* m4/libtool.m4, m4/ltdl.m4, m4/ltoptions.m4: Added commented out
empty macro definitions to match each AU_DEFUN to help aclocal
from automake-1.4 regexps find the right libtool macro files.
itself, does not link against an uninstalled library. It appears
that this can happen if you are building a project where:
(1) there is a library
(2) there is an executable that does not, itself, link against
that library
In this case, notinst_deplibs is empty, and when the wrapper
script is loaded by libtool during ./libtool mode=install, there
is a check for "is notinst_deplibs empty". In this case, it is,
so libtool reports an error.
I don't think it is wrong to have a wrapper script for an exe that
*technically* doesn't need it. But instead of checking for
notinst_deplib nonempty as a proxy for "did wrapper script get
sourced into my environment properly" a different var should be
used.
* config/ltmain.m4sh (func_mode_install): set new variable
`generated_by_libtool_version' in wrapper script when using
libtool_install_magic. When verifying that wrapper script was
properly sourced, check that `generated_by_libtool_version' is
non-empty, instead of using notinst_deplibs.
ensure that import libraries don't get stripped.
* m4/libtool.m4 (_LT_SYS_DYNAMIC_LINKER): On Windows based hosts,
use postinstall_cmds to strip DLL (not import lib!) when symbol
stripped installation was requested.
pass. When the output wrapper script is not in the current
directory, we can not use an rpath which is relative to the
current directory. Use an absolute path instead.
to escape $ properly in combination with double_quote_subst.
(func_quote_for_eval): Four space indent margin for functions!
(func_quote_for_expand): Use $sed_double_backslash.
(func_show_eval): New function that echos an escaped but variable
expanded command, and then evaluates it.
from the match character set.
(double_quote_subst): Ditto.
* lib/general.m4sh (sed_quote_subst): Ditto.
(double_quote_subst): New variable, copied from
m4/libtool.m4. Much the same as sed_quote_subst, but variable
references are not quoted.
(func_quote_for_eval): Given any $foo; $bar and
$foo must be identical after the following commands:
func_quote_for_eval "$foo"
eval "bar=$func_quote_for_eval_result"
(func_quote_for_expand): In a similar vein to to func_quote_for_eval,
except that variable expansions are not protected when quoting the
command for subsequent display.
* config/ltmain.m4sh: Factor away use of $sed_quote_subst now that
it has been subsumed into func_quote_for_eval.
(func_mode_link): Instead of `2004-10-14 Peter O'Gorman' below,
use func_quote_for_expand so that ${wl} is expanded properly in
the $show expression.
subdir-objects builds. This marks the beginning of a move to an
Autotest driven test framework, which will eventually allow us to
trim away many of the generated files in the current test demo
subdirectories:
* configure.ac (AC_CONFIG_TESTDIR): Re-enable commented out
Autotest macros.
(AUTOTEST, M4SH): Use --language long hand.
* bootstrap: Generate tests/testsuite, and by implication
tests/package.m4.
* Makefile.am (edit): Add substitutions needed to generate
tests/package.m4 from bootstrap.
(check-recursive): Add tests/testsuite.
(TESTSUITE): The new Autotest testsuite target.
(TESTSUITE_AT): The Autotest sources.
(EXTRA_DIST): Distribute the generated testsuite and sources.
(tests/package.m4): Use $(edit) to propagate package metadata to
the new tests.
(tests/testsuite): Build the testsuite.
(tests/atconfig): How to regenerate the file.
(check-local): Hook Autotest into `make check'.
(installcheck-local): Run the testsuite on the installed tree.
(clean-local): Tidy our file droppings.
* tests/Makefile.am: Remove bitrotted code from Autotest
experiments in Autoconf 2.52 era. Updated the rest and moved into
Makefile.am.
* tests/testsuite.at: New container for Autotest test groups.
* tests/am-subdir.at: New test group for compatibility with
Automake's subdir-objects builds. Currently holds two tests, one
for a C project and another similar project in C++.
* TODO: Add an entry to remind us that these new tests expose a
bug in libtool.
* NEWS: Updated.
in dllsearchpath when creating wrapper script. Fixes
depdemo-relink test failure.
