[llvm] r282433 - Allow StringRef to be constructed from a null pointer.

[Zachary Turner via llvm-commits]

Who is the best person to look at this?  It sounds like a codegen bug of
some kind.  I assume it's using an llvm toolchain?

On Wed, Sep 28, 2016 at 2:58 PM Evgenii Stepanov <eugeni.stepanov at gmail.com>
wrote:

> Stack trace:
> #0  0xb74b3850 in _GLOBAL__sub_I_DiagnosticInfo.cpp () at
> /code/llvm/include/llvm/Support/CommandLine.h:291
> #1  0xb747acd7 in __dl__ZN6soinfo16CallConstructorsEv.part.10 () from
> z/system/bin/linker
> #2  0xb747df8f in
> __dl__ZL29__linker_init_post_relocationR19KernelArgumentBlockj () from
> z/system/bin/linker
> #3  0xb74808c8 in __dl___linker_init () from z/system/bin/linker
> #4  0xb7481704 in __dl__start () from z/system/bin/linker
> #5  0x00000001 in ?? ()
>
> Source:
> 291  void setDescription(StringRef S) { HelpStr = S; }
>
> Asm:
> => 0xb74b3850 <+480>: movdqa %xmm2,0x40(%esp)
>
> This is an unaligned load. Your change must have triggered a bug in
> some other code.
>
>
> On Wed, Sep 28, 2016 at 2:49 PM, Zachary Turner <zturner at google.com>
> wrote:
> > Sorry, I take that back.  If this change had any effect on existing code,
> > then llvm-symbolizer should have been asserting before my patch
> >
> > On Wed, Sep 28, 2016 at 2:47 PM Zachary Turner <zturner at google.com>
> wrote:
> >>
> >> That is quite surprising, I admit I haven't seen the stack trace, but
> I'm
> >> inclined to say that something is wrong with the bisection method.  The
> only
> >> thing this change could possibly do is allow an object to be constructed
> >> where previously a construction would have failed to compile.  So if
> this
> >> change had any effect on existing code, then llvm-symbolizer shouldn't
> have
> >> even been compiling before.
> >>
> >> On Wed, Sep 28, 2016 at 2:36 PM Evgenii Stepanov
> >> <eugeni.stepanov at gmail.com> wrote:
> >>>
> >>> This change broke llvm-symbolizer on android/x86 (and only x86), SEGV
> >>> at startup. Trying to get a stack trace...
> >>>
> >>> On Mon, Sep 26, 2016 at 1:08 PM, Zachary Turner via llvm-commits
> >>> <llvm-commits at lists.llvm.org> wrote:
> >>> > Author: zturner
> >>> > Date: Mon Sep 26 15:08:05 2016
> >>> > New Revision: 282433
> >>> >
> >>> > URL: http://llvm.org/viewvc/llvm-project?rev=282433&view=rev
> >>> > Log:
> >>> > Allow StringRef to be constructed from a null pointer.
> >>> >
> >>> > Differential Revision: https://reviews.llvm.org/D24904
> >>> >
> >>> > Modified:
> >>> >     llvm/trunk/include/llvm/ADT/StringRef.h
> >>> >
> >>> > Modified: llvm/trunk/include/llvm/ADT/StringRef.h
> >>> > URL:
> >>> >
> http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ADT/StringRef.h?rev=282433&r1=282432&r2=282433&view=diff
> >>> >
> >>> >
> ==============================================================================
> >>> > --- llvm/trunk/include/llvm/ADT/StringRef.h (original)
> >>> > +++ llvm/trunk/include/llvm/ADT/StringRef.h Mon Sep 26 15:08:05 2016
> >>> > @@ -73,14 +73,14 @@ namespace llvm {
> >>> >      /// Construct an empty string ref.
> >>> >      /*implicit*/ StringRef() : Data(nullptr), Length(0) {}
> >>> >
> >>> > +    /// Disable conversion from nullptr.  This prevents things like
> >>> > +    /// if (S == nullptr)
> >>> >      StringRef(std::nullptr_t) = delete;
> >>> >
> >>> >      /// Construct a string ref from a cstring.
> >>> > +    LLVM_ATTRIBUTE_ALWAYS_INLINE
> >>> >      /*implicit*/ StringRef(const char *Str)
> >>> > -      : Data(Str) {
> >>> > -        assert(Str && "StringRef cannot be built from a NULL
> >>> > argument");
> >>> > -        Length = ::strlen(Str); // invoking strlen(NULL) is
> undefined
> >>> > behavior
> >>> > -      }
> >>> > +        : Data(Str), Length(Str ? ::strlen(Str) : 0) {}
> >>> >
> >>> >      /// Construct a string ref from a pointer and length.
> >>> >      LLVM_ATTRIBUTE_ALWAYS_INLINE
> >>> >
> >>> >
> >>> > _______________________________________________
> >>> > llvm-commits mailing list
> >>> > llvm-commits at lists.llvm.org
> >>> > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160928/9cea1bd6/attachment.html>