[PATCH ARC 3/3]: Fixed-linker-related-testsuite-for-ARC

[Cupertino Miranda]

Hi Alan,

Thanks for noticing this mistake.
Would it be reasonable to just change it to

+	 && ![istarget arc*-*-elf*]

Or should I change it in config.sub to default any elf* to elf.

The reason for the elf32 is our legacy build and regression testing
scripts.
Nevertheless, we do not consider it a requirement to keep this behavior
specially if it could lead us to more of this problems in the future.

Regarding "./config.sub arc", should we make it default to something. We
noticed that arm also gives "arm-unknown-none" and does not support the
target.

Looking forward to your reply and opinion on how we should address it.

Best regards,
Cupertino

On 05/12/2016 02:28 AM, Alan Modra wrote:
> On Wed, May 11, 2016 at 03:23:34PM +0000, Cupertino Miranda wrote:
>> @@ -1707,7 +1706,7 @@ proc check_gc_sections_available { } {
>>  
>>  proc check_shared_lib_support { } {
>>      if {![istarget aarch64*-*-elf]
>> -	 && ![istarget arc*-*-*]
>> +	 && ![istarget arc*-*-elf32]
>>  	 && ![istarget arm*-*-elf]
>>  	 && ![istarget avr-*-*]
>>  	 && ![istarget cr16-*-*]
> The other testsuite changes are OK, but this one looks wrong.
> --target=arc-elf is a valid configure option, resulting in a target
> triple of arc-unknown-elf.
>
> $ ./config.sub arc-elf
> arc-unknown-elf
>
> This one also builds
> $ ./config.sub arc-linux-uclibc
> arc-unknown-linux-uclibc
>
> But this one doesn't, and probably ought to be fixed in config.sub
> upstream.
> $ ./config.sub arc
> arc-unknown-none
>