Build notes for ds10
- October 19, 2012
- gcc (4.6.1), a normal
makecommand results in a link failure:
Could not find the cause of it, but can work around it by first building the compiler "gcc", and next the remaining stuff (support libraries etc):libbackend.a(tree-vect-data-refs.o): In function `get_insns': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:71: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o) libbackend.a(tree-vect-data-refs.o): In function `set_first_insn': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:80: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o) libbackend.a(tree-vect-data-refs.o): In function `get_last_insn': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:88: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o) libbackend.a(tree-vect-data-refs.o): In function `set_last_insn': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:97: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o) libbackend.a(tree-vect-data-refs.o): In function `get_max_uid': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:105: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o) libbackend.a(ddg.o): In function `get_insns': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:71: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o) libbackend.a(ddg.o): In function `set_first_insn': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:80: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o) libbackend.a(ddg.o): In function `get_last_insn': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:88: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o) libbackend.a(ddg.o): In function `set_last_insn': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:97: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o) libbackend.a(ddg.o): In function `get_max_uid': /usr/src/gcc-4.6.1_build/gcc/../../gcc-4.6.1/gcc/emit-rtl.h:105: relocation truncated to fit: GPREL16 against symbol `x_rtl' defined in .bss section in libbackend.a(emit-rtl.o)
This however results in a problem with the "clean" target of$ make all-gcc $ makemake. VariousMakefile's refer to the binary "xgcc" (located somewhere in the build tree) and this binary is apparently removed at an early stage. When executing the "clean" target manually theMakefiles trip over the missing "xgcc". One can either remove the entire build tree or clean every directory separately and point to an existing "gcc" binary, like this:$ for dir in `find . -type f -name Makefile -printf '%h\n'`; do make -C $dir CC=gcc clean; done
— UPDATE (September 2, 2020) —
Better solution is to create a build config in
<gcc-srcdir>/config/ and use that file as an additional
config during the configuration of GCC. The
build config file contains linker flags which will prevent the problem
above from occurring.
Create the build config file <gcc-srcdir>/config/alpha-linux.mk
with the following contents:
Now, configuring GCC and use the option
LDFLAGS = -Wl,--no-relax
BOOT_LDFLAGS = -Wl,--no-relax
--with-build-config to
choose the build configuration(s):
Build config
./configure [options]... --with-build-config='bootstrap-debug alpha-linux'
bootstrap-debug is the default config when
--with-build-config is omitted, so you need to specify the
default config in addition to the alpha-linux config.
Note that a make distclean "forgets" to remove the file
fixincludes/config.cache. Remove this file before doing a
new build.
- October 20, 2012
- jpeg-6b, applied some of the optimization hints described in
install.doc:
jconfig.h: add#define JDCT_DEFAULT JDCT_FLOAT
(on Alpha floats are faster than ints)jmorecfg.h: changed typedef ofJCOEFfrom "short" to "int"
(on Alpha 32 bits ops (i.e. int, long) are faster than 8 or 16 bits ops)
Need to replaceconfig.subandconfig.guessby a newer version since they do not recognize ev67 (only applicable when building libraries).
--infodirand--mandiroptions not honoured when running./configurescript, so modified mainMakefile.
- October 23, 2012
- pixman (0.26.2), optimizing with
-O2or higher causes the test suite to fail onlt-stress-testandlt-blitters-test.
- October 24, 2012
- mozilla
all:Makefilexpcom/reflect/xptcall/src/md/unix/Makefiledoes not match Alpha architecture correctly. Lookup Alpha definition section in the file and add a "%"-sign to the wordLinuxalphaof thefiltercommand.Makefileff-opt/xpcom/reflect/xptcall/src/md/unix/Makefilemay suffer from the exact same problem.
