Compiler bug or bad combo: WITH_DEBUG, FindKDE4Internals.cmake, devel/kdebindings4-python-pykde4 and bsd.cmake.mk

Sat May 16 15:24:57 UTC 2009

Hi,

I wanted to track down several issues with kde-4 and as such, started with 
this:
cat -n /etc/make.conf
    12  .if !empty(.CURDIR:M*/usr/ports/*)
    13  .if !defined(USE_STANDARD_CFLAGS)
    14  DEBUG_FLAGS=-ggdb
    15  WITH_DEBUG=yes
    20  .endif
        # more port specific stuff here, unrelated.

This presented me with a problem for the devel/kdebindings4-python-pykde4 
module. I've let the compilation of one object file (see below for details) 
run for >30 minutes wall clock /and/ CPU time and then pressed CTRL-C.

Adding the following:
    16  .if !empty(.CURDIR:M/usr/ports/devel/kdebindings4-python-pykde4)
    17  CMAKE_BUILD_TYPE=DEBUGFULL
    18  .else
    19  CMAKE_BUILD_TYPE=DEBUG

Fixed the problem and the entire port compiles and installs in under 10 
minutes.

This is on:
8.0-CURRENT FreeBSD 8.0-CURRENT #0 r192130: Fri May 15 11:19:04 CEST 2009

Things tried and tested:
- DISABLE_MAKE_JOBS: no effect
- CMAKE_BUILD_TYPE unset: no effect
- CMAKE_BUILD_TYPE=DEBUG: no effect
-   14  DEBUG_FLAGS=-ggdb -fno-strict-aliasing                                            
    15  WITH_DEBUG=yes                                                                    
    16  #.if !empty(.CURDIR:M/usr/ports/devel/kdebindings4-python-pykde4)                 
    17  #CMAKE_BUILD_TYPE=DEBUGFULL                                                       
    18  #.else                                                                            
    19  CMAKE_BUILD_TYPE=DEBUG                                                            
	No effect.

At present:
last pid: 89647;  load averages:  3.03,  2.60,  1.95               up 
1+04:13:57  16:55:42
195 processes: 4 running, 190 sleeping, 1 zombie
CPU: 74.0% user,  0.0% nice, 23.2% system,  2.2% interrupt,  0.6% idle
Mem: 868M Active, 285M Inact, 207M Wired, 28M Cache, 112M Buf, 103M Free
Swap: 3072M Total, 75M Used, 2997M Free, 2% Inuse

  PID USERNAME       THR PRI NICE   SIZE    RES STATE   C   TIME   WCPU 
COMMAND
89623 mel              1 115    0   241M   232M CPU1    1   9:09 87.79% 
cc1plus

cd /stable/usr/obj/usr/ports/devel/kdebindings4-python-
pykde4/work/kdebindings-4.2.3/python/pykde4 && 
/usr/local/libexec/ccache/world-c++   -D_GNU_SOURCE -DQT_NO_STL -
DQT_NO_CAST_TO_ASCII -D_REENTRANT -DKDE_DEPRECATED_WARNINGS -DQT3_SUPPORT -
D_REENTRANT -DQT_NO_DEBUG -DQT_CORE_LIB -DQT_GUI_LIB -
Dpython_module_PyKDE4_kdeui_EXPORTS -pipe -ggdb -fno-strict-aliasing -
Woverloaded-virtual -fvisibility=hidden -fvisibility-inlines-hidden -g -O2 -
fno-reorder-blocks -fno-schedule-insns -fno-inline -fPIC -
I/stable/usr/obj/usr/ports/devel/kdebindings4-python-
pykde4/work/kdebindings-4.2.3/python/pykde4 -
I/stable/usr/obj/usr/ports/devel/kdebindings4-python-
pykde4/work/kdebindings-4.2.3 -I/usr/local/kde4/include -
I/usr/local/kde4/include/KDE -I/usr/local/include/qt4/QtWebKit -
I/usr/local/include/qt4/QtHelp -I/usr/local/include/qt4/QtDBus -
I/usr/local/include/qt4/QtTest -I/usr/local/include/qt4/QtUiTools -
I/usr/local/include/qt4/QtScript -I/usr/local/include/qt4/QtSvg -
I/usr/local/include/qt4/QtXml -I/usr/local/include/qt4/QtSql -
I/usr/local/include/qt4/QtOpenGL -I/usr/local/include/qt4/QtNetwork -
I/usr/local/include/qt4/QtDesigner -I/usr/local/include/qt4/Qt3Support -
I/usr/local/include/qt4/QtGui -I/usr/local/include/qt4/QtCore -
I/usr/local/include/qt4/Qt -I/usr/local/share/qt4/mkspecs/default -
I/usr/local/include/qt4 -I/usr/local/include -I/usr/local/include/python2.6 -
I/usr/local/kde4/include/solid -I/usr/local/kde4/include/kio -
I/usr/local/kde4/include/kdeprint -I/usr/local/kde4/include/kdeprint/lpr -
I/usr/local/kde4/include/dom -I/usr/local/kde4/include/ksettings -
I/usr/local/kde4/include/knewstuff2 -I/usr/local/kde4/include/dnssd -o 
CMakeFiles/python_module_PyKDE4_kdeui.dir/sip/kdeui/sipkdeuipart0.o -c 
/stable/usr/obj/usr/ports/devel/kdebindings4-python-
pykde4/work/kdebindings-4.2.3/python/pykde4/sip/kdeui/sipkdeuipart0.cpp
load: 3.03  cmd: cc1plus 89623 [runnable] 413.90u 120.55s 87% 264348k

gdb yields nothing useful. A ktrace and subsequent:
kdump|egrep -v '(mmap|madvise|munmap)' >kdump.out

yields assembly code and writes in 4096 blocks.

Have not tried to reproduce this issue on -stable. I can make the kdump 
available for those interested.

Is this a compiler bug or will this code eventually finish (30 minutes CPU 
time still seems like a bug to me, especially when without -O2 it compiles in 
2 or 3 seconds).

As a sidenote, if anyone from kde@ is reading current:

The following assumption in bsd.cmake.mk is not true for kde4 builds:
# CMAKE_BUILD_TYPE  - Type of build (cmake predefined build types),
#                   affects on CFALGS and thus should not be set.
#                   Default: none (which respects CFLAGS)

As per kde4/share/apps/cmake/modules/FindKDE4Internals.cmake:
   set(CMAKE_C_FLAGS_RELEASE          "-O2 -DNDEBUG -DQT_NO_DEBUG")

Default CMAKE_BUILD_TYPE if unset is RELEASE, which then *adds* 
CMAKE_C(XX)?_FLAGS_RELEASE to ${CFLAGS}, negating the stripping of -O2 by 
WITH_DEBUG in bsd.ports.mk:
.if defined(WITH_DEBUG) && !defined(WITHOUT_DEBUG)
STRIP=  #none
STRIP_CMD=  ${TRUE}
DEBUG_FLAGS?=   -g
CFLAGS:=        ${CFLAGS:N-O*:N-fno-strict*} ${DEBUG_FLAGS}
.endif

So, it would be nice if bsd.cmake.mk would pick up on WITH_DEBUG and set 
CMAKE_BUILD_TYPE to DEBUGFULL, which has the expected effect from the end-user 
perspective.

-- 
Mel