* config/ltmain.m4sh (func_mode_link): don't relink
on cygwin/mingw; no need. But do ensure that wrappers
are created unless doing a purely static build.
gary@gnu.org--2004/cvs-utils--tla--1.0:
* commit, config/mailnotify: Removed. Now built from...
* clcommit.m4sh, config/mailnotify.m4sh: ...these new sources.
* Makefile.maint (commit, config/mailnotify): How to build them.
* bootstrap: If we have a checked out tree (i.e. clcommit.m4sh
is present), then build them.
and thus prevented it from finding itself:
* config/general.m4sh (progname): Don't actually change the name
of the script if it begins with a `-', just prepend `./'.
(progpath): Add a new clause that first searches the execution
PATH, and then finally assumes this script is in the current
working directory if $progname has no directory separators (e.g
when invoked as `$SHELL progname').
* config/ltmain.m4sh: Windows and Cygwin reserve the .lnk
extension to represent shortcuts. Use the extension ".lnkscript"
rather than ".lnk" for the piecewise linking ld script in order to
avoid the conflict.
around the deficiencies in AC_PROG_CXX and AC_PROG_F77.
* configure.ac: Cleaned up, having moved the checks for compilers
which do not exist to libtool.m4.
with $lt_ECHO, or we always get a libtool with ECHO=echo, rather
than testing for a mechanism that quotes properly.
* NEWS: Updated.
* config/ltmain.m4sh: Need to include getopt.m4sh earlier so that
progpath is set before it is called to --no-reexec.
* NEWS: Updated.
Without this change, the distributed libtool tarball has a
dependency on autom4te. There was some inconsistency in the
filenames: now we have foo.m4sh -> m4sh -> foo.in -> configure
-> foo[.sh]. Bootstrap runs the m4sh steps and the resulting .in
files are distributed:
* tests/Makefile.am (EXTRA_DIST): Distribute defs.m4sh.
* Makefile.am (EXTRA_DIST): Distribute libtoolize.m4sh.
(libtoolize.in, libtoolize): Split into bootstrap and runtime
stages.
($(top_srcdir)/tests/defs.in, $(top_srcdir)/tests/defs): Ditto.
($(top_srcdir)/config/ltmain.sh): Swap ltmain.m4sh and ltmain.in
names.
* bootstrap: Make libtoolize.in and tests/defs.in.
* NEWS: Updated.
libdlloader_la_OBJECTS, an undocumented Automake internal
variable...
(libltdlc_la_SOURCES): ...use libdlloader_la_SOURCES here
instead.
Reported by Bob Friesenhahn <bfriesen@simple.dallas.tx.us>
for mingw32.
* libltdl/lt__dirent.h: Ditto.
* libltdl/argz._h: Ditto. Be careful about not requiring
additional libltdl files when used outside of libltdl.
* libltdl/slist.h: Ditto.
* HACKING: Add a section on libltdl implementation layering.
* m4/libtool.m4: Correct obvious spelling errors in remarks.
(_LT_ENABLE_LOCK): Make s390x-ibm-tpf hosts adopt elf64_s390
conventions.
(_LT_SYS_DYNAMIC_LINKER): Set defaults for s390x-ibm-tpf host.
(_LT_CHECK_MAGIC_METHOD): Skip all dependency checks for
s390x-ibm-tpf hosts.
(_LT_LINKER_SHLIBS): Make s390x-ibm-tpf hosts adopt GNU/Linux OS
attributes.
* NEWS: Updated.
workstation doesn't hardcode the shared library path, although it
used to, so this is more likely a function of the binutils and/or
gcc I am using at the moment. The hardcode value setting code for
the configure script sucks at the moment though...
* (TODO): ...and needs to do a test compile instead of a table
lookup before 2.0 final.
libtool library on the command line using -lfoo we need to find
the library in the search paths and add it's dependency_libs to
the link in the conv pass so that any libtool libs listed are
correctly expanded in the link pass.
ltmain.in.
(ECHO): Audit for (and correct!) uses of $ECHO with more than one
argument, which breaks when ECHO='printf %sn', and with possible
`-' as first character of the argument, which breaks `print -r'
and some implementations of `echo'.
* config/ltmain.in, tests/defs.in, libtoolize.in (ECHO): Ditto.