1.9.2:- File
xpcom/glue/nsEnumeratorUtils.cpphas an uninitializedconstwhich fails compiling. Copied build command for this file, ran it manually, and added the-fpermissiveflag to the build command to allow uninitialized constants. Alternatively, a "target-specific variable" could be added toxpcom/glue/Makefile:
The following files suffer from the same issue:nsEnumeratorUtils.o: COMPILE_CXXFLAGS += -fpermissivexpcom/glue/standalone/nsEnumeratorUtils.cpp xpcom/base/nsDebugImpl.cpp xpcom/base/nsTraceRefcntImpl.cpp xpcom/io/nsUnicharInputStream.cpp xpcom/build/nsEnumeratorUtils.cpp storage/src/mozStorageStatement.cpp layout/style/nsCSSRuleProcessor.cpp toolkit/components/places/SQLFunctions.cpp toolkit/xre/nsAppRunner.cpp embedding/browser/gtk/src/EmbedPrivate.cpp --disable-ipc
No support for atomic operations on Alpha inipc/chromium/base/src/atomicops.h.--enable-oggor--enable-wave
DefinesMOZ_MEDIAwhich prevents an "undefined reference" incontent/base/src/nsContentSink.cpp.--disable-tests
NoRETURN_INSTRdefinition for Alpha inlayout/base/tests/TestPoisonArea.cpp.--enable-zipwriter
Prevents a build error due to missingnsIZipWriter.h.- Including (system) header
stddef.hin filegfx/ots/src/os2.cc:
The header file provides the macro#include <stddef.h>offsetofwhich is needed byos2.cc. Occurence of the problem depends on the GCC version being used (4.6.1 in this case).
6.0.1; 10.0.1; 20.0.1:- In
ipc/chromium/src/build/build_config.hadded Alpha support:#elif defined(__alpha__) #define ARCH_CPU_ALPHA 1 #define ARCH_CPU_64_BITS 1 - In
gfx/ycbcr/chromium_types.hadded Alpha support:#elif defined(__alpha__) #define ARCH_CPU_ALPHA_FAMILY 1 #define ARCH_CPU_ALPHA 1 #define ARCH_CPU_64_BITS 1 - In
xpcom/glue/nsID.h,struct nsID, set alignment for memberm0appropriate for your machine. E.g. for a DS10 (EV67) the preferred alignment is 8 bytes:
Without proper alignment you get tons of unalignment traps from the kernel.PRUint32 m0 __attribute__((aligned (8)));
10.0.1:- Applied patches.
- Removed
staticqualifiers inwidgets/src/gtk2/gtk2compat.hsince it's conflicting with the GTK header files. - In
js/src/yarr/BumpPointerAllocator.h, after the opening of the "WTF" namespace, added support for Alpha. The size corresponds to the machine's page size (refer to/proc/cpuinfo).#elif WTF_CPU_ALPHA #define MINIMUM_BUMP_POOL_SIZE 0x2000
20.0.1:- Applied patch to allow using "system" Cairo instead of Mozilla native.
