[llvm-bugs] [Bug 25907] Compiler seg-faults in lto for indirect goto

via llvm-bugs llvm-bugs at lists.llvm.org
Mon Dec 21 07:38:35 PST 2015


Teresa Johnson <tejohnson at google.com> changed:

           What    |Removed                     |Added
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #1 from Teresa Johnson <tejohnson at google.com> ---
The assert happened because some functions were materialized when creating the
lazy module reader due to blockaddress forward references (via
materializeForwardReferencedFunctions). The forward referenced function
contained a reference to module level TBAA metadata, resulting in creation of a
forward MD reference. However, since this was during gold's claim_file_hook
handler, the bitcode reader is destroyed without doing any more materialization
(including metadata materialization), since it is just looking to see if the
file is valid bitcode or not. Therefore, the forward MD reference remained and
the assert fired.

I have decided to remove the assert altogether. It was useful during
development of the ThinLTO metadata linking, but doesn't seem very robust or
necessary. There isn't any requirement that the full bitcode be materialized
before tossing out the bitcode reader as in situations like this one.

Fixed in r256161.

You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20151221/08ddd05b/attachment-0001.html>

More information about the llvm-bugs mailing list