[llvm-dev] ORC error: No callback manager available

[Lang Hames via llvm-dev]

Hi Greg,

Unfortunately there is no API to avoid the error message. The only way to
avoid it would be to patch LLVM 8. The patch itself is simple: On
OrcCBindingsStack.h line 455 (See
https://github.com/llvm/llvm-project/blob/89de0d8dfbb9a6ff1f8b141ed70b563ecc094878/llvm/lib/ExecutionEngine/Orc/OrcCBindingsStack.h#L455)
you just need to replace

logAllUnhandledErrors(CCMgr.takeError(), errs(), "ORC error: ");

with

consumeError(CCMgr.takeError());

It seems to me that the newer ORC APIs only exist in LLVM10. Were the older
> ORC APIs deprecated in LLVM10 only?


Yes. The old APIs were deprecated in 10 and will remain deprecated (but
present) in LLVM 11. I aim to remove them before LLVM 12.

Regards,
Lang.

On Mon, Jul 20, 2020 at 2:19 AM Greg Nancarrow <gregn4422 at gmail.com> wrote:

> Hi Lang,
>
> Thanks for your reply.
> I work on a product that is largely based on Postgres but we do
> contribute the odd fix/improvement/feature to the open-source Postgres
> where we can.
>
> It seems to me that the newer ORC APIs only exist in LLVM10.
> Were the older ORC APIs deprecated in LLVM10 only?
> Unfortunately our product has a dependency on LLVM8 (e.g. on RHEL8 on
> the systemz platform, the default LLVM version installed by the system
> package manager "yum" is version 8).
> For LLVM8, is there any other API that can be used to avoid the LLVM
> "No callback manager available" error seen on systemz?
>
> Regards,
> Greg
>
> On Thu, Jul 9, 2020 at 2:29 PM Lang Hames <lhames at gmail.com> wrote:
> >
> > Hi Greg,
> >
> > Thanks for your patience.
> >
> >> Looking through the LLVM source code, it appears that for the
> >> "systemz" architecture, there is  currently no local CCMgr
> >> implementation available, and this then prevents indirection support
> >> and COD layer support. BUT - apparently it can still function (maybe
> >> less efficiently?) without these?
> >> Is somebody able to clarify exactly what that error message means and
> >> what the limitations are? Is it unwise to use LLVM on architectures on
> >> which this error is reported?
> >> I looked through the LLVM V10 source code and there still seems to be
> >> this issue for systemz.
> >
> >
> > You are correct: The error is generated when a call is made to the
> createLocalCompileCallbackManager function. Platforms that don't support a
> callback manager (yet) can not use lazy compilation, but can still JIT
> whole modules.
> >
> > I suspect that postgres is using the older ORC C bindings. These always
> try to enable lazy compilation support and direct any errors to stderr.
> Postgress probably isn't using the lazy compilation support anyway, which
> is why everything works just fine despite the error.
> >
> > These older APIs are deprecated, so the best solution would be for
> postgres to move to the newer C APIs (see
> https://github.com/llvm/llvm-project/blob/master/llvm/include/llvm-c/Orc.h
> ).
> >
> > Are you working on Postgres yourself, or just building it?
> >
> > Regards,
> > Lang.
> >
> > On Thu, Jun 25, 2020 at 4:49 AM Greg Nancarrow via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
> >>
> >> Hi LLVM Developers,
> >>
> >> I have a query about an LLVM error message that I'm encountering and I
> >> hope it's appropriate to post to this mailing list about it.
> >>
> >> Postgres can be built with LLVM-based JIT support.
> >> I find that when running on Linux (RHEL8) on IBM Z hardware, using
> >> LLVM8+, I'm seeing many instances of the following LLVM error message
> >> appearing in the Postgres server log:
> >>
> >>     ORC error: No callback manager available for s390x-ibm-linux
> >>
> >> (strangely, it otherwise doesn't seem to be causing any problems)
> >>
> >> Looking through the LLVM source code, it appears that for the
> >> "systemz" architecture, there is  currently no local CCMgr
> >> implementation available, and this then prevents indirection support
> >> and COD layer support. BUT - apparently it can still function (maybe
> >> less efficiently?) without these?
> >> Is somebody able to clarify exactly what that error message means and
> >> what the limitations are? Is it unwise to use LLVM on architectures on
> >> which this error is reported?
> >> I looked through the LLVM V10 source code and there still seems to be
> >> this issue for systemz.
> >>
> >> Secondly, LLVM sends this error message to standard error. As a caller
> >> of the LLVM library, it would be nice to be able to redirect such
> >> messages from the library - is it possible?
> >>
> >> Hope you can advise,
> >> Greg
> >> _______________________________________________
> >> LLVM Developers mailing list
> >> llvm-dev at lists.llvm.org
> >> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200720/34d9d75a/attachment.html>