[LLVMdev] unwind's permanent residence

[Jonathan Roelofs]

On 4/22/15 10:04 AM, Reid Kleckner wrote:
> Really dumb question. Why do we need libcxxabi at all?
> Can we move it all into compiler-rt and have less repos?

Oh no, not this monster thread... again. :(

By the same rhetoric, why have separate repositories for Clang and LLVM?

Separation of libraries by repository is *good* for keeping out layering 
violations. It's hard enough to keep ourselves "honest" w.r.t. that 
as-is (hence moving the unwinder out of libcxxabi).

> I don't see what's all that different between the sanitizer run times
> and the bits of c++ rt in libcxxabi.

The fundamental difference is that the sanitizer runtimes don't fully 
implement the Itanium ABI, whereas libcxxabi does... they're only shims 
for a few "interesting" functions in that library. Also, can't the 
sanitizers be used with libcxxrt too?

As a sidenote, IMHO the sanitizers should be in their own repository 
too, not bundled with compiler_rt... but that's a whole other monster of 
a discussion.


Cheers,

Jon

>
> Sent from phone
>
> On Apr 22, 2015 3:17 AM, "C Bergström" <cbergstrom at pathscale.com
> <mailto:cbergstrom at pathscale.com>> wrote:
>
>     On Wed, Apr 22, 2015 at 5:12 PM, Renato Golin
>     <renato.golin at linaro.org <mailto:renato.golin at linaro.org>> wrote:
>      > On 22 April 2015 at 03:40, Saleem Abdulrasool
>     <compnerd at compnerd.org <mailto:compnerd at compnerd.org>> wrote:
>      >> So after a bit of a hiatus (sorry, other stuff has been eating
>     up my free
>      >> time), Id like to pick this up again.  I think that its a matter
>     of just
>      >> copying the unwind sources into the right place.  Im hoping to
>     do this
>      >> sometime this Friday (or perhaps Saturday).  Any objections?  I
>     can probably
>      >> try to take a stab with the CMake side of things once the repo
>     is copied
>      >> over.
>      >
>      > Hi Saleem,
>      >
>      > Thanks for looking at it again, this may simplify the FreeBSD
>     usage of
>      > compiler-rt a lot (they still use libgcc_s/eh with it).
>      >
>      > I imagine that copying the files will be the simple part. More
>     complex
>      > will be to make sure that they're built in the same way (so updating
>      > both CMale files to add/remove logic), and making sure to move all
>      > tests and get them to run when you do a make check on compiler-rt.
>      >
>      > However, the worse part will probably making sure that both
>      > compiler-rt and libc++abi have unwind for a period of time, to allow
>      > everyone to migrate whatever they do, and only kill it after a period
>      > of time. I'd vote for not having a release with the unwinding code on
>      > both places, but having them in trunk for a month or so, with weekly
>      > warnings will probably help a lot people that rely on it.
>
>     No way - that would be super confusing at best. "they" will have to
>     migrate eventually and the sooner that clean cut happens, the better.
>     Having a release with both would be a nightmare.
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>

-- 
Jon Roelofs
jonathan at codesourcery.com
CodeSourcery / Mentor Embedded