org
  Target Milestone: ---

If I use the patch candidate for PR107944 and do:

diff --git a/gcc/Makefile.in b/gcc/Makefile.in
index 615a07089ee..e95390169cf 100644
--- a/gcc/Makefile.in
+++ b/gcc/Makefile.in
@@ -2122,13 +2122,15 @@ selftest: $(SELFTEST_TARGETS)
 # This is used only if the user explicitly asks for it.
 compilations: $(BACKEND)

+$(foreach file,$(OBJS),$(eval CFLAGS-$(file) += -flto))
+
 # This archive is strictly for the host.
 libbackend.a: $(OBJS)
        -rm -rf libbackend.a
        @# Build libbackend.a as a thin archive if possible, as doing so
        @# significantly reduces build times.
 ifeq ($(USE_THIN_ARCHIVES),yes)
-       $(AR) $(AR_FLAGS)T libbackend.a $(OBJS)
+       -$(LINKER) $^ -flinker-output=nolto-rel -r -o $@ -flto=auto
 else
        $(AR) $(AR_FLAGS) libbackend.a $(OBJS)
        -$(RANLIB) $(RANLIB_FLAGS) libbackend.a

I get the following linker error:

$ make lto
g++ -no-pie   -g       -DIN_GCC     -fno-exceptions -fno-rtti
-fasynchronous-unwind-tables -W -Wall -Wno-narrowing -Wwrite-strings
-Wcast-qual -Wmissing-format-attribute -Woverloaded-virtual -pedantic
-Wno-long-long -Wno-variadic-macros -Wno-overlength-strings -fno-common 
-DHAVE_CONFIG_H -static-libstdc++ -static-libgcc   -o lto1 \
        lto/lto-lang.o lto/lto.o lto/lto-object.o attribs.o lto/lto-partition.o
lto/lto-symtab.o lto/lto-common.o libbackend.a main.o libcommon-target.a
libcommon.a ../libcpp/libcpp.a ../libdecnumber/libdecnumber.a  -lmpc -lmpfr
-lgmp -rdynamic  -L./../zlib -lz -lzstd  libcommon.a ../libcpp/libcpp.a  
../libbacktrace/.libs/libbacktrace.a ../libiberty/libiberty.a
../libdecnumber/libdecnumber.a 
/usr/bin/ld: libbackend.a: in function
`range_label_for_type_mismatch::range_label_for_type_mismatch(tree_node*,
tree_node*)':
/home/marxin/Programming/gcc2/gcc/gcc-rich-location.h:158: undefined reference
to `vtable for range_label_for_type_mismatch'
/usr/bin/ld: libbackend.a: in function
`maybe_range_label_for_tree_type_mismatch::get_text(unsigned int) const':
/home/marxin/Programming/gcc2/gcc/gcc-rich-location.cc:207: undefined reference
to `range_label_for_type_mismatch::get_text(unsigned int) const'
collect2: error: ld returned 1 exit status
make: *** [/home/marxin/Programming/gcc2/gcc/lto/Make-lang.in:96: lto1] Error 1

The error comes from gcc/gcc-rich-location.cc where we have:

label_text
maybe_range_label_for_tree_type_mismatch::get_text (unsigned range_idx) const
{
  if (m_expr == NULL_TREE
      || !EXPR_P (m_expr))
    return label_text::borrow (NULL);
  tree expr_type = TREE_TYPE (m_expr);

  tree other_type = NULL_TREE;
  if (m_other_expr && EXPR_P (m_other_expr))
    other_type = TREE_TYPE (m_other_expr);

  range_label_for_type_mismatch inner (expr_type, other_type);
  return inner.get_text (range_idx);
}

and yes, both symbols are undefined:

nm libbackend.a | grep range_label_for_type_mismatch
                 U _ZNK29range_label_for_type_mismatch8get_textEj
                 U _ZTV29range_label_for_type_mismatch

What's strange that also a normal build (non-LTO, no partial linking) has the
same definitions:

nm libbackend.a | grep range_label_for_type_mismatch
                 U _ZNK29range_label_for_type_mismatch8get_textEj
                 U _ZTV29range_label_for_type_mismatch

So any guess what can be bad?

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dmalcolm at gcc dot gnu.org
   Last reconfirmed|                            |2022-12-02
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
Oh, the function has 2 different definitions in C and C++ FE:

gcc/c/c-objc-common.cc:range_label_for_type_mismatch::get_text (unsigned
/*range_idx*/) const
gcc/cp/error.cc:range_label_for_type_mismatch::get_text (unsigned
/*range_idx*/) const

And so that's why I see the linker error for 'lto' binary, interesting..

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
I wonder why the linker complains?  Isn't this like

t.h

class X { foo (); bar (); baz (); };

and splitting the foo/bar/baz implementations to three TUs and then linking
two of them partially?

That is, it looks like we reference the vtable but don't emit it here or
we end up with an ODR violation because somehow LTO thinks it can
privatize some of it?