- Removed
staticqualifiers inwidget/gtk2/compat/gdk/gdkwindow.hsince it's conflicting with the GTK header files. In addition, I've commented out the function defintions ingdkwindow.has this header file is (indirectly) included in bothwidget/gtk2/gtk2drawing.candwidget/gtk2/mozcontainer.c. When thelibxul.solibrary is created, these two files cause duplicate function definitions and linkinglibxul.sofails (using GTK 2.24.13). - Speaking of
libxul.so, libraries of Ogg, Vorbis, Theroa are not included during the link oflibxul.socausing undefined references. Added these libraries to theMakefileoflibxul.soin . Add the following to variableEXTRA_DSO_LDOPTSfor theMOZ_ALSAconditional:
Furthermore a problem occurs when linking with the GCC 4.6 series ("relocation truncated to fit" error). TheEXTRA_DSO_LDOPTS += -logg -lvorbis -ltheora $(MOZ_ALSA_LIBS)--no-relaxflag needs to be passed to the linker, so in theMakefile.infile oflibxul.soI've added to the variableEXTRA_DSO_LDOPTSthe linker option-Wl,--no-relax. Alternatively, this could be added to the regularMakefileoflibxul.soinstead. - In
js/src/yarr/BumpPointerAllocator.h, after the opening of the "WTF" namespace, added support for Alpha. The size corresponds to the machine's page size (refer to/proc/cpuinfo).#elif WTF_CPU_ALPHA #define MINIMUM_BUMP_POOL_SIZE 0x2000
Also, there's a runtime check of the page size going wrong when jemalloc is
being used (which is the default). The check assumes a page size of 4k
while the page size on Alpha is 8k. Open file
memory/mozjemalloc/jemalloc.caround line 1115 and change the "define" forpagesize_2powto 13 instead of 12 (or wrap it in a conditional directive). If jemalloc is disabled (--disable-jemalloc), then the page size seems to be defined injs/public/HeapAPI.h, around line 25. - Xorg (X11R7.7)
Used build order:
X printing support is changed or missing, so copiedintltool (freedesktop.org) util/makedepend util/util-macros proto/glproto proto/dri2proto (freedesktop.org) proto/xf86driproto proto/bigreqsproto proto/compositeproto proto/dmxproto proto/fixesproto proto/fontsproto proto/inputproto proto/kbproto proto/randrproto proto/recordproto proto/renderproto proto/resourceproto proto/scrnsaverproto proto/videoproto proto/xcmiscproto proto/xextproto proto/xf86bigfontproto proto/xf86dgaproto proto/xf86vidmodeproto proto/xineramaproto proto/xproto xcb/xcb-proto xcb/libpthread-stubs xcb/libxcb lib/libdrm (freedesktop.org) lib/Mesa (freedesktop.org) lib/libFS lib/libICE lib/libSM lib/libX11 lib/libXScrnSaver lib/libXau lib/libXaw lib/libXcomposite lib/libXcursor lib/libXdamage lib/libXdmcp lib/libXext lib/libXfixes lib/libXfont lib/libXi lib/libXinerama lib/libXmu lib/libXpm lib/libXrandr lib/libXrender lib/libXres lib/libXt lib/libXtst lib/libXv lib/libXvMC lib/libXxf86dga lib/libXxf86vm lib/libdmx lib/libfontenc lib/libpciaccess lib/libxkbfile lib/xtrans app/* data/* font/font-util font/* xserver/xorg-server driver/* doc/xorg-docslibXp,libXprintUtil,libXprintAppUtil, andprintprotofrom the 7.3 release.
Theunsupporteddirectory requires tools likeimakewhich is no longer part of the 7.7 release, so copied the latest version ofimake,gccmakedep,lndir, andxorg-cf-filesfrom the 7.3 release.
- October 29, 2012
- glib (2.32.4), old header file
/usr/include/glib-2.0/gio/gperiodic.hmay be left after installing. This header file causes a link error forgobject-introspection(1.31.22) (undefined referenceg_periodic_get_type).
- November 16, 2012
- lynx (2.8.7), when configuring with SSL support
(
--with-ssloption), linking the final executable may be missing thedllibrary. The./configurescript can use thepkg-configutility to determine compile and link flags for SSL; the link flags may be missing thedllibrary resulting in undefined references whenlynxis being linked.
Specifying--with-ssl=<path>where<path>is/usrfor instance, triggers additional tests in the./configurescript in which case thedllibrary will be used during the final link oflynx.
- November 21, 2012
- libgsf (1.14.4), the files
gsf/gsf-infile-stdio.c,gsf/gsf-output-csv.c, andgsf/gsf-output-iconv.care directly including Glib-specific header files which doesn't seem to be allowed by Glib. It seems you only have to includeglib.h. - abiword (2.8.6), files
goffice-bits/goffice/app/goffice-app.handsrc/af/util/xp/ut_go_file.hinclude Glib-specific headers. Must useglib.hheader file instead.