Reported by Albert Chin-A-Young <china@thewrittenword.com>
for a different host, it can't be shipped. That also means that
defs will always be generated in the build tree, so each test case
can have the srcdir discovery factored away into tests/defs.in:
* Makefile.am (EXTRA_DIST): s/defs/defs.in/g
* tests/Makefile.am (defs): New rule to remake defs from the
parent directory's Makefile.
* tests/cdemo-conf.test, tests/cdemo-exec.test,
tests/cdemo-make.test, tests/cdemo-shared.test,
tests/cdemo-static.test, tests/demo-conf.test,
tests/demo-deplibs.test, tests/demo-exec.test,
tests/demo-hardcode.test, tests/demo-inst.test,
tests/demo-make.test, tests/demo-nofast.test,
tests/demo-noinst-link.test, tests/demo-nopic.test,
tests/demo-pic.test, tests/demo-relink.test,
tests/demo-shared.test, tests/demo-static.test,
tests/demo-unst.test, tests/depdemo-conf.test,
tests/depdemo-exec.test, tests/depdemo-inst.test,
tests/depdemo-make.test, tests/depdemo-nofast.test,
tests/depdemo-relink.test, tests/depdemo-shared.test,
tests/depdemo-static.test, tests/depdemo-unst.test,
tests/f77demo-conf.test, tests/f77demo-exec.test,
tests/f77demo-make.test, tests/f77demo-shared.test,
tests/f77demo-static.test, tests/link-2.test, tests/link.test,
tests/mdemo-conf.test, tests/mdemo-dryrun.test,
tests/mdemo-exec.test, tests/mdemo-inst.test,
tests/mdemo-make.test, tests/mdemo-shared.test,
tests/mdemo-static.test, tests/mdemo-unst.test,
tests/mdemo2-conf.test, tests/mdemo2-exec.test,
tests/mdemo2-make.test, tests/nomode.test, tests/pdemo-conf.test,
tests/pdemo-exec.test, tests/pdemo-inst.test,
tests/pdemo-make.test, tests/quote.test, tests/sh.test,
tests/suffix.test, tests/tagdemo-conf.test,
tests/tagdemo-exec.test, tests/tagdemo-make.test,
tests/tagdemo-shared.test, tests/tagdemo-static.test,
tests/tagtrace.test: Removed srcdir discovery boilerplate...
* tests/defs.in: ...and put it here.
* config/general.m4sh, config/getopt.m4sh: New files with factored
out functionality from...
* config/ltmain.in, libtoolize.in: ...here. m4_include the shared
code, and call AS_SHELL_SANITIZE. Also quote all `[' and `]' to
survive m4 run.
(func_serial): No need to double up m4sh quotes after a `#'.
(func_check_macros): Escape `dnl' to stop m4sh interpreting it as
a macro.
* tests/defs.in: New file expanded into tests/defs by autom4te.
* m4/libtool.m4 (_LT_DECL_EGREP): Also require fgrep for the tests
that use it.
* Makefile.am ($(top_srcdir)/config/ltmain.sh): Now built with
M4SH.
(libtoolize): Ditto.
* bootstrap: Set M4SH for call to make.
* HACKING: No need to sync to AS_SHELL_SANITIZE by hand any more.
* config/ltmain.in: Use it throughout.
(func_mkdir_p): Now matches the other implementations in
tests/defs and libtoolize.in.
* tests/sh.test: Updated to detect ECHO misuse.
$deplibs list. This causes problems when the link pass tries to
find libraries when they are located in non-standard places
denoted by -L options. Due to the reversed order these -L options
occur after the libraries that need them, and they are not found:
* config/ltmain.in: (Un)Reverse $deplibs list at the start of the
link pass in lib mode.
a .exp file for an executable, also create a .def file, and add
that file to the link command via SYMFILES. This causes binutils
to generate an .edata section, and allows self dlopening to work.
* NEWS: Updated.
_preloaded_symbols[] cannot go into .rdata section if symbols are
DATA imported from DLL, on windows, because runtime relocations
must happen.
* m4/libtool.m4 (_LT_LINKER_SHLIBS([TAGNAME])): Build exports for
symbols in .rdata sections.
