[Bug c++/53570] New: Unused symbols are not removed for fully static targets (avr-gcc)

[Bug c++/53570] New: Unused symbols are not removed for fully static targets (avr-gcc)

[jwatte at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53570

             Bug #: 53570
           Summary: Unused symbols are not removed for fully static
                    targets (avr-gcc)
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: jwatte@gmail.com


Using built-in specs.
COLLECT_GCC=avr-gcc
COLLECT_LTO_WRAPPER=/usr/lib/gcc/avr/4.7.0/lto-wrapper
Target: avr
Configured with: /build/src/gcc-4.7.0/configure --prefix=/usr --libdir=/usr/lib
--libexecdir=/usr/lib --mandir=/usr/share/man --infodir=/usr/share/info
--enable-languages=c,c++ --disable-libssp --disable-nls --target=avr
--with-as=/usr/bin/avr-as --with-ld=/usr/bin/avr-ld --with-gnu-as --with-gnu-ld
Thread model: single
gcc version 4.7.0 (GCC) 



This is a problem report for how the linker is not smart enough 
to remove unused functions, even with -Os. This makes it very 
hard to create libraries of functions where only some functions 
may be used in any particular binary. Note: "r" option to ar is 
passed in, so the linker should have all the symbol information 
it needs to make the determination.

See the attached test program, especially the "lib.cpp" file.

There is nothing in the C++ standard that requires the compiler 
or linker to include the not_called symbol, or the 
some_large_function symbol, when they are not used. On a small 
target like the AVR, including those symbols is too wasteful.

To reproduce:
- untar this archive
- run ./make.sh
- avr-objdump --disassemble main, and find the unused symbols 
  still included

[Bug c++/53570] Unused symbols are not removed for fully static targets (avr-gcc)

[jwatte at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53570

--- Comment #1 from jwatte at gmail dot com 2012-06-04 07:08:03 UTC ---
Created attachment 27551
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27551
Reproduction case (gzipped tar archive)

[Bug c++/53570] Unused symbols are not removed for fully static targets (avr-gcc)

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53570

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|major                       |normal

[Bug c++/53570] Unused symbols are not removed for fully static targets (avr-gcc)

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53570

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-06-04 08:36:53 UTC ---
This is expected as there is only one section in the object file.
If you use -ffunction-sections -fdata-sections and link with -Wl,--gc-sections.
 The unused functions/data will be removed.

Another way is to use the link time optimizer (-flto on both the compile
command line and the link command line; note you need to have the optimization
options passed to the link command line also).

[Bug c++/53570] Unused symbols are not removed for fully static targets (avr-gcc)

[jwatte at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53570

--- Comment #3 from jwatte at gmail dot com 2012-06-04 18:07:04 UTC ---
I will try -flto and report back. Thanks!

Btw: I think "that's expected" is in the eye of the beholder. If all you've
ever done is VAX/GCC-style UNIX, then it's expected. If you've worked with
other toolchains, like AIX, CodeWarrior, Microsoft, etc, then you'd actually
expect function-level linking by default.

[Bug c++/53570] Unused symbols are not removed for fully static targets (avr-gcc)

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53570

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |ebotcazou at gcc dot
                   |                            |gnu.org
         Resolution|                            |WORKSFORME

--- Comment #4 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-06-09 16:49:59 UTC ---
> Btw: I think "that's expected" is in the eye of the beholder. If all you've
> ever done is VAX/GCC-style UNIX, then it's expected. If you've worked with
> other toolchains, like AIX, CodeWarrior, Microsoft, etc, then you'd actually
> expect function-level linking by default.

Not sure what VAX/GCC-style UNIX means... it's the default ELF behavior.  And
what you call function-level linking is achieved by means of
-ffunction-sections -fdata-sections -Wl,--gc-sections.  -flto is something
else.