- November 27, 2012
- abiword (2.8.6), encountered link error. In
src/Makefileremoved the flag-avoid-versionfrom theLDFLAGSdefinitions since this flag is not recognized by the compiler or the linker. Other linker flags used in this makefile are--no-undefinedand-export-dynamic. They are passed to the compiler, g++, but since they are linker flags they should be passed told. Therefore, these options are prefixed with the-Wlflag:
and-Wl,--no-undefined
Note the double dash in front of the-Wl,--export-dynamic--export-dynamicflag.
- December 10, 2012
- lcms (2-2.4), must compile with IEEE floating point compatibility else
the test suite breaks with a floating point exception. So, added
-mieeetoCFLAGS.
- January 3, 2013
- firefox (6.0.1), compiling Firefox with binutils 2.21 requires the
option
--no-relaxto be added to the linker flags. The flag can be added to the variableOS_LDFLAGSin the fileconfig/autoconf.mk(located in directory where Firefox is being built). E.g:
Previous versions of binutils did not need theOS_LDFLAGS = -lpthread -Wl,--no-relax--no-relaxoption, because they apparently use it by default.
- January 6, 2013
- libtool. When a library is installed in a new location,
libtoolarchives (i.e.*.lafiles) depending on the library may contain the path to the old library. For example, if X11 libraries were once installed in/usr/liband now you want to have a separate tree for all X11 stuff, say/usr/X11, then archives likelibcairo*.laandlibgtk*.lastill expect certain X11 libraries in/usr/lib. So, the*.lafiles are not automatically kept in sync.
In case a library has been relocated without the depending libraries know about it,libtoolemits errors like "no such file or directory" and "not a valid libtool archive".
- January 8, 2013
- pixman (0.26.2), compile flag
-O3and-funroll-loopsseem a bit too much, assembler bails out at some point. - libgsf (1.14.4), when running the test suite, a duplicate definition of
the function
clonemay cause the test suite to prematurely fail. An unrelatedclonefunction is also defined in one of the system header files, so this seems to be an unfortunate choice of names. There are 4 files inlibgsfusing aclonefunction:
In each file I've renamed thetests/test-cp-msole.c tests/test-cp-zip.c tests/test-dump-msole.c tests/test-restore-msole.cclonefunction togsf_clone. - python (2.5.2), may need to add compiler
flag
-mieeetoCFLAGSvariable. Without this flag the floating point test will fail.
- January 14, 2013
- Xaw3d (1.5E), in order to build the (separate) Athena widget library
(Xaw3d) I've followed the instructions to build the library without an X11
source tree. This will create a proper
Makefileto build the library.
- January 15, 2013
- babl (0.1.10), due to many unaligned traps in code generated by GCC,
I've used Compaq Alpha compiler
cccto buildbabl. Simply passedCC=cccwhen running the./configurescript. - ccc, in header file
/usr/include/bits/pthreadtypes.hthere may be a member variable in use by the name of__align. This happens to be a reserved keyword forccc, so when usingcccthe member variable must be renamed.
