Re: PING: PATCH: PR target/46770: Use .init_array/.fini_array sections

["H.J. Lu" <hjl.tools@gmail.com>]

On Mon, Aug 22, 2011 at 11:59 AM, Joseph S. Myers
<joseph@codesourcery.com> wrote:
> On Mon, 22 Aug 2011, H.J. Lu wrote:
>
>> > Require a good assembler on ELF targets and just enable this by default
>> > for them without trying a configure test that won't work for cross
>> > compilation (AC_RUN_IFELSE is bad).
>> >
>> > The toplevel config/elf.m4 provides a good notion of what is or is not
>> > ELF (if there are problems, we can fix that file).  Only a handful of
>> > targets support non-GNU assemblers; for the vast bulk of targets we should
>> > assume a not-too-old GNU assembler.  That way, the configure test can be
>> > used to cause a configure-time error if the assembler is defective and it
>> > doesn't matter that the test is late.
>> >
>>
>> A working .init_array support needs assembler, linker and libc.
>> That is why AC_RUN_IFELSE is used.
>
> A working .init_array is a standard part of ELF.  The correct default for
> cross compilation to an ELF target (as determined by elf.m4) is to assume
> it is present; this is not a matter of a GNU extension that could affect
> interoperation with other tools.
>

I wish we can make  .init_array support mandatory.  One problem is
to support mixing input .ctors/.init_array sections to generate
output .init_array section.  Not all linkers support it.  This is a new
feature.  I don't want to cause any more trouble.

Currently, one can force to enable .init_array support by setting
enable_initfini_array to yes.

-- 
H.J.