integrating the arm bytecode interpreter into IcedTea - gcc version portability issue

[Andrew Haley]

On 07/16/2009 09:57 AM, Xerxes Rånby wrote:
> Matthias Klose skrev:
>> On 15.07.2009 10:43, Gary Benson wrote:
>>   Matthias Klose wrote:
>>>> What does it take to integrate the arm bytecode interpreter written
>>>> by Edward into IcedTea? I currently build it as a set of extra
>>>> patches applied to the Ubuntu packaging, which works ok. Ed provided
>>>> another fix to make it work on armv4 as well, so we don't need any
>>>> explicit configuration option anymore (?).
> Right!
>> Here is the patch that I'm currently testing. Pending the confirmation
>> from Xerxes the code should be compatible with armv4. I didn't add any
>> conditionals to apply it selectively.
>
> Portability issue:
>
> The code contains some lines that makes function calls using c++ mangled
> method names. These lines will cause linking failures when using most
> c++ compilers. The code currently work using gcc 4.3.3 and possibly some
> gcc 4.4 versions.
> Are these some better way to do these calls with portability presered to
> work with any c++ compiler?

Sure: it should easy enough to write a script to mangle the names.  But I
would question whether it's really worth doing.  I think this was just a
gcc bug that caused volatile not to be properly encoded in a mangled name.
It's not as if the mangling algorithm is going to change.

> some parts of the code contains gcc version specific code to deal with
> selected gcc versions.

Maybe we should just declare unmangled equivalents.  Can you provide a list
of the entry points it uses?

Andrew.