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #3 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #2)
> I wonder why the linker complains?  Isn't this like
> 
> t.h
> 
> class X { foo (); bar (); baz (); };
> 
> and splitting the foo/bar/baz implementations to three TUs and then linking
> two of them partially?

I've just tried using -flto-partition=one and it hasn't helped me.

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #4 from Jonathan Wakely <redi at gcc dot gnu.org> ---
The vtable will only be emitted in the TU containing the key function, which is
that get_text function defined in the front end.

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
It might be the case the object files were unused in the archive so they are
not linked in the LTO front-end but now with using LTO partial linking, they
are pulled in always.

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #6 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #5)
> It might be the case the object files were unused in the archive so they are
> not linked in the LTO front-end but now with using LTO partial linking, they
> are pulled in always.

Yes, that's likely what's happening. So the question is if the issue is caused
by partial linking, or if the function definition should be provided in
libbackend.a itself?

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #7 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Thanks for looking into the incremental link of libbackend. I had it in my tree
for a while but never got around implementing correct way to enable it only
during bootstrap since host compiler may not support it. It would be nice to
have it in since it should reduce WPA memory use and also test this code path.

I also think it is the case where partial linking makes the symbol to be pulled
into LTO binary at the initial link time.  It should be optimized away if
linker was not complaining.

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #8 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to Jan Hubicka from comment #7)
> Thanks for looking into the incremental link of libbackend. I had it in my
> tree for a while but never got around implementing correct way to enable it
> only during bootstrap since host compiler may not support it. It would be
> nice to have it in since it should reduce WPA memory use and also test this
> code path.

Feel free to grab my initial patch in c#0 and upstream it. I tried that some
time ago in the following email thread:
https://gcc.gnu.org/pipermail/gcc/2021-May/236096.html

> 
> I also think it is the case where partial linking makes the symbol to be
> pulled into LTO binary at the initial link time.  It should be optimized
> away if linker was not complaining.

Optimize away during the partial linking? So you think it's a GCC issue when it
comes to partial linking?

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #9 from Jan Hubicka <hubicka at ucw dot cz> ---
> 
> Feel free to grab my initial patch in c#0 and upstream it. I tried that some
> time ago in the following email thread:
> https://gcc.gnu.org/pipermail/gcc/2021-May/236096.html

Actually I was shooting for partial linking LTO streams just to make
individual WPAs chaper.  With -flinker-output=nolto-rel we are
effectively disabling most of benefits of LTO.
My main problem was arranging the partial link only when stage1/2
compiler is used so bootstrap works with older GCCs which suports LTO
but not partial linking yet.
> 
> > 
> > I also think it is the case where partial linking makes the symbol to be
> > pulled into LTO binary at the initial link time.  It should be optimized
> > away if linker was not complaining.
> 
> Optimize away during the partial linking? So you think it's a GCC issue when it
> comes to partial linking?
No with partial linking you merge all the individual object files into
single. So if lto frontned is not using some of libbackend entry point,
you still get that code biult into it. So I think it is correct
behaviour.

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
I suppose a fix would be to provide a dummy implementation for
range_label_for_type_mismatch::get_text in lto/, but I wonder how
for example the fortran frontend avoids this issue?

f951 has rich_location uses but no range_label_for_type_mismatch, it looks
like range_label_for_* is used only from the C family frontends, so maybe
another fix would be to move that class and implementation somewhere to
c-family/?

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #11 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #10)
> I suppose a fix would be to provide a dummy implementation for
> range_label_for_type_mismatch::get_text in lto/, but I wonder how
> for example the fortran frontend avoids this issue?
> 
> f951 has rich_location uses but no range_label_for_type_mismatch, it looks
> like range_label_for_* is used only from the C family frontends, so maybe
> another fix would be to move that class and implementation somewhere to
> c-family/?

AIUI class range_label_for_type_mismatch is only used:
  * by the C and C++ frontends (directly, and by
range_label_for_format_type_mismatch), and
  * by maybe_range_label_for_tree_type_mismatch::get_text in
gcc/gcc-rich-location.cc, which is used by class binary_op_rich_location in
gcc/gcc-rich-location.{h,cc}; the latter is only used by the C/C++ frontends.

So moving range_label_for_type_mismatch *and* binary_op_rich_location to
c/cp/c-family might be a fix for this; I'm not sure.

[Bug bootstrap/107950] partial LTO linking of libbackend.a: gcc/gcc-rich-location.cc:207: undefined reference to `range_label_for_type_mismatch::get_text(unsigned int) const'

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107950

--- Comment #12 from Martin Liška <marxin at gcc dot gnu.org> ---
> So moving range_label_for_type_mismatch *and* binary_op_rich_location to
> c/cp/c-family might be a fix for this; I'm not sure.

I don't think so as any other (non-C family) FE will fail to link against
libbackend.a.