- January 17, 2013
- NSS/NSPR (3.14.1), used combined tarball of NSS and NSPR to build
correctly (or must move NSPR tree in NSS tree). There's no
./configurescript to build NSS; the build starts by runningmakein the directory<nss-release>/mozilla/security/nss. However, NSS expects NSPR to be located in the directory<nss-release>/mozilla/nsprpuband it will attempt to automatically build NSPR. If you want a specific configuration for NSPR then you must build NSPR separately in a "platform" sub-directory. The name of the "platform" directory can be determined by using the NSSMakefile:
Return to the NSS directory and build NSS:# cd <nss-release>/mozilla/security/nss # make BUILD_OPT=1 USE_64=1 platform Linux2.6_alpha_glibc_PTH_DBG.OBJ # cd <nss-release>/mozilla/nsprpu # mkdir Linux2.6_alpha_glibc_PTH_DBG.OBJ # cd Linux2.6_alpha_glibc_PTH_DBG.OBJ # ../configure <option-list>
Note that build options influence the "platform" string, so ensure the "platform" target and the "nss_build_all" use the same build options. Also applies to "install" target.make BUILD_OPT=1 USE_64=1 nss_build_all
- January 18, 2013
- ccc, installation addendum
- In
/usr/lib/compaq/<ccc>/alpha-linux/bin, the scriptprobe_linux.shdoes not run completely correct. Modified the following:probe_gcc(), first line, determining GCCspecsfile. Original line commented and replaced by:specs=`gcc -print-search-dirs 2>&1 | grep install: | sed -e 's|install: ||g'`/specsprobe_taso(),cccommands for linkinglibtaso$$.soandlibtaso2$$.so: Must add-fPICflag else linker will have relocation errors.probe_demangle(),cccommand withdemangleflag: Demangling stylecompaqis no longer recognized, so usedhpinstead.probe_version(), the firstelifstatement is missing a "$" operator in front of variableDIST_FILES.
- In
/usr/lib/compaq/<ccc>/alpha-linux/bin, scriptcreate-comp-config.shrunsgccwith the-V <version>option. Recent versions of igccno longer recognize this option, so may need to remove the-Voption and its argument ifcreate-comp-config.shfails to run. - GCC
specsfile is needed byprobe_linux.sh, but recent GCC versions use a builtin version of that file. Createspecsfile in GCC's install directory:gcc -dumpspecs > /usr/lib/gcc/<arch-os_name>/<gcc-version>/specs - In
ccc'sbindirectory, adjust symlinks forcrtbegin.oandcrtend.o. Link them to thecrtbegin.oandcrtend.oof GCC. - In
/usr/bin, adjust symlink ofcccfile. - Copied the DEC specific header files
float.handlimits.hto the compiler's privateincludedirectory (i.e./usr/lib/compaq/<ccc>/alpha-linux/include). - Encountered some build errors in
/usr/include/bits/stdio-ldbl.hwhich were due to__LDBL_REDIR1_DECLbeing undefined. This defininion is defined in/ust/include/sys/cdefs.h. "Fixed" the problem by adding a dummy definition for__LDBL_REDIR1_DECL. - In
ccc'sbindirectory, added the following defines tocomp.config(at end of line):-D__LONG_DOUBLE_128__ -D_STDC_ -D_GNU_SOURCE - Post-installation step (performed by
rpmcommand): creating shared library oflibcpml.
(change Alpha EV variant as needed, ev5 or ev6)$ ld -shared -o libcpml_ev5.so -soname libcpml.so -whole-archive \ libcpml_ev5.a -no-whole-archive -lots
- In
- gegl (0.2.0), file
operations/common/matting-global.chas aminandmaxdefinition.cccseems to choke on the construction, so I've replaced it with a simpler construction forccc.
Also, in the filegegl/opencl/cl_platform.hthe macroCL_ALIGNEDis not defined forccc(i.e.__DECC). The keyword__aligncould be used in this case, however,CL_ALIGNEDis used in numerousuniontypes andcccdoes not accept the__alignkeyword there. Therefore, I've turned theCL_ALIGNEDmacro into a dummy, empty definition when usingccc.
During./configuresome tests are performed to check whether certain warning flags (-W) are supported by the compiler (variableDESIRED_CFLAGSin./configurescript).cccsupports some of these flags and ignores the unknown ones. However, whenccccalls the linkerld, it seems to pass the unknown flags to the linker which causes an error. In order to prevent this I've adjusted theDESIRED_CFLAGSvariable to only list the flags that are actually supported byccc.
