Re: Ping: java: Align data in .rodata.jutf8.* sections.

[David Daney <ddaney at caviumnetworks dot com>]

On 04/30/2010 01:50 PM, Tom Tromey wrote:
> > > > > > "David" == David Daney<ddaney@caviumnetworks.com>  writes:
>
> David>  So it has been a little while since I sent this patch:
> David>  http://gcc.gnu.org/ml/java-patches/2009-q1/msg00044.html
> David>  I still think it is a good idea, and would re-test it on the trunk.
>
> I looked at this.  I don't know what it means to set DECL_USER_ALIGN
> when there is no explicit alignment on the decl.  Is this ok?

There is no source decl. We generate these things as a side effect of 
compiling string constants.

My feeling is that gcj is acting in the role of a USER as our proxy when 
it is emitting things that are part of the runtime.  The java language 
doesn't allow the programmer to specify alignments, but we as authors of 
the runtime do know how these data structures should be aligned.  If we 
were emitting C instead of assembly, we would be emitting __attribute__ 
((aligned (??))), so DECL_USER_ALIGN is appropiate.

That's my take on it.

> (This isn't an area of gcc I know well.)
>
> Also it is missing a space before the open paren.

Right, I will fix that one.


> Assuming the generated tree is ok, it seems fine to me.
>
> Tom

Thanks,

I will wait a few more days to see if there are other comments and if 
not re-test and commit it.

David Daney