make test failures on Solaris (SPARC)

[ggreif]

Bugzilla Link	1682
Resolution	FIXED
Resolved on	Mar 12, 2010 00:57
Version	unspecified
OS	Solaris
CC	@efriedma-quic,@ggreif

Extended Description

Here is the list I got today:

Sema/cocoa.m
******************** TEST 'Sema/cocoa.m' FAILED! ********************
Command:
clang Sema/cocoa.m -arch ppc
Output:
Sema/cocoa.m:3:10: error: 'Cocoa/Cocoa.h' file not found
#include <Cocoa/Cocoa.h>
^
1 diagnostic generated.
******************** TEST 'Sema/cocoa.m' FAILED! ********************

Sema/switch.c
******************** TEST 'Sema/switch.c' FAILED! ********************
Command:
clang -parse-ast-check Sema/switch.c
Output:
Errors expected but not seen:
Line 11: previous case value
Line 13: duplicate case value
Errors seen but not expected:
Line 11: duplicate case value '42'
Line 13: previous case value occurrence defined here
******************** TEST 'Sema/switch.c' FAILED! ********************

Sema/carbon.c
******************** TEST 'Sema/carbon.c' FAILED! ********************
Command:
clang Sema/carbon.c -fsyntax-only -arch ppc
Output:
Sema/carbon.c:3:10: error: 'Carbon/Carbon.h' file not found
#include <Carbon/Carbon.h>
^
1 diagnostic generated.
******************** TEST 'Sema/carbon.c' FAILED! ********************

Sema/format-strings.c
******************** TEST 'Sema/format-strings.c' FAILED! ********************
Command:
clang -parse-ast-check Sema/format-strings.c
Output:
Errors seen but not expected:
Line 10: use of undeclared identifier '__builtin_va_alist'
Line 51: use of undeclared identifier '__builtin_va_alist'
Line 60: use of undeclared identifier '__builtin_va_alist'
******************** TEST 'Sema/format-strings.c' FAILED! ********************

The switch.c test seems to give the same diagnostics, but attributes them to the exchanged lines. Very strange. Maybe it is some indeterministic ordering issue.

My system compiler is

ggreif@dengsvr18 [#273] g++ --version
g++ (GCC) 3.4.6
Copyright (C) 2006 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Cheers,

Gabor

[lattner]

The Sema/cocoa.m and carbon.c failures are unfortunate, because clang doesn't have a configure process, the predefined macros assume you're on a mac (see Driver/Targets.cpp). We need to fix this by building up the driver infrastructure.

The Sema/switch.c failure is very strange and I'm not sure why it fails.

The Sema/format-strings.c failure seems related to not handling something in the solaris headers. I think andersca has a patch or an approach for this.

-Chris

[ggreif]

Sema/switch.c is not so mysterious when you look
at

Sema/SemaStmt.cpp:337

// Sort all the scalar case values so we can easily detect duplicates.
std::stable_sort(CaseVals.begin(), CaseVals.end());

if (!CaseVals.empty()) {
for (unsigned i = 0, e = CaseVals.size()-1; i != e; ++i) {
if (CaseVals[i].first == CaseVals[i+1].first) {
// If we have a duplicate, report it.
Diag(CaseVals[i+1].second->getLHS()->getLocStart(),
diag::err_duplicate_case, CaseVals[i].first.toString());
Diag(CaseVals[i].second->getLHS()->getLocStart(),
diag::err_duplicate_case_prev);
// FIXME: We really want to remove the bogus case stmt from the substmt,
// but we have no way to do this right now.
CaseListIsErroneous = true;
}
}
}

stable_sort is not allowed to reorder inside the equivalence classes,
so if the CaseVals has the order before wrong, then they will
stay wrong. This does not explain the problem, but shifts it
into the original routine that fills CaseVals.

The switchcases come from this loop:

for (SwitchCase *SC = SS->getSwitchCaseList(); SC;
SC = SC->getNextSwitchCase()) {

my guess is that the SS->getSwitchCaseList() has undefined order.

Maybe this lights it up a bit.

Cheers,

Gabor

[ggreif]

AAAAHHHHH!

Got it. stable_sort uses std::pair's comparison predicate,
and the second component is a pointer!

You should supply a custom predicate that looks into
the Location.

Do you want that I come up with a patch?

BTW, that loop does some unneeded indexing,
one could remember the previous value and start
at index 1.

