User of autotools-generated tarball gets error message: aclocal-1.13: command not found
I am distributing a tarball with installation scripts generated by autoconf version 2.69. Works fine on many different machines. Now a user, working on a fresh Arch Linux system, reports that configure executes properly, but make immediately terminates with the following error message:
/home/user/project/build-aux/missing: line 81: aclocal-1.13: command not found
WARNING: 'aclocal-1.13' is missing on your system.
You should only need it if you modified 'acinclude.m4' or
'configure.ac' or m4 files included by 'configure.ac'.
The 'aclocal' program is part of the GNU Automake package:
<http://www.gnu.org/software/automake>
It also requires GNU Autoconf, GNU m4 and Perl in order to run:
<http://www.gnu.org/software/autoconf>
<http://www.gnu.org/software/m4/>
<http://www.perl.org/>
make: *** [aclocal.m4] Error 127
There is no acinclude.m4 in the project directory. The user assures that he has not modified aclocal.m4, configure.ac, files in m4/, or anything else in the project.
The error message makes no sense for me: as I understand, aclocal is executed when I run autoreconf -if; it generates a file aclocal.m4 that is part of the tarball I distribute; there is no reason why a user's make command should request aclocal.
The user further reports that he can solve the problem for himself by running aclocal. However, that is not a clean solution: packagers are not supposed to request their users to have autotools installed, right?
For completeness, I attach the full configure.ac: is there anything wrong?
################################################################################
## FRIDA: fast reliable interactive data analysis ##
## configure.ac: used by 'autoreconf -i' to prepare for 'configure' ##
## (C) Joachim Wuttke, Sebastian Busch 2008- ##
## http://apps.jcns.fz-juelich.de/frida ##
################################################################################
################################################################################
## Generic initialization ##
################################################################################
AC_INIT([frida],[post-2.1.8c],[[email protected]])
# ^^^^^ delete "post-" for public releases
# project name must be "frida"; this determines the installation directories
AC_CONFIG_AUX_DIR([build-aux])
AC_CONFIG_MACRO_DIR([m4])
AM_INIT_AUTOMAKE([foreign]) # don't insert GNU blala files
################################################################################
## Select compiler and preprocessors ##
################################################################################
AC_PROG_CXX
AC_LANG(C++)
AX_CXX_COMPILE_STDCXX_11 # provided in directory m4
AC_SUBST(AM_CXXFLAGS,"-g -pedantic -Wall -Wno-sign-compare -Wno-unused-result -Werror")
# for valgrind --leak-check=full frida, use -O0 -fno-inline
# source files that use -D settings must #include "../config.h"
AC_CONFIG_HEADERS([config.h]) # also needed to prevent endless -D option lists
AC_PROG_LEX # LEX,LEXLIB = "flex","-lfl" or "lex","-ll"
if test "$LEX" = :; then
AC_MSG_ERROR([Cannot find lex. Aborting.])
fi
AC_PROG_YACC # YACC = "bison -y" or "byacc" or "yacc"
if test "$YACC" = "yacc"; then
AC_MSG_ERROR([Cannot find yacc. Aborting.])
fi
## I put this one here only because qmake does: Make sure off_t is 64-bit in *nix, taken from http://www.google.com/search?q=cache:wlNJ8Qe4dBgJ:www.sfr-fresh.com/unix/privat/libfb-v0.18.4b-src.zip:a/src/rtlib/configure.ac+configure.ac+file+offset+bits&hl=de&ct=clnk&cd=5&gl=de&client=firefox-a
AC_DEFINE([_FILE_OFFSET_BITS],64,[File offset bits])
################################################################################
## Check for headers, defs needed at compile time ##
################################################################################
# Consistency check: is source code present?
AC_CONFIG_SRCDIR([src/frida2.cpp])
AC_CONFIG_SRCDIR([man/frida.pod])
# Checks for typedefs, structures, and compiler characteristics
AC_C_CONST
AC_C_INLINE
AC_TYPE_SIZE_T
AC_STRUCT_TM
AC_HEADER_STDBOOL
AC_HEADER_STDC
# Checks for standard includes
AC_FUNC_ALLOCA
AC_FUNC_MALLOC
AC_FUNC_REALLOC
AC_FUNC_MKTIME
AC_FUNC_STRFTIME
AC_CHECK_FUNCS([gettimeofday],,AC_MSG_ERROR([Cannot find gettimeofday.]))
