Re: [boost] Problem cross-compiling boost.context for Raspberry Pi (2)

Subject: Re: [boost] Problem cross-compiling boost.context for Raspberry Pi (2)
From: Vladimir Prus (vladimir.prus_at_[hidden])
Date: 2015-06-16 04:12:09

On 6/16/2015 10:48 AM, Oliver Kowalke wrote:
> 2015-06-16 9:28 GMT+02:00 Vladimir Prus <vladimir.prus_at_[hidden]>:
>
> Depends on what value is correct. If the check can be formulated as
>> compiling some
>> source, or running some tools, then it can be done.
>>
>
> AFAIK it is very hard or nearly impossible for code to detect for which ABI
> and/or binary format it is compiled for.
> ABI/binary format are set/specified by flags in the compiler command line.
> The address model can be detected over pointer size - but ABI ...?!

Right, at least GDB appear to have APCS vs AAPCS as a user option.

>> However, unless I am mistaken, all ARM variants use <abi>aapcs - so maybe
>> it
>> does not need to be detected at all, and can be removed from conditions?
>>
>
> For ARM this is true but not for the other architectures (MIPS for instance
> has 4 different ABIs).

Correct, but we're talking about ARM right now - there, adding <abi> to condition
appears unnecessary and problematic. There is no chance of ambiguity between ARM
and MIPs variants. Do you see any issues if <abi> property is removed from
condition for all ARM variants?

> boost.context determines the default values for ABI/binary-format on the
> values returned by 'os.name' and 'os.platform'.
> I don't know another way for reliable ABI/binary-format detection in
> boost.build.

For avoidance of doubt, do you mean "boost.build" specifically? From your first
observation, it appears that it's hard to detect ABI in any build system not
using mind reading technology?

- Volodya