[PATCH][Solaris] Default to -fno-cxa-finalize
Saleem Abdulrasool via cfe-commits
cfe-commits at lists.llvm.org
Mon Sep 7 10:38:08 PDT 2015
On Mon, Sep 7, 2015 at 10:34 AM, Xan López <xan at igalia.com> wrote:
> OK! So here's the patch with a test.
The patch LGTM with a minor request for tweaking the commit message to have
the context about the fact that cxa_finalize.o never shipped, and so this
doesn't really cause any problems for older releases.
> Xan
>
> On Mon, Sep 07, 2015 at 09:14:05AM -0700, Saleem Abdulrasool wrote:
> > On Mon, Sep 7, 2015 at 2:28 AM, Xan López <xan at igalia.com> wrote:
> >
> > > On Sat, Sep 05, 2015 at 12:25:28PM -0700, Saleem Abdulrasool wrote:
> > > > > Ping?
> > > > >
> > > >
> > > > Does this break with older Solaris releases? How far back did this
> > > change
> > > > in Solaris? The change itself should really be accompanied with a
> test.
> > >
> > > Hi,
> > >
> > > turns out I'm not really sure cxa_finalize.o was ever available on
> > > Solaris. The initial code for Solaris was pushed in 2012, but it was
> > > hardcoded to work on a pretty specific Solaris/x86 environment. One
> > > commit from that time is r151648, which claims -fno-cxa-atexit
> > > produces broken code and says they'll ship their own
> > > cxa_finalize.o. So it seems it was not available back then
> > > either. I've also seen commits in OpenSolaris from 2013 creating dummy
> > > cxa_finalize.o files so that clang will pass the bare minimum tests.
> > >
> >
> > That seems less than ideal. If there seems to be no release of Solaris
> > with this mythical support file, it doesn't seem too terrible to just get
> > rid of it.
> >
> >
> > > Now that I'm slowly pushing changes to make clang/Solaris work again
> > > (it was totally broken) I've found this issue again. My patch allows
> > > me to compile clang with clang itself, and other simple C++ programs
> > > compile and work. So if there was a bug here it was either fixed or I
> > > have not found it yet. I'd say that we should probably fix whatever
> > > issue there is with -fno-cxa-atexit when it shows up instead of
> > > working around it, or maybe *actually* bundle an implementation for
> > > Solaris with clang itself? But suggestions are welcome.
> > >
> >
> > Okay, so, AIUI, no release ever contained this file, __cxa_exit isn't
> > provided, and that using -fno-cxa-exit does (and really should) works
> well
> > enough for non-trivial applications.
> >
> >
> > > Also, what kind of test would you have in mind for this?
> >
> >
> > That -fno-cxa-exit is included by the driver when compiling for Solaris.
> >
> >
> > >
> > > Xan
> > >
> > > >
> > > >
> > > > > > From 014ddb164689a3452b76f85079f213d607d07840 Mon Sep 17 00:00:00
> > > 2001
> > > > > > From: =?UTF-8?q?Xan=20L=C3=B3pez?= <xan at igalia.com>
> > > > > > Date: Fri, 14 Aug 2015 11:36:56 +0200
> > > > > > Subject: [PATCH 2/4] [Solaris] Default to -fno-cxa-finalize
> > > > > >
> > > > > > There is no __cxa_finalize symbol available on recent Solaris OS
> > > > > > versions, so we need this flag to make non trivial C++ programs
> run.
> > > > > >
> > > > > > Also stop looking for cxa_finalize.o, since it won't be there.
> > > > > > ---
> > > > > > lib/Driver/Tools.cpp | 9 +++------
> > > > > > 1 file changed, 3 insertions(+), 6 deletions(-)
> > > > > >
> > > > > > diff --git a/lib/Driver/Tools.cpp b/lib/Driver/Tools.cpp
> > > > > > index b204961..cf01d7b 100644
> > > > > > --- a/lib/Driver/Tools.cpp
> > > > > > +++ b/lib/Driver/Tools.cpp
> > > > > > @@ -4506,8 +4506,9 @@ void Clang::ConstructJob(Compilation &C,
> const
> > > > > JobAction &JA,
> > > > > > if (!Args.hasFlag(options::OPT_fuse_cxa_atexit,
> > > > > > options::OPT_fno_use_cxa_atexit,
> > > > > > !IsWindowsCygnus && !IsWindowsGNU &&
> > > > > > - getToolChain().getArch() !=
> > > > > llvm::Triple::hexagon &&
> > > > > > - getToolChain().getArch() !=
> > > > > llvm::Triple::xcore) ||
> > > > > > + getToolChain().getTriple().getOS() !=
> > > > > llvm::Triple::Solaris &&
> > > > > > + getToolChain().getArch() !=
> > > llvm::Triple::hexagon &&
> > > > > > + getToolChain().getArch() !=
> > > llvm::Triple::xcore) ||
> > > > > > KernelOrKext)
> > > > > > CmdArgs.push_back("-fno-use-cxa-atexit");
> > > > > >
> > > > > > @@ -6882,10 +6883,6 @@ void
> solaris::Linker::ConstructJob(Compilation
> > > > > &C, const JobAction &JA,
> > > > > >
> > > Args.MakeArgString(getToolChain().GetFilePath("values-Xa.o")));
> > > > > > CmdArgs.push_back(
> > > > > >
> > > Args.MakeArgString(getToolChain().GetFilePath("crtbegin.o")));
> > > > > > -
> > > > > > - if (getToolChain().getDriver().CCCIsCXX())
> > > > > > - CmdArgs.push_back(
> > > > > > -
> > > > > Args.MakeArgString(getToolChain().GetFilePath("cxa_finalize.o")));
> > > > > > }
> > > > > >
> > > > > > const ToolChain::path_list &Paths =
> getToolChain().getFilePaths();
> > > > > > --
> > > > > > 2.4.3
> > > > > >
> > > > >
> > > > > > _______________________________________________
> > > > > > cfe-commits mailing list
> > > > > > cfe-commits at lists.llvm.org
> > > > > > http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
> > > > >
> > > > > _______________________________________________
> > > > > cfe-commits mailing list
> > > > > cfe-commits at lists.llvm.org
> > > > > http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
> > > > >
> > > >
> > > >
> > > >
> > > > --
> > > > Saleem Abdulrasool
> > > > compnerd (at) compnerd (dot) org
> > >
> >
> >
> >
> > --
> > Saleem Abdulrasool
> > compnerd (at) compnerd (dot) org
>
--
Saleem Abdulrasool
compnerd (at) compnerd (dot) org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150907/76343f1c/attachment.html>
More information about the cfe-commits
mailing list