Cheers++,

Gabor

[ggreif]

raw patch

[lattner]

Great approach. I finished it off and applied it:
http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20070917/002063.html

Thanks!

[efriedma-quic]

There hasn't been any activity on this for ages; what's still failing?

[ggreif]

Here are the failures on Solaris:

XFAIL: CodeGen/Generic/2007-04-08-MultipleFrameIndices.ll for #1680 ,1557
Failed with exit(1) at line 1
while running: llvm-as < llvm/test/CodeGen/Generic/2007-04-08-MultipleFrameIndices.ll | llc
Couldn't allocate output reg for contraint 'r'!

XFAIL: CodeGen/Generic/2007-04-14-EHSelectorCrash.ll for #1880 ,1326
Failed with exit(1) at line 1
while running: llvm-as < llvm/test/CodeGen/Generic/2007-04-14-EHSelectorCrash.ll | llc -enable-eh
Assertion failed: noAssert && "Invalid LLVM intrinsic name", file Function.cpp, line 304
llc: bitcode didn't read correctly.
Reason: Invalid bitcode signature

XFAIL: CodeGen/Generic/2007-04-27-InlineAsm-X-Dest.ll for #1929
Failed with exit(1) at line 1
while running: llvm-as < llvm/test/CodeGen/Generic/2007-04-27-InlineAsm-X-Dest.ll | llc
Couldn't allocate output reg for contraint 'r'!

XFAIL: CodeGen/Generic/2007-04-27-LargeMemObject.ll for #1929
Failed with exit(1) at line 1
while running: llvm-as < llvm/test/CodeGen/Generic/2007-04-27-LargeMemObject.ll | llc
Could not match memory address. Inline asm failure!

XFAIL: CodeGen/Generic/2007-11-21-UndeadIllegalNode.ll
Failed with signal(SIGABRT) at line 1
while running: llvm-as < llvm/test/CodeGen/Generic/2007-11-21-UndeadIllegalNode.ll | llc -o -
Assertion failed: NVT > VT && MVT::isInteger(NVT) == MVT::isInteger(VT) && "Cannot promote to smaller type!", file LegalizeDAG.cpp, line 4073

FAIL: CodeGen/Generic/2008-02-20-MatchingMem.ll for #1505
Failed with exit(1) at line 1
while running: llvm-as < llvm/test/CodeGen/Generic/2008-02-20-MatchingMem.ll | llc
Could not match memory address. Inline asm failure!

FAIL: CodeGen/Generic/asm-large-immediate.ll
Failed with exit(1) at line 1
while running: llvm-as < llvm/test/CodeGen/Generic/asm-large-immediate.ll | llc | grep 68719476738
Invalid operand found in inline asm: '/* result: ${0:c} /'
INLINEASM <es:/ result: ${0:c} */>, 11, 68719476738

[efriedma-quic]

Umm, this was originally a clang bug, wasn't it?

[ggreif]

I have built a pretty recent clang and did:

make report
make[1]: Entering directory `llvm/tools/clang/test'
******************** TEST 'CodeGen/mandel.c' FAILED! ********************
Command:
clang -emit-llvm CodeGen/mandel.c
Incorrect Output:
CodeGen/mandel.c:44:72: error: use of undeclared identifier '_Imaginary_I'
double complex c = (START_X+xstep_X) + (START_Y+ystep_Y) * I;
^
1 diagnostic generated.
******************** TEST 'CodeGen/mandel.c' FAILED! ********************
******************** TEST 'Parser/objc-try-catch-1.m' FAILED! ********************
Command:
clang -fsyntax-only -verify Parser/objc-try-catch-1.m
Incorrect Output:
Bus Error
******************** TEST 'Parser/objc-try-catch-1.m' FAILED! ********************
******************** TEST 'Sema/builtins.c' FAILED! ********************
Command:
clang Sema/builtins.c -fsyntax-only -verify -pedantic
Incorrect Output:
Errors expected but not seen:
Line 39: too few arguments to function
Warnings seen but not expected:
Line 39: implicit declaration of function '__builtin_ia32_vec_ext_v4si' is invalid in C99
******************** TEST 'Sema/builtins.c' FAILED! ********************

The Bus Error in Parser/objc-try-catch-1.m looks serious.