AC_CHECK_FUNCS([memset], ,AC_MSG_ERROR([Cannot find memset.]))
AC_CHECK_FUNCS([mkfifo], ,AC_MSG_ERROR([Cannot find mkfifo.]))
AC_CHECK_FUNCS([strchr], ,AC_MSG_ERROR([Cannot find strchr.]))
AC_CHECK_HEADERS([fcntl.h], ,AC_MSG_ERROR([Cannot find fcntl.h.]))
AC_CHECK_HEADERS([math.h], ,AC_MSG_ERROR([Cannot find math.h.]))
# The following headers are needed for code produced by lexx and yacc
AC_CHECK_HEADERS([inttypes.h], ,AC_MSG_ERROR([Cannot find inttypes.h.]))
AC_CHECK_HEADERS([libintl.h], ,AC_MSG_ERROR([Cannot find libintl.h.]))
AC_CHECK_HEADERS([malloc.h], ,AC_MSG_ERROR([Cannot find malloc.h.]))
AC_CHECK_HEADERS([unistd.h], ,AC_MSG_ERROR([Cannot find unistd.h.]))
AC_CHECK_HEADERS([stddef.h], ,AC_MSG_ERROR([Cannot find stddef.h.]))
# Non-standard includes
AC_CHECK_HEADERS([gsl/gsl_rng.h gsl/gsl_randist.h gsl/gsl_math.h \
gsl/gsl_errno.h gsl/gsl_integration.h gsl/gsl_roots.h \
gsl/gsl_sf.h gsl/gsl_sf_debye.h],,
AC_MSG_ERROR([Cannot find required gsl headers.]))
# C includes of our own
AC_CHECK_HEADERS([kww.h],, AC_MSG_ERROR([Cannot find kww.h.]))
AC_CHECK_HEADERS([cerf.h],, AC_MSG_ERROR([Cannot find cerf.h.]))
# C++ includes of our own
AC_TRY_CPP([#include<boost/format.hpp>],,
AC_MSG_ERROR([Cannot find boost/format.hpp.]))
AC_TRY_CPP([#include<boost/shared_ptr.hpp>],,
AC_MSG_ERROR([Cannot find boost/shared_ptr.hpp.]))
AC_TRY_CPP([#include<trivia/file_ops.hpp>],,
AC_MSG_ERROR([Cannot find trivia/file_ops.hpp.]))
AC_TRY_CPP([#include<readplus/readln.hpp>],,
AC_MSG_ERROR([Cannot find readplus/readln.hpp.]))
AC_TRY_CPP([#include<yamlfreeze/yaml.hpp>],,
AC_MSG_ERROR([Cannot find yamlfreeze/yaml.hpp.]))
################################################################################
## Check for libraries, needed at link time ##
################################################################################
# From standard packages
AC_SEARCH_LIBS([cos], [m],, [AC_MSG_ERROR(libm not found or corrupted)])
AC_SEARCH_LIBS([cblas_dgemm], [gslcblas],,
[AC_MSG_ERROR(libgslcblas not found or out of sync)])
AC_SEARCH_LIBS([gsl_blas_dgemm], [gsl],,
[AC_MSG_ERROR(libgsl not found or out of sync)])
AC_SEARCH_LIBS([fftw_plan_r2r_1d], [fftw3],,
[AC_MSG_ERROR(libfftw not found or out of sync)])
AC_SEARCH_LIBS([rl_completion_matches], [readline],,
[AC_MSG_ERROR(libreadline not found or corrupted)])
# From our own C packages
AC_CHECK_LIB([lmfit], [lmmin], ,
[AC_MSG_ERROR(liblmfit not found or out of sync)])
AC_CHECK_LIB([kww], [kwwp], ,
[AC_MSG_ERROR(libkww not found or out of sync)])
AC_CHECK_LIB([cerf], [voigt], ,
[AC_MSG_ERROR(libcerf not found or out of sync)])
