[cfe-commits] [llvm-commits] (autoconf) Build ASan runtime for ARM/Android

[Evgeniy Stepanov]

Thanks, r166559, r166560.

I'd be happy to rewrite this once we have agreement on how to do this
in a more maintainable way. I'm not really happy with the out-of-tree
approach you mentioned, because it puts additional burden on the
people who do the actual build, forcing them to remember long, magical
"make" lines. Making it more complex with make it less well-tested.

Multiple SDKs and build environments are a fact of life, and the best
we can do is embed the knowledge of them in the build system without
an explosion in its complexity (linear growth is fine in my book).

On Sat, Oct 20, 2012 at 12:04 AM, Eric Christopher <echristo at gmail.com> wrote:
> Not particularly happy about it, but it seems like it'll be fine for
> now if it unblocks some progress you'd like to make.
>
> -eric
>
> On Thu, Oct 18, 2012 at 8:16 AM, Evgeniy Stepanov
> <eugeni.stepanov at gmail.com> wrote:
>> ping
>>
>> On Wed, Oct 17, 2012 at 6:34 PM, Evgeniy Stepanov
>> <eugeni.stepanov at gmail.com> wrote:
>>> OK, let's get rid of the configure option for the sake of progress.
>>> Please take another look.
>>> This way android runtime can be built with an extra make invocation in
>>> the same build tree.
>>> make -C tools/clang/runtime/ LLVM_ANDROID_TOOLCHAIN_DIR=/path/to/ndk/toolchain
>>>
>>>
>>> On Thu, Oct 11, 2012 at 1:50 AM, Eric Christopher <echristo at gmail.com> wrote:
>>>>>
>>>>> Otherwise I think it starts to become an unwieldy set of configure flags
>>>>> to set any sdk/sysroot that any particular target triple needs in building
>>>>> any of the runtime libraries (libcxx will necessarily have this problem too).
>>>>>
>>>>> Thoughts?
>>>>>
>>>>
>>>> FWIW Jim, Chandler and I have been discussing this as well. Been going
>>>> over how to deal with in versus out of tree builds and options for a full
>>>> toolchain. Just wanted to let you know it's more than just a couple of
>>>> people chatting :)
>>>>
>>>> -eric