Re: [gccsdk] Absolute binaries giving incorrect __riscosify_control

On 19/04/18 12:51, Theo Markettos wrote:
> On Wed, Apr 18, 2018 at 01:44:45PM +0100, Duncan Moore wrote:
>> This is a bug report, but since the bug reporting list has been out of
>> action for some time, I'm sending it here.
>
> Thanks. We should probably start a new issue tracker on github and migrate
> across. (apologies for closing some of your bugs on the old system BTW - I
> wasn't denying they were valid, just trying to reduce the amount of spamming
> I was getting from bugzilla)
>
>> I have ELF and Absolute binaries getting different values of
>> __riscosify_control from UnixEnv$progname$nonametrans. It's the Absolute
>> binary that is giving the wrong result, and it only seems to happen when the
>> binary is run via a RISC OS variable containing the full pathname of the
>> binary.
>
> Hmm... I'm not familiar with that code, but I'm puzzled. As far as I can
> see elf2aif is simply slapping a pre-baked header on the ELF which does
> zero-initialisation and then jumps to the ELF's entry point. Beyond a
> handful of instructions they're identical.
>
> __riscosify_control is a weak symbol so there *could* be something different
> in linking, but here the linked binary is static and the difference is shown
> with the same linked binary, with and without header.
>
> So my guess is either that the header has some subtle side effect on the OS
> that causes it to do different things, or that SOManager is changing things
> when a statically linked ELF is run directly.
>
> Lee, do you have any thoughts?

Yes, in the last AIF one, the code that constructs the system variable
name in Unixlib takes the program name as <Test$pathname>, so it's
looking for a variable called UnixEnv$<Test$pathname>$nonametrans which
obviously fails.

When SOManager is responsible for starting the program, Test$pathname
is being expanded to the full filename, probably by the DDEUtils module,
so Unixlib finds the leaf name correctly.

So you're write that SOManager is changing things, but I don't think
it could be called a bug.

The best work around is to explicitly name the program in the code:

const char *const __program_name="testbin";

Unixlib will then always use this instead of the CLI.

Lee.

_______________________________________________
GCCSDK mailing list gcc@gccsdk.riscos.info
Bugzilla: http://www.riscos.info/bugzilla/index.cgi
List Info: http://www.riscos.info/mailman/listinfo/gcc
Main Page: http://www.riscos.info/index.php/GCCSDK