# From our own C++ packages (why checking for main? how else?)
AC_CHECK_LIB([trivia], [main], ,
[AC_MSG_ERROR(libtrivia not found or out of sync)])
AC_CHECK_LIB([readplus], [main],,
[AC_MSG_ERROR(library readplus not found)])
AC_CHECK_LIB([yamlfreeze], [main],,
[AC_MSG_ERROR(library yamlfreeze not found)])
################################################################################
## Check for facilities needed at run time ##
################################################################################
AC_CHECK_PROGS([GNUPLOT], [gnuplot], [:])
if test "$GNUPLOT" = :; then
AC_MSG_ERROR([Cannot find gnuplot. Aborting.])
fi
################################################################################
## make Makefiles ##
################################################################################
AC_CONFIG_FILES([Makefile src/Makefile man/Makefile share/Makefile])
AC_OUTPUT
Could the problem be due to the presence of files that do not belong in a tarball? Here the contents of the tgz archive, except for the source files themselves:
frida2.1.8c/aclocal.m4
frida2.1.8c/build-aux/
frida2.1.8c/build-aux/depcomp
frida2.1.8c/build-aux/ylwrap
frida2.1.8c/build-aux/ltmain.sh
frida2.1.8c/build-aux/missing
frida2.1.8c/build-aux/install-sh
frida2.1.8c/build-aux/config.guess
frida2.1.8c/build-aux/config.sub
frida2.1.8c/CHANGELOG
frida2.1.8c/config.h.in
frida2.1.8c/configure
frida2.1.8c/configure.ac
frida2.1.8c/COPYING
frida2.1.8c/INSTALL
frida2.1.8c/m4/
frida2.1.8c/m4/libtool.m4
frida2.1.8c/m4/m4_ax_boost_regex.m4
frida2.1.8c/m4/ltversion.m4
frida2.1.8c/m4/ltoptions.m4
frida2.1.8c/m4/lt~obsolete.m4
frida2.1.8c/m4/m4_ax_boost_base.m4
frida2.1.8c/m4/ax_cxx_compile_stdcxx_11.m4
frida2.1.8c/m4/ltsugar.m4
frida2.1.8c/Makefile.am
frida2.1.8c/Makefile.in
frida2.1.8c/man/
...
frida2.1.8c/share/
...
frida2.1.8c/src/
...
frida2.1.8c/test/
...
Answer
Ok, so I had this exact problem too, it was driving me nuts. The issue seems to be that I was doing an svn export of the source code to make my src tarball.
Now that is fine, but correct me if I am wrong, but I think svn only commits files that have been modified, not touched. This means that when you come to check out the files using an svn export the files can be chronologically in the wrong order, even though files like:
configure.ac aclocal.m4 configure Makefile.am Makefile.in
Have all been touched recently, svn's date stamps wont be updated; so when you come to release the source package, the dates are in the wrong order and the configure stage will fail (not too obviously).
After doing svn export, remember always to touch these files before the src is bundled into a tar.gz using the commands:
svn export <repo address>
cd <repo name>
touch configure.ac aclocal.m4 configure Makefile.am Makefile.in
Otherwise, the user will be forced to either touch the files themselves:
touch configure.ac aclocal.m4 configure Makefile.am Makefile.in
./configure
make
or run:
autoreconf
./configure
automake
make
which will retouch these files, instead of just:
./configure
make
Which is much easier for the end user.