Next problem, when creating a shared object withccc, object symbols may get stripped away despite the sequence
The libraries specified by-Wl,--whole-archive <libs>... -Wl,--no-whole-archive<libs>...are meant for the linker but it seemscccuses<libs>...for itself and leaves an empty "-Wl,--whole-archive" / "-Wl,--no-whole-archive" sequence. A simple workaround is to prefix each filename in<libs>...with-Wl, as such options are passed to the linker. (use-vflag to verify the commandscccruns)
Note on the__alignkeyword.cccexpects the alignment as a power of 2 which is different to how GCC expects the alignment (which is the number of bytes). When using some alignment macro, the following can be used to convert a GCC alignment specification tocccalignment./* * Translate GCC alignment (in bytes) * to CCC (DECC) alignment (in 2^x). */ #if defined( __DECC ) #define DECC_ALIGN_BYTE_1 0 #define DECC_ALIGN_BYTE_2 1 #define DECC_ALIGN_BYTE_4 2 #define DECC_ALIGN_BYTE_8 3 #define DECC_ALIGN_BYTE_16 4 #if defined( __alpha__ ) #define DECC_ALIGN_BYTE_32 5 #define DECC_ALIGN_BYTE_64 6 #define DECC_ALIGN_BYTE_128 7 #define DECC_ALIGN_BYTE_256 8 #endif /* __alpha__ */ #define DECC_ALIGN_BYTE_512 9 #if defined( __alpha__ ) #define DECC_ALIGN_BYTE_1024 10 #define DECC_ALIGN_BYTE_2048 11 #define DECC_ALIGN_BYTE_4096 12 #define DECC_ALIGN_BYTE_8192 13 #define DECC_ALIGN_BYTE_16384 14 #define DECC_ALIGN_BYTE_32768 15 #define DECC_ALIGN_BYTE_65536 16 #endif /* __alpha__ */ #define ALIGN(B) __align(DECC_ALIGN_BYTE_##B) #endif /* __DECC */ unsigned int ALIGN(2) i; /* 2-byte alignment */
- February 11, 2013
- sysbench (0.4.12), build error with sysbench's private
libtool. Replaced the privatelibtoolwith thelibtoolfrom/usr/binafter./configurehas run.
- March 26, 2013
- nmap (6.25),
nmapbugs out with a floating point exception, probably due to Alpha not being IEEE-compliant. Adding-mieeeflag toCCOPTvariable in the mainMakefileseems sufficient to resolve the problem.
- June 6, 2013
- guile (2.0.9), long build especially when
guilecompiler is busy. Total build time was about 2 hours.
- September 14, 2013
- wterm (6.2.9),
./configurescript fails to determineutmppath correctly. Search forac_cv_header_utmpx_h. There are two cases handling theutmppath (if/elseclause). Both are identical expect for the number ofutmppaths they check. So, to determine theutmppath, either make theifandelseclause test the sameutmppaths, or turn theelseclause into a separateifclause. Also, need to setuid or setgid thewtermbinary to an appropriate user or group such that theutmpfile can be updated.
- October 16, 2013
- ncompress (4.2.4), when using the
./buildscript to generate aMakefile(genmakeoption), run./buildwith a shell that does not support the modern command substitution syntax (i.e.$(some_cmd)). This kind of command substitution is used inside a HERE-document and causes some variables in the resultingMakefileto get "eliminated." TheMakefilewill fail to run. Also, compiler options need to be passed as "Special options..." (option 7), but only when aMakefileis to be generated. When using optionc(Compile compress), the options set by "Compiler options" (option 5) will be used correctly. In either case though, I've added the compiler option-DNOFUNCDEF=1to prevent "incompatible declaration" errors.
Manually installedcompress,uncompress(symlink) and their man pages.
- May 1, 2014
- lshw (B.02.17), added optimization level 2. Removed
pci.idsfrom variableDATAFILESbecause it's installed bypciutilspackage. Also changed the "install" target to install theDATAFILESin/usr/share.
