Updated: libtool-devel-1.5, libltdl3-1

Charles Wilson cwilson@ece.gatech.edu
Wed Apr 16 22:41:00 GMT 2003

libtool-devel has been updated to the 1.5 release, as well as libltdl3.

At long last, the super spiffy auto-import-based, mostly "just like 
unix" DLL support for cygwin is available in the official, stable, 
libtool release.  Now is the time to start bugging your favorite 
upstream maintainers to update their packages to the latest autotools -- 
and in most cases, they'll get cygwin and mingw DLLs for free.

Note that libtool-devel-1.5 depends on autoconf-devel-2.57 or newer, and 
automake-devel-1.7.3 or newer.

Changes from 20020216-1

o updated to the official 1.5 release



To update your installation, click on the "Install Cygwin now" link
on the http://cygwin.com/ web page.  This downloads setup.exe to
your system.  Save it and run setup, answer the questions and pick
up 'libtool-devel' from the 'Devel' category.

Note that downloads from sources.redhat.com (aka cygwin.com) aren't
allowed due to bandwidth limitations.  This means that you will need
to find a mirror which has this update.

In the US, ftp://mirrors.rcn.net/mirrors/sources.redhat.com/cygwin/
is a reliable high bandwidth connection.

In Japan, ftp://ftp.u-aizu.ac.jp/pub/gnu/gnu-win32/ is already

In DK, http://mirrors.sunsite.dk/cygwin/ is usually up-to-date.

If one of the above doesn't have the latest version of this package
you can either wait for the site to be updated or find another

Please  send questions or comments to the Cygwin mailing list at:
cygwin@sources.redhat.com .  If you want to subscribe go to:
http://cygwin.com/lists.html I would appreciate if you would use
this mailing list rather than emailing me directly.  This includes
ideas and comments about the setup utility or Cygwin in general.

If you want to make a point or ask a question the Cygwin mailing
list is the appropriate place.


To unsubscribe to the cygwin-announce mailing list, look at the
"List-Unsubscribe: " tag in the email header of this message. Send
email to the address specified there.  It will be in the format:




1) Building an executable,
that links against a shared library that is part of the same package, 
where the source file(s) are in a different subdir than the makefile 
that builds the executable, doesn't work. (yeah, real "common" case. 
Known to affect expat)  Example:

<top-sourcedir>/sharedlib/<library files, own Makefile.in>

<top-sourcedir>/Makefile.in contents:
default:  buildlib foo/foo

buildlib: $(LIBRARY)

all: $(LIBRARY) foo/foo

For more info, see

2) Open issue: pseudo-cross compiler setup doesn't work (e.g. setting 
CC='gcc -mno-cygwin' and expecting a mingw output doesn't work, although 
it's unclear whether it SHOULD work, even if at some point in the past 
it did.)  The binary wrapper program hardcodes "/bin/sh" -- which is not 
a valid pathname for a -mno-cygwin program.  It'd be best to compile the 
wrapper with plain "gcc" -- But without making a distincution between 
the buildplatform compiler and the hostplatform compiler (e.g. without 
using --build= --host=), we can't know about the need for this 
distinction. IMO, fixing the "relink .exe ad nauseum" problem is more 
important than this corner case -- and there are other ways of 
addressing this corner case, which do not involve changes to libtool.


o adds partial support for including .rc files and using windres
   to add resources to executables and libraries.  You still must
   add something like the following to your Makefile.am file:

   LTRCCOMPILE = $(LIBTOOL) --mode=compile $(RC) \
      `echo $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) |\
          sed -e 's/-I/--include-dir /g;s/-D/--define /g'`

   and then explicitly include a build rule:

   %.o : %.rc
         $(LTRCCOMPILE) -i $@ -o @^

o -export-symbols <a def file> now works properly (it's been broken
   for over two years...)

o DESIGN DECISION: (this is a a change from pre-20021111 behavior)
     libtool will refuse to create a shared library if any of its
     dependencies are available only as static archives.  DLLs may
     only depend on other DLLs (*)

(*) This is a good idea.  But, we need workarounds for the standard 
runtime libs like libgcc.a, libstdc++.a, etc.  These workarounds are 
implemented in this libtool release.

o correctly identifies import libraries as "shared objects", so that
   libtool's new "DLLs may only depend on other DLLs" behavior isn't
   triggered with fatal results when a target DLL's dependencies are
   satisfied by import libs.

o Skips the "DLLs may only depend on other DLLs" check when the
   dependency is one of the "standard" runtime libs which are currently
   available only in static form (libgcc.a, libstdc++.a, etc) on

o Isn't "confused" by the libtool .la files supplied with cygwin's gcc,
   even though they only specify static archives  (e.g. libstdc++.la
   lists "libstdc++.a" but not "cygstdc++.dll" [which is good because
   cygstdc++.dll doesn't exist]).

o No longer records compiler builtin library paths or compiler-generated
   deplibs (like -luser32 -lgcc) in the "dependency_libs" variable in
   generated .la files.  (**)

(**) e.g. if cygwin's gcc-3.2 had been built with this version of 
libtool, then /usr/lib/libstdc++.la wouldn't be so "weird" -- even 
though cgf removed it entirely from the more recent releases of gcc-3.2 
for cygwin. However, if you've got an old version of cygwin-gcc-3.2, 
this libtool won't croak.


Three selftest failures (two longstanding, not regressions; one recent 
due to changes [bug?] in cygwin1.dll)

It seems there is a race condition with regards to 'ls -l' computing the 
total size of all files in a directory.  This cause the "before" 
snapshot to be incorrect -- while the "after" snapshot is correct.  And 
during a dryrun test, "before" should equal "after" -- and they don't. 
but not because the file-on-disk are different; that "before" snapshot 
of those files-on-disk is simply wrong.  I'm pretty sure this isn't a 
libtool bug...and it won't affect actual libtool --mode=dry-run operation.


I fixed one bug, but another showed up: $PATH doesn't
get set properly when running this test...This test
used to get skipped on cygwin, but no longer?

compile mode seems okay
install mode seems okay
link mode *always* fails -- like this:
   "failed: mkdir .libs
    gcc -o hell.exe -g -O -Wl,-someflag=test foo.o"
?? *mkdir* fails??
But it works fine in compile mode---
   "passed: mkdir .libs
    gcc -c "-DVAR= test    " foo.c  -DPIC -o .libs/foo.o
    gcc -c "-DVAR= test    " foo.c -o foo.o >/dev/null 2>&1"

More information about the Cygwin-announce mailing list