[Bug other/111544] New: [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[Bug other/111544] New: [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

            Bug ID: 111544
           Summary: [14 regression] assignment of read-only location after
                    r14-4111-g6e92a6a2a72d3b
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: other
          Assignee: unassigned at gcc dot gnu.org
          Reporter: seurer at gcc dot gnu.org
  Target Milestone: ---

Starting with r14-4111-g6e92a6a2a72d3b I am seeing a whole bunch of errors when
compiling some test code like this:

/home/seurer/gcc/git/install/gcc-test//bin/g++ -c -o WFXMLScanner.o -DSPEC_CPU
-DNDEBUG  -DAPP_NO_THREADS -DXALAN_INMEM_MSG_LOADER -I. -Ixercesc -Ixercesc/dom
-Ixercesc/dom/impl -Ixercesc/sax -Ixercesc/util/MsgLoaders/InMemory
-Ixercesc/util/Transcoders/Iconv -Ixalanc/include -DPROJ_XMLPARSER
-DPROJ_XMLUTIL -DPROJ_PARSERS -DPROJ_SAX4C -DPROJ_SAX2 -DPROJ_DOM
-DPROJ_VALIDATORS -DXML_USE_NATIVE_TRANSCODER -DXML_USE_INMEM_MESSAGELOADER
-m64 -O3 -mcpu=power8 -fpeel-loops -funroll-loops -ffast-math -mpopcntd
-mrecip=all    -DSPEC_CPU_LP64 -DSPEC_CPU_LINUX      WFXMLScanner.cpp
In file included from ./xercesc/util/NameIdPool.hpp:338,
                 from ./xercesc/internal/XMLScanner.hpp:311,
                 from ./xercesc/internal/WFXMLScanner.hpp:93,
                 from WFXMLScanner.cpp:65:
./xercesc/util/NameIdPool.c: In member function
'xercesc_2_5::NameIdPoolEnumerator<TElem>&
xercesc_2_5::NameIdPoolEnumerator<TElem>::operator=(const
xercesc_2_5::NameIdPoolEnumerator<TElem>&)':
./xercesc/util/NameIdPool.c:416:20: error: assignment of read-only location
'((xercesc_2_5::NameIdPoolEnumerator<TElem>*)this)->xercesc_2_5::NameIdPoolEnumerator<TElem>::fMemoryManager'
  416 |     fMemoryManager = toAssign.fMemoryManager;
      |     ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~

This is old, old code (one of the tests in spec 2006) and appears to run just
fine.  Is there actually something wrong in the code or is the error message in
error?


commit 6e92a6a2a72d3b7a5e1b29042d8a6a43fe1085aa (HEAD)
Author: Patrick Palka <ppalka@redhat.com>
Date:   Mon Sep 18 14:47:52 2023 -0400

    c++: non-dependent assignment checking [PR63198, PR18474]

[Bug other/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #1 from Sam James <sjames at gcc dot gnu.org> ---
Maybe attach a preprocessed version for completeness?

[Bug other/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #2 from seurer at gcc dot gnu.org ---
I can't attach the whole thing but I am working on cutting it down.

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|other                       |c++
           Keywords|rejects-valid               |
   Target Milestone|---                         |14.0

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I am suspecting it is invalid code but maybe we should accept for -fpermissive
if possible ...

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #4 from Sam James <sjames at gcc dot gnu.org> ---
Am trying to build xerces-c-2.5.0 but am struggling for other reasons due to
its age...

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #5 from Sam James <sjames at gcc dot gnu.org> ---
Created attachment 55969
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=55969&action=edit
DGXMLScanner.ii

Reproduced w/ DGXMLScanner.ii

1. wget
https://archive.apache.org/dist/xml/xerces-c/Xerces-C_2_5_0/xerces-c-src_2_5_0.tar.gz
2. tar ... && cd xerces*
3. export XERCESCROOT=$(pwd)
4. cd src/xercesc && ./configure && make -j1


```
# c++ -fPIC -DLINUX -D_REENTRANT -c
-I/root/xerces-c-src_2_5_0/src/xercesc/../../include -g -O2   -o
/root/xerces-c-src_2_5_0/src/xercesc/../../obj/LINUX/DGXMLScanner.o
DGXMLScanner.cpp -save-temps
In file included from
/root/xerces-c-src_2_5_0/src/xercesc/../../include/xercesc/util/NameIdPool.hpp:338,
                 from
/root/xerces-c-src_2_5_0/src/xercesc/../../include/xercesc/internal/XMLScanner.hpp:311,
                 from
/root/xerces-c-src_2_5_0/src/xercesc/../../include/xercesc/internal/DGXMLScanner.hpp:102,
                 from DGXMLScanner.cpp:65:
/root/xerces-c-src_2_5_0/src/xercesc/../../include/xercesc/util/NameIdPool.c:
In member function ‘xercesc_2_5::NameIdPoolEnumerator<TElem>&
xercesc_2_5::NameIdPoolEnumerator<TElem>::operator=(const
xercesc_2_5::NameIdPoolEnumerator<TElem>&)’:
/root/xerces-c-src_2_5_0/src/xercesc/../../include/xercesc/util/NameIdPool.c:416:20:
error: assignment of read-only location
‘((xercesc_2_5::NameIdPoolEnumerator<TElem>*)this)->xercesc_2_5::NameIdPoolEnumerator<TElem>::fMemoryManager’
  416 |     fMemoryManager = toAssign.fMemoryManager;
      |     ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
```

