gfortran | 9.3.1 missing FINDLOC?

gfortran | 9.3.1 missing FINDLOC?

[Evan Cooch]

Weird one (perhaps). Colleague and I trade back a large slug of FORTRAN 
code -- he is on a Windows machine, running mingw gfortran 9.2. Code 
compiles fine on his end.

But, on my end, no such luck -- I'm on Linux boxes (CentOS 7), running 
gfortran 9.3.1 under scl devtoolset. When I try to compile the exact 
same code, I get a fail at the end -- a few lines referring to a call to 
FINDLOC in one of the modules (something called estmat).

/opt/rh/devtoolset-9/root/usr/libexec/gcc/x86_64-redhat-linux/9/ld: 
estmat.o: in function `filldm.3876': estmat.f90:(.text+0x1201): 
undefined reference to `_gfortran_findloc0_s1'


Was wondering if anyone can give me some advice? Is there a way to 'look 
for' FINDLOC on my box (irony accidental), or (more likely) if there is 
a packaging issue with devtoolkit for CentOS, is there something I can 
do to correct the problem?

Many thanks in advance...

Re: gfortran | 9.3.1 missing FINDLOC?

[Jim Wilson]

On Sun, Mar 28, 2021 at 11:28 AM Evan Cooch via Gcc-help <
gcc-help@gcc.gnu.org> wrote:

> Was wondering if anyone can give me some advice? Is there a way to 'look
> for' FINDLOC on my box (irony accidental), or (more likely) if there is
> a packaging issue with devtoolkit for CentOS, is there something I can
> do to correct the problem?
>

findloc0_s1 should be in the libgfortran.so file.  This is a new function
in GCC 9, so maybe you are linking with the wrong library version?  You can
use nm (or nm -D) to look for symbols in a shared library.  You can use
"gcc --print-file-name=libgfortran.so" to find the library file that gcc is
using by default.  The gcc-9 compiler should point at a gcc-9 version of
this library.  You can add -v to a compiler command to see what the
compiler driver is doing, and -Wl,--verbose to see what the linker is
doing, to see exactly which libgfortran.so file that the linker is using.

Jim

Re: gfortran | 9.3.1 missing FINDLOC?

[Evan Cooch]

On 3/30/2021 11:08 AM, Jim Wilson wrote:
> On Sun, Mar 28, 2021 at 11:28 AM Evan Cooch via Gcc-help 
> <gcc-help@gcc.gnu.org <mailto:gcc-help@gcc.gnu.org>> wrote:
>
>     Was wondering if anyone can give me some advice? Is there a way to
>     'look
>     for' FINDLOC on my box (irony accidental), or (more likely) if
>     there is
>     a packaging issue with devtoolkit for CentOS, is there something I
>     can
>     do to correct the problem?
>
>
> findloc0_s1 should be in the libgfortran.so file.  This is a new 
> function in GCC 9, so maybe you are linking with the wrong library 
> version?  You can use nm (or nm -D) to look for symbols in a shared 
> library.  You can use "gcc --print-file-name=libgfortran.so" to find 
> the library file that gcc is using by default.  The gcc-9 compiler 
> should point at a gcc-9 version of this library.  You can add -v to a 
> compiler command to see what the compiler driver is doing, and 
> -Wl,--verbose to see what the linker is doing, to see exactly which 
> libgfortran.so file that the linker is using.
>
> Jim
>

Many thanks. That gave me a clue. Turns out that if I statically link 
libgfortran (-static-libgfortran), findloc is found (unavoidable irony). 
But, if I use a dynamic link, it is not staying with the scl env, and is 
using the libgfortran that CentOS 7 (and RHEL 7) defaults to, which is 
<< gcc 9. Which, if I'd stoped to think about it, sort of makes sense.

Re: gfortran | 9.3.1 missing FINDLOC?