`foo=bar break', but unfortunately they also choke on `foo=bar;
break' and `foo=bar && break'. Writing a sophisticated test to
catch just the intended case seems like more trouble than it's
worth, but leaving the test in causes the testsuite to fail on
valid m4sh output:
* tests/assign.test: Removed; chokes on perfectly valid shell
syntax.
* tests/Makefile.am (COMMON_TESTS): Remove assign.test.
* tests/defs.in (scripts): Don't check the m4sh inputs, go back
to checking the generated ltmain.sh script.
patch; fix them carefully. Introduce an opt_dry_run to ltmain.in
so that the implementations of func_mkdir_p can converge, and a
func_mktempdir to do a better job of temporary directory creation:
* libtoolize.in (func_mkdir_p): Don't fail if the directory wasn't
created in dry run mode.
* tests/defs (func_mkdir_p): Ditto. We don't actually have a dry
run mode for the tests, but the function is written carefully to
be kept in synch and work correctly here too.
* config/ltmain.in (func_mkdir_p): Ditto. This copy of the
function now only differs in its use of $echo over $ECHO.
(func_extract_archive): Removed first redundant mkdir call.
(func_mktempdir): New function that tries to avoid races when
making temporary directories.
(opt_dry_run): Set this if --dry-run is given at the CLI, or if
tests/mdemo-dryrun.test has forced the value of $run.
(func_mode_install): Call $MKDIR directly and error out if the
directory cannot be created.
(func_mode_link): Rather than copying func_mkdir_p into the
wrapper script as a replacement for $MKDIR_P, we know that the
script won't be called my `make -j', so write the current value of
$MKDIR.
below is horrendously broken since it makes the installed libtool
try to run the mkinstalldirs or install-sh helper scripts. This
patch fixes the `make -j' mkdir race condition internally:
* libtoolize.in (func_mkdir_p): New slice and dice algorithm
to build the directory tree one dir at a time, ignoring errors
until the end incase they are transient due to a concurrent
identical mkdir.
* tests/defs (func_mkdir_p): Ditto.
* config/ltmain.in (func_mkdir_p): Ditto, except that ltmain.in
uses $echo rather than $ECHO, and uses $show and $run to interact
correctly with the command line.
* configure.ac (AM_INIT_AUTOMAKE): 1.8 is sufficient again.
* Makefile.am (edit): Don't substitute automake's $(mkdir_p).
* NEWS: Updated.
from Gary V. Vaughan) to compile under MinGW.
* config/mkstamp: Translate input data so that it is assured to
use the Unix line terminations. This is necessary under Windows if
the files are checked out with CR/NL line terminations. A
side-effect of this change is that input must always be from
stdin. Previously a filename argument was accepted as well.
avoid forking at every call. Changed all callers.
(func_mode_link): Simplified, and removed the final non-portable
nested escaped double quotes in back quotes usage in the process.
compiler flags need an argument or not, and can disect options and
pass parts through incorrectly. This changeset reverts both
2004-09-06 Albert Chin-A-Young (topmost) and 2004-09-05 Albert
Chin-A-Young below, and recognises more options explicitly:
* config/ltmain.in (func_mode_link): Pass through options needed
to compile in 64-bit mode with gcc, and the SGI, Sun, HP and IBM
compilers.
were convenience libs, so we did not test the creation of shared
libraries on any platform. Also none of the libs used any actual
fortran library calls, adding a call to write(*,*) in each lib
causes the tests to fail on darwin (and presumably other platforms).
These new tests would likely cause many more fortran test failures,
so if there is an error during make it is reported as SKIP.
* tests/f77demo-make.test: Don't fail hard, skip on failure.
* tests/f77demo/foof2.f: New file.
* tests/f77demo/foof3.f: New file.
* tests/f77demo/fprogram.f: Call routine in new lib.
* tests/f77demo/foof.f: Call routine in another new lib.
* tests/f77demo/Makefile.am: Make a couple of new libraries, add
$(FLIBS) to cprogram LDADD.
unmolested will prove to be a good plan for the 2.0 release. If
it goes awry, we can get back to where we were by reverting this
patch and the 2004-09-05 Albert Chin-A-Young patch, aka
gary@gnu.org--2004/libtool--devo--1.0--patch-169,172:
* config/ltmain.in (func_mode_link): Because we now pass through
compiler arguments we doesn't recognize, the code to pass
through GCC's -m* arguments is not needed.
