Loading ...
Sorry, an error occurred while loading the content.

Re: [hackers-il] Freecell Solver's Test Suite + Conversion to CMake

Expand Messages
  • Nadav Har'El
    ... Hi Shlomi, can you please tell us a bit about this CMake , of which I have to admit I never heard? What makes it better than autoconf (forget automake, I
    Message 1 of 5 , Aug 3, 2008
    • 0 Attachment
      On Thu, Jul 31, 2008, Shlomi Fish wrote about "[hackers-il] Freecell Solver's Test Suite + Conversion to CMake":
      > I recently resumed work on Freecell Solver (FCS) somewhat more intensively.
      > Having gotten tired of Autoconf/Automake/Libtool (a.k.a Autohell), I decided
      > to convert to CMake ( http://www.cmake.org/ ) for its configuration stage.

      Hi Shlomi, can you please tell us a bit about this "CMake", of which I have
      to admit I never heard? What makes it better than autoconf (forget automake,
      I consider it cosmetics) - at least for you? Does it have any downsides
      compared to autoconf?

      For reference, here is what I use autoconf for. In the (good?) old days,
      not all machines I had access to were PCs with Intel CPUs running Linux.
      I found myself compiling the same program I wrote on Linux, Solaris,
      DEC Alpha, AIX, Silicon Graphics, and AT&T System 5, all in the same week.
      And each of those systems had its own compiler peculiarity, different
      installed software, library differences, and most notorious - compiler bugs
      (demonstrating, by the way, the point that free software *does* have fewer
      bugs). This is where autoconf came in extremely handy. Does CMake also
      aim to help you in this scenario - of people who don't use Windows or Linux?

      Nadav.

      --
      Nadav Har'El | Monday, Aug 4 2008, 3 Av 5768
      nyh@... |-----------------------------------------
      Phone +972-523-790466, ICQ 13349191 |Those who beat their swords into
      http://nadav.harel.org.il |plowshares will plow for those who don't.
    • Shlomi Fish
      Hi Nadav! ... Well, I have mentioned it in passing in the Haifux mailing list in relevance to the Autotools presentation:
      Message 2 of 5 , Aug 4, 2008
      • 0 Attachment
        Hi Nadav!

        On Monday 04 August 2008, Nadav Har'El wrote:
        > On Thu, Jul 31, 2008, Shlomi Fish wrote about "[hackers-il] Freecell
        Solver's Test Suite + Conversion to CMake":
        > > I recently resumed work on Freecell Solver (FCS) somewhat more
        > > intensively. Having gotten tired of Autoconf/Automake/Libtool (a.k.a
        > > Autohell), I decided to convert to CMake ( http://www.cmake.org/ ) for
        > > its configuration stage.
        >
        > Hi Shlomi, can you please tell us a bit about this "CMake", of which I have
        > to admit I never heard?

        Well, I have mentioned it in passing in the Haifux mailing list in relevance
        to the Autotools presentation:

        http://hamakor.org.il/pipermail/haifux/2008-July/000356.html

        And Artium mentioned it on his blog:

        http://art-blog.no-ip.info/newpress/blog/post/146

        You can also read about it in the Wikipedia:

        http://en.wikipedia.org/wiki/CMake

        > What makes it better than autoconf (forget
        > automake, I consider it cosmetics) - at least for you? Does it have any
        > downsides compared to autoconf?

        OK, here are the advantages as I see them:

        1. Its files are written in a consistent high-level language, whose
        interpreter is written in C++. This is instead of Autoconf's
        m4-generating-Bourne-shell setup, where both m4 and /bin/sh are quirky and
        get broken very easily.

        2. Configuration and compilation are much faster than in Autoconf, yielding a
        faster development process.

        3. It has a BSD License instead of the Autotools' GPL (which may not be very
        relevant, but it's still a nice bonus).

        4. It causes much less unexpected problems and regressions than Autoconf.
        Autoconf requires constant maintenance, while CMake tends to keep very good
        backwards compatibility.

        5. It has some user-interfaces for the configuration: ccmake (Curses-based),
        cmake-gui (Qt/Win32), etc.

        6. It merges in the functionality of Automake and libtool, both in a more
        consistent and less error-prone way.

        7. CMake contains CPack, which allows to prepare source and binary
        distributions of the projects with ease.

        8. It also supports CTest which is an interface to an automated testing
        framework.

        9. Has native support for Win32.

        10. Since CMake is installed separately on the system, the source pacakges are
        smaller. I was able to reduce the size of the FCS tar.gz from over 500,000
        bytes to slightly above 200,000 bytes.

