This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: Compile and build tests with too many sections.
- From: Alan Modra <amodra at gmail dot com>
- To: Shankar Easwaran <shankarke at gmail dot com>
- Cc: binutils at sourceware dot org
- Date: Thu, 28 Jan 2016 23:53:14 +1030
- Subject: Re: Compile and build tests with too many sections.
- Authentication-results: sourceware.org; auth=none
- References: <loom dot 20160122T193809-749 at post dot gmane dot org>
On Fri, Jan 22, 2016 at 06:54:52PM +0000, Shankar Easwaran wrote:
> I have a test which generates too many sections and we expect the linker to
> resolve a COMMON symbol that exists in multiple files.
>
> Test :-
>
> cat > gen.py << \!
> for i in range (0, 90000):
> print "int foo%d() { return 0; }" % (i)
> print "int mycommon[10];"
> !
>
> cat > gen2.py << \!
> for i in range (0, 90000):
> print "int bar%d() { return 0; }" % (i)
> print "int mycommon[20];"
> !
>
> python gen.py > gen.c
> python gen2.py > gen2.c
>
> g++ -c -ffunction-sections gen.c
> g++ -c -ffunction-sections gen2.c
> ld gen.o gen2.o
> gen2.o:(.bss+0x0): multiple definition of `mycommon'
> gen.o:(.bss+0x0): first defined here
>
> Further analyzing the problem, it looks like ths symbol mycommon gets
> allocated in the section .bss as opposed to being a COMMON symbol.
Well, yes, you did compile using g++ after all. C++ is not C, and one
of the differences is how uninitialized global variables behave.
> Digging into the object file, When I see the generated file, the reserved
> section index SHN_COMMON gets used up in the object file for symbols
> outside COMMON symbols.
You might need to look a little deeper into the object than readelf
shows you. You'll find that symbols defined in sections with indices
of 0xff00 and higher use SHN_XINDEX in ".symtab". The actual index
for these symbols is found in ".symtab_shndx".
--
Alan Modra
Australia Development Lab, IBM