Re: New auxiliary archive script

[Peter Rosin]

Hi Ralf,

and thank you right back for _your_ work on this! :-)

Den 2010-08-04 22:18 skrev Ralf Wildenhues:
> thanks for your work on this!
> * Peter Rosin wrote on Wed, Aug 04, 2010 at 12:14:28PM CEST:
> > Den 2010-08-01 20:06 skrev Ralf Wildenhues:
> > > I wonder whether we can and should call it 'archive', or whether that
> > > would clash with user file names in case they don't use
> > > AC_CONFIG_AUX_DIR.  OTOH, a script named 'ar' may not be installable
> > > anywhere due to the ar binary already there.  Thoughts?
> >
> > Let's use the prefix am, and append the short form ar ->  amar. Because
> > who can resist a little bit of love in the source trees? FWIW, I came up
> > with no relevant hits for that in codesearch.
>
> I'm really not that excited about the name: It doesn't make it too clear
> what the tool does (unlike am-ar maybe) and we don't use the 'am' prefix
> anywhere else: depcomp, compile, mkinstalldirs, gendocs, gnupload.  Why
> should that be different here?  What's wrong with 'archive', is it taken
> already?  If you care about it wrapping MS lib only for now: it could
> wrap other junk as well in the future, no?
>
> Now if you insist, then maybe we can just find another compromise name.
> Please set $me to the name and use that throughout, so at least a change
> is easily done.  ;-)

I'm perfectly fine with 'archive', so let's go back to that. I found only
one problematic project on codesearch, and that was an ancient version
of gimp (0.54, 1997?):
http://archive.debian.org/debian-archive/debian/dists/Debian-1.3.1/contrib/source/gimp_0.54.1.orig.tar.gz
A quick peek at more recent versions of gimp shows that they no longer have
any clashing archive script. But I don't know when it was actually removed
from from the gimp source. It does not appear to be there in gimp 2.4 from
a couple of years ago.

Should the AM_PROG_AR macro perhaps be required when AC_PROG_RANLIB is
seen? Hmmm, but the ranlib macro is from autoconf so probably not...

> I found a few nits, but nothing substantial, so if you can format your
> next iteration as git patch, and given testing, it can just go in.

What branch should I base this on? master, maint or msvc?

> >    -h | --h*)
> >      cat<<\EOF
> > Usage: amar [--help] [--version] PROGRAM ACTION ARCHIVE [MEMBER...]
> >
> > Members can be specified on separate lines in a file named with @<file>.
>
> I think  s/can/may/  sounds better because that is not mandatory.
>
> Please use @FILE not @<file>; the capitalization is already meant to
> denote metasyntactic variables (as is done by info pages).

I wanted to, but binutils ar has this in its --help:
  @<file>      - read options from <file>
and I thought that when @FILE support for the archiver is added to
libtool it would be done similar to how it was done for the name
lister, i.e.

if test "$lt_cv_nm_interface" = "MS dumpbin"; then
  nm_file_list_spec='@'
elif $NM --help 2>/dev/null | grep 'address@hidden' >/dev/null; then
  nm_file_list_spec='@'
fi

So, I just wanted to save a future bifurcation in libtool. Should
I still change the help text to @FILE?

> > EOF
>
> > # strip leading dash in $action
> > case $action in
> >    -*) action="${action#-}" ;;
> > esac
>
> The case statement is unnecessary, you can use just
>    action=${action#-}

Right.

> > while test -n "$action"
> > do
> >    case $action in
> >      d*)
> >        delete=yes
> >        action="${action#d}"
> >        ;;
> >      x*)
> >        extract=yes
> >        action="${action#x}"
> >        ;;
> >      t*)
> >        list=yes
> >        action="${action#t}"
> >        ;;
> >      r*)
> >        replace=yes
> >        action="${action#r}"
> >        ;;
> >      c*)
> >        create=yes
> >        action="${action#c}"
> >        ;;
> >      u*)
> >        # TODO: don't ignore update modifier
> >        action="${action#u}"
> >        ;;
>
> You can factor out all the action=... lines to after the case statement
> as
>    action=${action#?}

Indeed.

> >      ?*)
> >        echo "amar: unknown action specified" 1>&2
> >        exit 1
> >        ;;
> >    esac
> > done
>
> > if test -n "$delete"; then
> >    if test ! -f "$orig_archive"; then
>
> Can it be a symbolic link or another non-regular file type?
> Won't $AR detect a non-existing file?