        -----------------

        As for the downsides, Artium mentioned them in his blog:

        http://art-blog.no-ip.info/newpress/blog/post/146

        1. No support for "make uninstall".

        http://www.cmake.org/Wiki/CMake_FAQ#Can_I_do_.22make_uninstall.22_with_CMake.3F

        2. No support for building both static and shared libraries.

        http://xrl.us/omn7i

        3. Poor documentation (I personally feel that the Autotools documentation is
        much worse).

        4. Inadequate Defaults (building a debug version without optimisation as
        default).

        ---------------

        Personally, I feel that CMake is still better.

        >
        > For reference, here is what I use autoconf for. In the (good?) old days,
        > not all machines I had access to were PCs with Intel CPUs running Linux.
        > I found myself compiling the same program I wrote on Linux, Solaris,
        > DEC Alpha, AIX, Silicon Graphics, and AT&T System 5, all in the same week.
        > And each of those systems had its own compiler peculiarity, different
        > installed software, library differences, and most notorious - compiler bugs
        > (demonstrating, by the way, the point that free software *does* have fewer
        > bugs). This is where autoconf came in extremely handy. Does CMake also
        > aim to help you in this scenario - of people who don't use Windows or
        > Linux?

        Yes, it does. Reading from the wikipedia:

        {{{{{{{{{{{{
        Support for cross-platform builds, and known to work on Linux, other POSIX
        systems (including AIX, *BSD systems, HP-UX, IRIX/SGI, MinGW/MSYS and
        Solaris), Mac OS X and Windows 95/98/NT/2000/XP,
        }}}}}}}}}}}}

        CMake is a GNU Autotools replacement, on everything that is implied from that.

        Regards,

        Shlomi Fish

        -----------------------------------------------------------------
        Shlomi Fish http://www.shlomifish.org/
        Rethinking CPAN - http://xrl.us/bjn7p

        I met a guy in the bar, talked to her and she gave me her phone number.
      • Nadav Har'El
        ... The whole idea of Autoconf s m4-generating-Bourne-shell setup is that the result (the bourne shell) can run on any computer without needing to install
        Message 3 of 5 , Aug 4, 2008
        • 0 Attachment
          On Mon, Aug 04, 2008, Shlomi Fish wrote about "Re: [hackers-il] Freecell Solver's Test Suite + Conversion to CMake":
          > 1. Its files are written in a consistent high-level language, whose
          > interpreter is written in C++. This is instead of Autoconf's
          > m4-generating-Bourne-shell setup, where both m4 and /bin/sh are quirky and
          > get broken very easily.

          The whole idea of Autoconf's "m4-generating-Bourne-shell setup" is that
          the result (the bourne shell) can run on any computer without needing to
          install anything. Therefore, whatever input language your autoconf alternative
          has, generating Bourne shell from it is a *good* idea, I think...

          So, if CMake doesn't generate Bourne shell, does it mean your users will
          need to install it before they can compile your code? Don't you consider
          this a problem?

          See more on this below.

          > 2. Configuration and compilation are much faster than in Autoconf, yielding a
          > faster development process.

          This surprises me. Compilation speed really has nothing to do with autoconf -
          CMake can't make gcc (or whatever) any speedier. And while running "configure"
          for the first time isn't very quick, normally running for the second time is
          indeed quick (because autoconf caches its results).

          > 6. It merges in the functionality of Automake and libtool, both in a more
          > consistent and less error-prone way.

          Sounds interesting. Like I said, I'm not a big automake fan, so indeed I'd
          like to see a better approach .

          > 9. Has native support for Win32.

          What is Win32? Who uses that system anyway? :-)

          > 10. Since CMake is installed separately on the system, the source pacakges are
          > smaller. I was able to reduce the size of the FCS tar.gz from over 500,000
          > bytes to slightly above 200,000 bytes.

          What does this matter?
          As a user, given the choice between downloading an extra 300,000 bytes
          (which takes a few seconds on today's internet) and having to install an
          entire package I don't have (CMake), I'd definitely prefer the first option..


