[LLVMdev] MDNodeFwdDecl

[Duncan P. N. Exon Smith]

(Wow, 3.3 is old.)

I think some of the interfaces in DIBuilder now return a temporary
node that didn't before (I wasn't involved in that work, so I don't
know if it's from new API, or if the old API changed).

Otherwise, I can't think of anything off the top of my head.  Let
me know what you discover.

> On 2015-Jun-02, at 14:07, Hui Zhang <wayne.huizhang at gmail.com> wrote:
> 
> Hello, Duncan
> 
> Thanks for the help。
> 
> I've double checked my code and I did call finalize(), and it worked for llvm 3.3, but failed on 3.6. Is there any other possibility that can lead to this error ?
> 
> Thanks
> 
> On Tue, Jun 2, 2015 at 4:29 PM, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:
> 
> > On 2015-Jun-02, at 11:14, Hui Zhang <wayne.huizhang at gmail.com> wrote:
> >
> > Any ideas ？
> >
> > Thanks
> >
> > On Tue, Jun 2, 2015 at 12:40 PM, Hui Zhang <wayne.huizhang at gmail.com> wrote:
> > Hello,
> >
> > I'm having this assertion fail when I was using DIBuilder to create the debug information:
> >
> > Expected no forward declarations!
> > !719
> >
> > The assertion condition is: !isa<MDNodeFwdDecl> (MD)
> >
> > So My questions are:
> > 1. What is exactly the MDNodeFwdDecl ? I saw it's in the latest release 3.6.1, but I also found this link http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20150119/253730.html
> > saying that MDNodeFwdDecl has been removed, and neither can I find class reference of it on the llvm doxgen website. So is it still valid ??
> >
> > 2. How and when will it be generated when I use DIBuilder module ?
> 
> I suspect you've neglected to call `DIBuilder::finalize()`.
> 
> On trunk, `isa<MDNodeFwdDecl>(N)` has become `N.isTemporary()`.  There
> are a number of temporaries created in the debug info graph to facilitate
> building cycles, and `DIBuilder::finalize()` (1) cleans them up, and (2)
> resolves other nodes involved in the cycles.
> 
> It looks like we don't archive old doxygen files, but you should be
> able to build it yourself for old versions.
> 
> 
> 
> -- 
> Best regards
> 
> 
> Hui Zhang