I doubt that lib will work with anything but regular files.
What I'm trying to accomplish is that lib will recreate the
archive every time if you do
  lib -out:foo.lib foo.obj
  lib -out:foo.lib bar.obj
  lib -out:foo.lib baz.obj
so that you end up with a foo.lib that only contains baz.obj.
You have to do this:
  lib -out:foo.lib foo.obj
  lib -out:foo.lib foo.lib bar.obj
  lib -out:foo.lib foo.lib baz.obj
to create an archive that contains all three objects (unless
you create it all in one go of course). If foo.lib doesn't
exist, lib falls over if you use the latter syntax. So, you
have to know in advance if you are adding to an existing
archive or if you are creating a new one.

Per POSIX, ar should print a diagnostic if it has to create
an archive and the c modifier isn't specified, so I just hooked
that into the existing test.

> >      echo "amar: archive not found" 1>&2
> >      exit 1
> >    fi
> >    for member
> >    do
> >      case $1 in
> >        @*)
> >          # When interpreting the content of the @file, do NOT
> >          # use func_file_conv, since the user would need to
> >          # supply preconverted file names to binutils ar, at
> >          # least for MinGW.
> >          cat "address@hidden" | while read file
> >          do
> >            $AR -NOLOGO -REMOVE:"$file" "$archive"
> >          done
>
> You can omit the cat and pipe and just
>            done<  "address@hidden"
>
> OTOH, it is common (at least in newer GCC and binutils) to allow @file
> contents to be just whitespace-separated, including single or double
> quoting.  That could be done with something like
>    atfile_contents=`cat "address@hidden"`
>    eval set x "$atfile_contents"
>    shift
>
> which you might want to factor out into a function to reuse below.
> I have no idea whether native w32 tools do something similar, though.

You are correct, lib supports @FILEs with content like:
"some object.obj" "other object.obj"
foo.obj
bar.obj baz.obj

So, I'm going with the function approach.

> Shouldn't the script stop and fail if one of the $AR commands fails?

Probably, will fix.

> >          ;;
> >        *)
> >          func_file_conv "$1"
> >          $AR -NOLOGO -REMOVE:"$file" "$archive"
> >          ;;
> >      esac
> >    done
> >
> > elif test -n "$extract"; then
> >    if test ! -f "$orig_archive"; then
> >      echo "amar: archive not found" 1>&2
> >      exit 1
>
> Repeated code, could be in an 'error' function.

Yes, *fixing*.

> >    fi
> >    if test $# -gt 0; then
> >      for member
> >      do
> >        case $1 in
> >          @*)
> >            cat "address@hidden" | while read file
> >            do
> >              $AR -NOLOGO -EXTRACT:"$file" "$archive"
> >            done
> >            ;;
> >          *)
> >            func_file_conv "$1"
> >            $AR -NOLOGO -EXTRACT:"$file" "$archive"
> >            ;;
> >        esac
> >      done
> >    else
> >      $AR -NOLOGO -LIST "$archive" | while read member
> >      do
> >        func_file_conv "$member"
> >        $AR -NOLOGO -EXTRACT:"$file" "$archive"
>
> Is this file conversion necessary, or even right?
> I mean, is it really that '$AR -list' prints names that
> '$AR -extract:' cannot eat?  Or am I missing something?

Oh. Good catch! It's not right of course. *fixing*

> Does lib have no way to just extract the whole archive?

No. Come on, what do you expect? :-)

> >      done
> >    fi
> >
> > elif test -n "$replace"; then
> >    if test ! -f "$orig_archive"; then
> >      if test -z "$create"; then
> >        echo "amar: creating $orig_archive"
> >      fi
> >      orig_archive=
> >    else
> >      orig_archive=$archive
> >    fi
> >
> >    for member
> >    do
> >      case $1 in
> >      @*)
> >        func_file_conv "address@hidden"
> >        set x "$@" "@$file"
> >        shift
> >        ;;
> >      *)
> >        func_file_conv "$1"
> >        set x "$@" "$file"
> >        shift
>
> The two shifts can be factored outside the case.

*fixing*

> >        ;;
> >      esac
> >      shift
> >    done
> >
> >    if test -n "$orig_archive"; then
> >      $AR -NOLOGO -OUT:"$archive" "$orig_archive" "$@"
> >    else
> >      $AR -NOLOGO -OUT:"$archive" "$@"
> >    fi
> >
> > elif test -n "$list"; then
> >    if test ! -f "$orig_archive"; then
> >      echo "amar: archive not found" 1>&2
> >      exit 1
> >    fi
> >    $AR -NOLOGO -LIST "$archive"
> > fi

Cheers,
Peter