[LLVMbugs] [Bug 2038] New: Ruby-1. 9 builds and passes light checks but explodes in the regression suite.

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=2038

           Summary: Ruby-1.9 builds and passes light checks but explodes in
                    the regression suite.
           Product: libraries
           Version: 2.2
          Platform: PC
               URL: http://snakeratpig.blogspot.com/2008/02/alternative-
                    compiler-suites.html
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: Backend: X86
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: ttmrichter at gmail.com
                CC: llvmbugs at cs.uiuc.edu


Executive summary:
Ruby 1.9 compiles under llvm-gcc-4.2/llvm-2.2 just fine and passes sanity
checks.  Running the full regression suite fails with an "Illegal instruction
(core dumped)" message.  Running the same full regression suite with plain old
GCC causes a segmentation fault far later in the process.

Details:
Currently the test case, such as it is, is as follows:
<ol>
<li>svn co http://svn.ruby-lang.org/repos/ruby/trunk ruby-1.9</li>
<li>cd ruby-1.9</li>
<li>autoconf</li>
<li>CC=llvm-gcc ./configure</li>
<li>make</li>
<li>make test</li>
<li>make test-all</li>
</ol>

In the "make test" there are 7 errors out of 800-something tests.  This is
identical to what happens when GCC is used.  I consider this a validation of
llvm-gcc or a condemnation of badly-written test suites (your choice).  It is
the "make test-all" that shows behaviour sharply divergent from GCC.  About 2
minutes into the test (on my machine) the test suite fails with a sudden and
very complete core dump.  The nature of the test is such that I haven't yet
figured out which file is being run and what it's doing, but that is on my "to
do" list.  Any further information I get will be appended to this report.

Incidentally, GCC also fails on that test suite, but it fails after going
through a lot more tests and with a completely different error that reports
where it was and what it was doing, so whatever is happening with LLVM-GCC is
pretty drastic given its inability to report where it was when it died.


-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.