[Jonathan Wakely]

On Tue, 30 Mar 2021 at 16:14, Evan Cooch wrote:
> Many thanks. That gave me a clue. Turns out that if I statically link
> libgfortran (-static-libgfortran), findloc is found (unavoidable irony).
> But, if I use a dynamic link, it is not staying with the scl env, and is
> using the libgfortran that CentOS 7 (and RHEL 7) defaults to, which is
> << gcc 9. Which, if I'd stoped to think about it, sort of makes sense.

If you're using gfortran from devtoolset then linking to the
libgfortran.so from the base OS is the whole point of devtoolset. New
symbols that aren't in the base libgfortran.so are supposed to be
statically linked into the executable. Please open a bug at
bugzilla.redhat.com about this.

Re: gfortran | 9.3.1 missing FINDLOC?

[Evan Cooch]

On 3/30/2021 11:22 AM, Jonathan Wakely wrote:
> On Tue, 30 Mar 2021 at 16:14, Evan Cooch wrote:
>> Many thanks. That gave me a clue. Turns out that if I statically link
>> libgfortran (-static-libgfortran), findloc is found (unavoidable irony).
>> But, if I use a dynamic link, it is not staying with the scl env, and is
>> using the libgfortran that CentOS 7 (and RHEL 7) defaults to, which is
>> << gcc 9. Which, if I'd stoped to think about it, sort of makes sense.
> If you're using gfortran from devtoolset then linking to the
> libgfortran.so from the base OS is the whole point of devtoolset. New
> symbols that aren't in the base libgfortran.so are supposed to be
> statically linked into the executable. Please open a bug at
> bugzilla.redhat.com about this.

OK, will do once I've worked out the details.

Re: gfortran | 9.3.1 missing FINDLOC?

[Jonathan Wakely]

On Tue, 30 Mar 2021 at 17:42, Evan Cooch <evan.cooch@gmail.com> wrote:
>
>
>
> On 3/30/2021 11:22 AM, Jonathan Wakely wrote:
> > On Tue, 30 Mar 2021 at 16:14, Evan Cooch wrote:
> >> Many thanks. That gave me a clue. Turns out that if I statically link
> >> libgfortran (-static-libgfortran), findloc is found (unavoidable irony).
> >> But, if I use a dynamic link, it is not staying with the scl env, and is
> >> using the libgfortran that CentOS 7 (and RHEL 7) defaults to, which is
> >> << gcc 9. Which, if I'd stoped to think about it, sort of makes sense.
> > If you're using gfortran from devtoolset then linking to the
> > libgfortran.so from the base OS is the whole point of devtoolset. New
> > symbols that aren't in the base libgfortran.so are supposed to be
> > statically linked into the executable. Please open a bug at
> > bugzilla.redhat.com about this.
>
> OK, will do once I've worked out the details.

I'm told that this is already fixed in devtoolset-10.

Re: gfortran | 9.3.1 missing FINDLOC?

[Evan Cooch]

On 3/30/2021 1:55 PM, Jonathan Wakely wrote:
> On Tue, 30 Mar 2021 at 17:42, Evan Cooch <evan.cooch@gmail.com> wrote:
>>
>>
>> On 3/30/2021 11:22 AM, Jonathan Wakely wrote:
>>> On Tue, 30 Mar 2021 at 16:14, Evan Cooch wrote:
>>>> Many thanks. That gave me a clue. Turns out that if I statically link
>>>> libgfortran (-static-libgfortran), findloc is found (unavoidable irony).
>>>> But, if I use a dynamic link, it is not staying with the scl env, and is
>>>> using the libgfortran that CentOS 7 (and RHEL 7) defaults to, which is
>>>> << gcc 9. Which, if I'd stoped to think about it, sort of makes sense.
>>> If you're using gfortran from devtoolset then linking to the
>>> libgfortran.so from the base OS is the whole point of devtoolset. New
>>> symbols that aren't in the base libgfortran.so are supposed to be
>>> statically linked into the executable. Please open a bug at
>>> bugzilla.redhat.com about this.
>> OK, will do once I've worked out the details.
> I'm told that this is already fixed in devtoolset-10.

...which, I've heard, won't be available for CentOS 7. And probably not 
CentOS 8 since its gone to a rolling distro model.

Ah well.

Thanks for the update.