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
    MemoryManager* const fMemoryManager;


So reduced:
```
template <int>
struct a
{
        int * const t;
        void f();
};

template <int n>
void a<n>::f()
{
        t = 0;
}
```

Yes this is invalid code that GCC accepted before.

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Actually this is the reduced testcase:
```
struct bs
{
        int * const t;
};
template <int n>
struct a
{
        int * const t;
        a &f(const a&, int *const tt, const a *c, const bs&);
};

template <int n>
a<n> & a<n>::f(const a &b, int *const tt, const a *c, const bs &d)
{
        t = c->t;
        t = b.t;
        // t = d.t;
        // t = tt;
        return *this;
}
```

clang accepts each of the above statements except for the commented out ones.
Not even in pedantic mode they reject them.

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #8 from Sam James <sjames at gcc dot gnu.org> ---
https://github.com/apache/xerces-c/commit/bc3189892c2e700bd3298b77cd8a523080fa74bb

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #9 from Sam James <sjames at gcc dot gnu.org> ---
(In reply to Sam James from comment #8)
> https://github.com/apache/xerces-c/commit/
> bc3189892c2e700bd3298b77cd8a523080fa74bb

https://issues.apache.org/jira/projects/XERCESC/issues/XERCESC-1259

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #10 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #7)
> Actually this is the reduced testcase:
> ```
> struct bs
> {
>         int * const t;
> };
> template <int n>
> struct a
> {
>         int * const t;
>         a &f(const a&, int *const tt, const a *c, const bs&);
> };
> 
> template <int n>
> a<n> & a<n>::f(const a &b, int *const tt, const a *c, const bs &d)
> {
>         t = c->t;
>         t = b.t;
>         // t = d.t;
>         // t = tt;
>         return *this;
> }
> ```
> 
> clang accepts each of the above statements except for the commented out
> ones. Not even in pedantic mode they reject them.

Clang does reject it if it is instantiated:
```
void g(a<1> &a1, bs &d)
{
        a1.f(a1,nullptr,&a1, d);
}
```
So clang most likely thinks b.t and c->t are still type depedent even though
they don't need to be ...

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #11 from seurer at gcc dot gnu.org ---
From, the original I cut it down to this.  Compiles OK with r14-4110, error
with r14-4111.


bad3.c: In member function 'NameIdPoolEnumerator<TElem>&
NameIdPoolEnumerator<TElem>::operator=(const NameIdPoolEnumerator<TElem>&)':
bad3.c:23:20: error: assignment of read-only location
'((NameIdPoolEnumerator<TElem>*)this)->NameIdPoolEnumerator<TElem>::fMemoryManager'
   23 |     fMemoryManager = toAssign.fMemoryManager;
      |     ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~



class MemoryManager;



template <class TElem> class NameIdPoolEnumerator// : public XMemory
{
public :
    NameIdPoolEnumerator<TElem>& operator=
    (
        const NameIdPoolEnumerator<TElem>& toAssign
    );

private :
    MemoryManager* const fMemoryManager;
};




template <class TElem> NameIdPoolEnumerator<TElem>&
NameIdPoolEnumerator<TElem>::
operator=(const NameIdPoolEnumerator<TElem>& toAssign)
{
    fMemoryManager = toAssign.fMemoryManager;
    return *this;
}

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #12 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #10)
> So clang most likely thinks b.t and c->t are still type depedent even though
> they don't need to be ...

Which is fine. It's QoI whether non-dependent errors in uninstantiated
templates are diagnosed. So Clang is not wrong to accept it. And GCC is not
wrong to reject it.

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

--- Comment #13 from Jonathan Wakely <redi at gcc dot gnu.org> ---
But the code is "ill-formed, no diagnostic required" and GCC is now being more
helpful by diagnosing it.

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

seurer at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |INVALID

--- Comment #14 from seurer at gcc dot gnu.org ---
I modified the code to remove the "const" as that what was done to xerces
itself.

[Bug c++/111544] [14 regression] assignment of read-only location after r14-4111-g6e92a6a2a72d3b

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111544

Sam James <sjames at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |macro at orcam dot me.uk

--- Comment #15 from Sam James <sjames at gcc dot gnu.org> ---
*** Bug 113454 has been marked as a duplicate of this bug. ***