          --
          Nadav Har'El | Monday, Aug 4 2008, 4 Av 5768
          nyh@... |-----------------------------------------
          Phone +972-523-790466, ICQ 13349191 |Are you still here? The message is over.
          http://nadav.harel.org.il |Shoo! Go away!
        • Shlomi Fish
          Hi all! All hail my new sig. I replaced my old one that was mis-understood in several ways due to being taken out of its original context. ... Bourne shell
          Message 4 of 5 , Aug 4, 2008
          • 0 Attachment
            Hi all!

            All hail my new sig. I replaced my old one that was mis-understood in several
            ways due to being taken out of its original context.

            On Monday 04 August 2008, Nadav Har'El wrote:
            > On Mon, Aug 04, 2008, Shlomi Fish wrote about "Re: [hackers-il] Freecell
            Solver's Test Suite + Conversion to CMake":
            > > 1. Its files are written in a consistent high-level language, whose
            > > interpreter is written in C++. This is instead of Autoconf's
            > > m4-generating-Bourne-shell setup, where both m4 and /bin/sh are quirky
            > > and get broken very easily.
            >
            > The whole idea of Autoconf's "m4-generating-Bourne-shell setup" is that
            > the result (the bourne shell) can run on any computer without needing to
            > install anything. Therefore, whatever input language your autoconf
            > alternative has, generating Bourne shell from it is a *good* idea, I
            > think...

            Bourne shell (and especially the one generated from Autoconf) is very
            limiting, error-prone, and tends to break a lot. I recall many times when I
            needed to tweak ./configure scripts by hand due to a bug or limitation there.

            So it's not a good idea to rely on it. If I can:

            1. Make sure I have prepared more consistent, usable and high-level (e.g:
            CMake, Perl, Python, etc.) framework.

            2. Make sure that it is portable enough.

            3. Require the user to install it there.

            4. This in turn will run the "./configure" stage and possibly other stages.

            It will save a lot of future frustrations on my and my user's part.

            >
            > So, if CMake doesn't generate Bourne shell, does it mean your users will
            > need to install it before they can compile your code? Don't you consider
            > this a problem?

            Yes, the users will need to install it before they compile my code. But it's a
            simple matter, and then they can enjoy CMake for life with all the programs
            that require it (see http://www.cmake.org/Wiki/CMake_Projects ).

            So I don't consider it a problem. Forcing one to rely on the least common
            denominator that is pre-installed on all UNIX-like systems is a bit absurd in
            this day and age.

            >
            > See more on this below.
            >
            > > 2. Configuration and compilation are much faster than in Autoconf,
            > > yielding a faster development process.
            >
            > This surprises me. Compilation speed really has nothing to do with autoconf
            > - CMake can't make gcc (or whatever) any speedier. And while running
            > "configure" for the first time isn't very quick, normally running for the
            > second time is indeed quick (because autoconf caches its results).

            Maybe it just feels faster, or the makefiles are less bloated, and don't have
            the long GCC command line. Don't know. I'm doing some stuff that's not
            GCC-related.

            >
            > > 6. It merges in the functionality of Automake and libtool, both in a more
            > > consistent and less error-prone way.
            >
            > Sounds interesting. Like I said, I'm not a big automake fan, so indeed I'd
            > like to see a better approach .
            >
            > > 9. Has native support for Win32.
            >
            > What is Win32? Who uses that system anyway? :-)

            A few people, apparently. It's a priority for me, and for many other FOSS
            projects.

            >
            > > 10. Since CMake is installed separately on the system, the source
            > > pacakges are smaller. I was able to reduce the size of the FCS tar.gz
            > > from over 500,000 bytes to slightly above 200,000 bytes.
            >
            > What does this matter?
            > As a user, given the choice between downloading an extra 300,000 bytes
            > (which takes a few seconds on today's internet) and having to install an
            > entire package I don't have (CMake), I'd definitely prefer the first
            > option..

            CMake is only one package, and it's probably avialable from your
            distribution's sources (or else they won't be able to build KDE 4). And you
            can save the extra 350,000 bytes (possibly more as more Autoconf
            functionality is there) times and again when downloading source packages and
            their new versions. A stitch in time saves nine.

            Regards,

            Shlomi Fish

            -----------------------------------------------------------------
            Shlomi Fish http://www.shlomifish.org/
            What Make Software Apps High Quality - http://xrl.us/bkeuk

            Shlomi, so what are you working on? Working on a new wiki about unit testing
            fortunes in freecell? -- Ran Eilam
          Your message has been successfully submitted and would be delivered to recipients shortly.