Re: libtool 2.2.2 cross-compile to mingw problems

[Charles Wilson]

Simon Josefsson said:
> address@hidden:~/gnutls4win/build/gnutls-2.3.7/src$ ./certtool.exe
> Wine failed with return code 127
> address@hidden:~/gnutls4win/build/gnutls-2.3.7/src$ 
>
> I need to run them in-place, because they are invoked from 'make check'.
>
> Re-compiling the wrapper with DEBUGWRAPPER yields:

> (main) libtool shell wrapper name: certtool_ltshwrapperTMP
> (make_executable)   : 
> Z:/home/jas/gnutls4win/build/gnutls-2.3.7/src/./.libs/certtool_ltshwrapperTMP
> (main) newargz[0]   : .exe

well, here's your problem.

> As far as I can understand from the .libs/lt-certtool.c code, it writes
> a bash script into .libs/certtool_ltshwrapperTMP and then tries to
> execute that.
>
> How is this intended to work?  You can't execute bash scripts under
> Windows nor Wine, as far as I know.

Yes and no. You need some win32 shell installed in your Wine 
environment. (For native win32 builds, this is no problem: if you are 
running libtool, you have a shell).  Long-term, it would be great to use 
the wrapper executable ONLY, without a script, to execute the real 
target exe -- at least for win32 hosts.  (we'd still need the script for 
one purpose, even then; libtool "reads" the script by sourcing it back 
it.)  However, we don't that yet -- to do so, we'd need to create an env 
structure, copy the existing env, then manipulate the LT_LIBRARY_PATH 
and PATH (and, and, etc) entries...that's probably a libtool-2.4 idea.

Anyway, the following bits in ltmain kick in, which try to determine the 
name of the shell that the (win32 wrapper executable) should invoke, as if:
   system("some-win32-sh.exe the-wrapper-script <other args?>")


case "$host" in
  *mingw* )
    # awkward: cmd appends spaces to result
    lt_sed_strip_trailing_spaces="s/[ ]*\$//"
    lt_newargv0=`( cmd //c echo $SHELL | $SED -e 
"$lt_sed_strip_trailing_spaces" ) 2>/dev/null || echo $SHELL`
    case $lt_newargv0 in
      *.exe | *.EXE) ;;
      *) lt_newargv0=$lt_newargv0.exe ;;
    esac
    ;;
  * ) lt_newargv0=$SHELL ;;
esac

However, I see the problem:

(1) the construct `(cmd //c echo $SHELL ...)` assumes that the shell 
used by the backticks is the MSYS shell -- which is true for native 
mingw builds only -- because it is using the doubled '/' to bypass MSYS 
path-translation logic (/c is a switch, not a posix path that needs to 
be win32-ified).
(2) in your case, the shell used by the backticks is your linux shell. 
Even if you could invoke the Wine 'cmd.exe' in this way, your shell will 
pass '//c' literally, unlike MSYS shell which says "Oh, this is not a 
path, so I'll replace '//' with '/' and move on". In your case, the Wine 
cmd.exe will complain about the illegal argument '//c'
(3) the "|| echo $SHELL" part isn't working, because the return status 
of a subshell is the status of the LAST command -- and $SED is perfectly 
happy.
(4) end result: you have lt_newargv0=<EMPTY>, which then gets ".exe" 
appended.

Boom.

(5) Also, there is a hidden assumption that the win32 shell that should 
be used by Wine has the same name as the host shell.

Now, I can probably address some of these isses -- but the real solution 
is actually found in the enclosing block of code:

if test -n "$TARGETSHELL" ; then
  # no path translation at all
  lt_newargv0=$TARGETSHELL
else
 <the stuff above>
fi

If, when compiling your code, you have TARGETSHELL defined in your 
envinrment to the win32 (e.g. internal Wine) path, then the wrapper 
executable will use that. That is,

export TARGETSHELL="C:\\Util\\gnuwin\\bin\\sh.exe"
export TARGETSHELL="C:\\msys\\1.0\\bin\\sh.exe"

*before* building your project.

Assuming that setting TARGETSHELL solves your problem, I think something 
like the following (untested) patch might be useful. The second hunk 
tries to address points (1), (2) and (3) above. This still won't fix 
your problem -- but TARGETSHELL should -- because you'd really just end 
up trying to exec the linux shell from the (win32-native, Wine) wrapper 
exe: I don't know if Wine lets you do that.

--
Chuck


--- ltmain.m4sh.orig    2008-04-19 12:30:30.125000000 -0400
+++ ltmain.m4sh 2008-04-19 13:00:08.750000000 -0400
@@ -2683,9 +2683,20 @@
            else
              case "$host" in
                *mingw* )
+                  case "$build" in
+                   *mingw* ) ;;
+                   *cygwin* ) ;; # don't warn if doing cygwin->mingw cross
+                   * ) 
+                     func_warning "Cross-compile for mingw detected, but 
\$TARGETSHELL is not set."
+                     func_warning "Wrapper executables may not function 
correctly unless \$TARGETSHELL"
+                     func_warning "is set to the mingw emulation environment 
path to a posix shell"
+                     func_warning "while building this executable."
+                     ;;
+                 esac 
                  # awkward: cmd appends spaces to result
                  lt_sed_strip_trailing_spaces="s/[ ]*\$//"
-                 lt_newargv0=`( cmd //c echo $SHELL | $SED -e 
"$lt_sed_strip_trailing_spaces" ) 2>/dev/null || echo $SHELL`
+                 lt_newargv0=`( cmd //c echo $SHELL ) 2>/dev/null || ( cmd /c 
echo $SHELL ) 2>/dev/null || $ECHO $SHELL`
+                 lt_newargv0=`$ECHO $lt_newargv0 | $SED -e 
"$lt_sed_strip_trailing_spaces"`
                  case $lt_newargv0 in
                    *.exe | *.EXE) ;;
                    *) lt_newargv0=$lt_newargv0.exe ;;