[ggreif]

Here is the reason for above Bus Error:

(gdb) run -fsyntax-only -verify test/Parser/objc-try-catch-1.m
Starting program: /home/ggreif/llvm/Debug/bin/clang -fsyntax-only -verify test/Parser/objc-try-catch-1.m
warning: Temporarily disabling breakpoints for unloaded shared library "/usr/lib/ld.so.1"
warning: Lowest section in /lib/libpthread.so.1 is .dynamic at 00000074

Program received signal SIGSEGV, Segmentation fault.
0x004d2958 in clang::Stmt::DestroyChildren (this=0x77f790, C=@0xffbfee58)
at Stmt.cpp:47
47 if (Stmt* Child = *I) Child->Destroy(C);
(gdb) bt
#​0 0x004d2958 in clang::Stmt::DestroyChildren (this=0x77f790, C=@0xffbfee58)
at Stmt.cpp:47
#​1 0x004d29b0 in clang::Stmt::Destroy (this=0x77f790, C=@0xffbfee58) at Stmt.cpp:51
#​2 0x004d2974 in clang::Stmt::DestroyChildren (this=0x77f560, C=@0xffbfee58)
at Stmt.cpp:47
#​3 0x004d29b0 in clang::Stmt::Destroy (this=0x77f560, C=@0xffbfee58) at Stmt.cpp:51
#​4 0x004d2974 in clang::Stmt::DestroyChildren (this=0x783d88, C=@0xffbfee58)
at Stmt.cpp:47
#​5 0x004d29b0 in clang::Stmt::Destroy (this=0x783d88, C=@0xffbfee58) at Stmt.cpp:51
#​6 0x004d2974 in clang::Stmt::DestroyChildren (this=0x79cba0, C=@0xffbfee58)
at Stmt.cpp:47
#​7 0x004d29b0 in clang::Stmt::Destroy (this=0x79cba0, C=@0xffbfee58) at Stmt.cpp:51
#​8 0x004bb32c in clang::FunctionDecl::Destroy (this=0x79c6e0, C=@0xffbfee58)
at Decl.cpp:453
#​9 0x004ec730 in ~TranslationUnit (this=0xffbfee40) at TranslationUnit.cpp:40
#​10 0x00439844 in clang::ParseAST (PP=@0x785298, Consumer=0x781ba0, PrintStats=false)
at ParseAST.cpp:75
#​11 0x003048e0 in clang::CheckASTConsumer (PP=@0x785298, C=0x781ba0)
at DiagChecker.cpp:231
#​12 0x00337018 in ProcessInputFile (PP=@0x785298, PPF=@0xffbff1d8, InFile=@0x781b50)
at clang.cpp:1266
#​13 0x003380f0 in main (argc=4, argv=0xffbff44c) at clang.cpp:1480

[efriedma-quic]

I have built a pretty recent clang and did:

make report
make[1]: Entering directory `llvm/tools/clang/test'
******************** TEST 'CodeGen/mandel.c' FAILED! ********************
Command:
clang -emit-llvm CodeGen/mandel.c
Incorrect Output:
CodeGen/mandel.c:44:72: error: use of undeclared identifier '_Imaginary_I'
double complex c = (START_X+xstep_X) + (START_Y+ystep_Y) * I;
^
1 diagnostic generated.
******************** TEST 'CodeGen/mandel.c' FAILED! ********************

Mandel is nothing but trouble :( fails on Linux too.

******************** TEST 'Parser/objc-try-catch-1.m' FAILED!

---

Command:
clang -fsyntax-only -verify Parser/objc-try-catch-1.m
Incorrect Output:
Bus Error
******************** TEST 'Parser/objc-try-catch-1.m' FAILED!

Real cross-platform bug; valgrind says it's a conditional jump on an uninitialized value.

******************** TEST 'Sema/builtins.c' FAILED! ********************
Command:
clang Sema/builtins.c -fsyntax-only -verify -pedantic
Incorrect Output:
Errors expected but not seen:
Line 39: too few arguments to function
Warnings seen but not expected:
Line 39: implicit declaration of function '__builtin_ia32_vec_ext_v4si' is
invalid in C99
******************** TEST 'Sema/builtins.c' FAILED! ********************

Perhaps a bug in the test? I don't think we should be defining __builtin_ia32_vec_ext_v4si for a SPARC target.

[efriedma-quic]

The failure on Parser/objc-try-catch-1.m should be fixed in r51551.

[efriedma-quic]

And the Sema/builtins.c failure should be fixed in r51552.

[efriedma-quic]

And CodeGen/mandel.c should be "fixed" in r51554. Please comment if there are still any issues.

[ggreif]

Nice work Eli!

make report is clean now :-)