- February 25, 2015
- xterm (223), mismatching include-guard name. File
xutf8.hmay trigger a preprocessor error due to an incorrect name of the_XLIB_H_conditional. The related header file,X11/Xlib.h, probably uses_X11_XLIB_H_as the name of its include-guard. - rstart (1.0.5), error in
./configurescript. Line 10435 defines the variableconfigdir:
The variable value contains a command substitution whereas a parameter substitution is intendent. Remove parenthesis or replace them with braces.configdir=$(sysconfdir)/X11/rstart
- March 12, 2015
- abiword (2.8.6), the main edit screen shows a very small cursor and
text is misaligned. This is due to some missing TrueType fonts. Installing
the font package
liberation-ttffixed the problem.
- March 16, 2015
- xpaint (2.7.6), mismatching include-guard name. In file
image.hthe structimageprocessinfois enclosed by the preprocessor directive#ifdef _XLIB_H_. This may trigger a build error due to an incorrect name of the_XLIB_H_conditional. The related header file,X11/Xlib.h, probably uses_X11_XLIB_H_as the name of its include-guard, so need to adjustimage.h. - xv (3.10a), disabled PNG support because
libpng15.patchfailed. Also note that in theMakefile, the variableLIBDIRspecifies the location of the documentation.
- September 16, 2015
- nagios-plugins (1.4.9), undefined identifier
NAMEDATALENinplugins/check_pgsql.c:72(PostgreSQL plugin). Using Postgres version 9.2.3 and apparently the mentioned define moved to a different file. Fixed by adding the following line toplugins/check_pgsql.c:
#include <pg_config_manual.h>
- September 17, 2015
- nrpe (2.8.1), there's a typo in the
./configurescript related to the option--with-nagios-group. Setting a group with this option is not honoured due to a typo in a variable name. Edit the./configurescript at line 6157. Rename variablenagios_grptonagios_group.
- November 15, 2015
- lame (3.99.5), build system detects the presence of
cccand uses it to compile source files but usesgccwhen linking a shared library. Some command options specific tocccare passed togccwhich causes a link failure. As a workaround I've disabled building shared libraries (--disable-shared).
- February 17, 2016
- id3lib (3.8.3),
./configurescript checks for incorrect header fileiomanip.h; this file is callediomanip. libc functionsmemmove()andmemcpy()not found in the filesinclude/id3/id3lib_strings.handinclude/id3/writers.h. Include header filestring.hin these files. In fileexamples/test_io.cppthestdnamespace is missing forcin,hex, anddec.
- August 20, 2016
- WindowMaker (0.95.7), if X11 libraries have been installed in a
non-standard location then the
./configurescript may fail the tests for XShape, XShm, Xmu, and RandR. The tests do not take into account the search paths for X11 libraries/includes specified by the--x-includesand--x-librariesflags. Use this patch to fix. In addition, a non-standard X11 location causes the linker to fail onlibWMaker.so.1due to being unable to find libX11. Linking can continue without this library. In filewmlib/Makefile, locate the definition oflibWMaker_la_LIBADD(line 374) and remove the X11 library (-lX11). - gimp (2.8.2), SIGFPE problems. Added
-mieeetoCFLAGSduring package configuration.
- December 29, 2016
- console-tools (0.2.3), updated
config.guessandconfig.subsince they did not recognize ev67 CPU. - In
kbdtools/showkey.c, line 269, added abreakkeyword for thedefaultclause of theswitchstatement. - In
contrib/codepage.c, line 230, fixed quoting.
- January 31, 2017
- amanda (2.5.1p3), requires static Kerberos libraries, but this
caused a build error with
libcom_err.a. In the./configurescript the list of (static) Kerberos libraries is set. This can be changed to a list of-loptions of the correct Kerberos libraries. For Kerberos 5, search the./configurescript for the variable assignments ofK5CRYPTO,K5SUPPORT, andKRB5LIBS, and convert to a-loption.
Note: version 2.5.1p3 has connection problems with Kerberos; moving to version 2.5.2p1 seem to resolve these problems.
