Re: A suggestion for Sun to take up. - SUN

This is a discussion on Re: A suggestion for Sun to take up. - SUN ; On Tue, 13 Dec 2005 02:24:41 +0000, Dave wrote: > CJT wrote: >> Dave wrote: >>> I don't know about others, but I get rather fed up trying to compile >>> open-source software >> >> It's open source. You're part ...

+ Reply to Thread
Results 1 to 4 of 4

Thread: Re: A suggestion for Sun to take up.

  1. Re: A suggestion for Sun to take up.

    On Tue, 13 Dec 2005 02:24:41 +0000,
    Dave <[...]@southminster-branch-line.org.uk> wrote:
    > CJT wrote:
    >> Dave wrote:
    >>> I don't know about others, but I get rather fed up trying to compile
    >>> open-source software

    >>
    >> It's open source. You're part of the community, so work on it.
    >> B*tching is no help.

    >
    > I'm not just B*tching - I have spent a lot of time (~15-20 hours) on it,
    > passing comments, code, patches etc back to the author etc.
    >
    > But the fact remains it is a pain, and whilst it WILL get there, the
    > time is considerable.
    >
    > Just spent half an hour on the latest release with Sun Studio 11
    > compiling it fine, then to get:
    >
    > gcc -I../include -I. -O3 -fPIC -c MakeDescAux.c
    > sh: gcc: not found
    > *** Error code 1
    >
    > So now it decides it *must* have gcc, whereas I'd made sure gcc was not
    > in the path.


    So, someone's makefiles are broken and hardcore the compiler name, flags
    and other options. This is definitely not Sun's or Solaris' fault and
    can only be fixed by working with the developers of the original
    software. If that turns out to be too slow for you, there's always the
    option of a locally maintained mini-fork.

    About a year ago, I had to make changes to libnet for some local tests I
    wanted to run. Working with the author of libnet was actually pretty
    fast, but not as fast as I'd have liked it. I imported libnet in a CVS
    tree and kept working on it, occasionally sending patches to Mike
    D. Schiffman too. This helped both of us, as he had the time to review
    the patches whenever he wanted and I could move on to the real work I
    had to do with libnet. All hail open source :-)

    Having said that, your idea about something like HP's Testdrive program
    is actually pretty cool!


  2. Re: A suggestion for Sun to take up.

    Giorgos Keramidas wrote:
    > On Tue, 13 Dec 2005 02:24:41 +0000,
    > Dave <[...]@southminster-branch-line.org.uk> wrote:


    >>Just spent half an hour on the latest release with Sun Studio 11
    >>compiling it fine, then to get:
    >>
    >>gcc -I../include -I. -O3 -fPIC -c MakeDescAux.c
    >>sh: gcc: not found
    >>*** Error code 1
    >>
    >>So now it decides it *must* have gcc, whereas I'd made sure gcc was not
    >>in the path.

    >
    >
    > So, someone's makefiles are broken and hardcore the compiler name, flags
    > and other options. This is definitely not Sun's or Solaris' fault


    Of course I am aware it is not Sun's fault, but Sun could increase the
    likelihood of Solaris being an "easy" system to build programs on by
    making some machines available for test purposes.

    > and
    > can only be fixed by working with the developers of the original
    > software.


    The problem with this particular package is that it has the sources
    (sometimes slightly modified) for perhaps 20 different libraries, hence
    there are likely to be 20 different developers. Several have some sort
    of problems on Solaris, ranging from the trivial to quite difficult.

    One particularly important one (GMP) which I expect to run 4x faster on
    64-bit hardware, refuses so far to build as a 64-bit library. Still,
    building the complete program as 32-bit is the first priority. Then I'll
    worry about speeding it up.

    > If that turns out to be too slow for you, there's always the
    > option of a locally maintained mini-fork.


    Yes, I realise that.

    > About a year ago, I had to make changes to libnet for some local tests I
    > wanted to run. Working with the author of libnet was actually pretty
    > fast, but not as fast as I'd have liked it. I imported libnet in a CVS
    > tree and kept working on it, occasionally sending patches to Mike
    > D. Schiffman too. This helped both of us, as he had the time to review
    > the patches whenever he wanted and I could move on to the real work I
    > had to do with libnet. All hail open source :-)


    The main developer of SAGE (http://modular.ucsd.edu/sage/) is very fast
    at responding. We have exchanged 20+ emails in a couple of days. I've
    set him up an account on my Sun, he has done the same for me on a Linux
    box. It makes testing easier, which is why I made the suggestion.

    > Having said that, your idea about something like HP's Testdrive program
    > is actually pretty cool!


    Perhaps someone from Sun will consider it. No doubt it is politically
    more difficult than practically, but I suspect it can be justified
    financially.

    Compaq (then HP) have done it for several years and I expect they would
    discontinue it should they feel the cost outweighed the benefits.

    --
    Dave K

    http://www.southminster-branch-line.org.uk/

    Please note my email address changes periodically to avoid spam.
    It is always of the form: month-year@domain. Hitting reply will work
    for a couple of months only. Later set it manually. The month is
    always written in 3 letters (e.g. Jan, not January etc)

  3. Re: A suggestion for Sun to take up.

    On 13/12/05 8:38, in article 439e8879@212.67.96.135, "Dave"

    wrote:

    > Giorgos Keramidas wrote:
    >> On Tue, 13 Dec 2005 02:24:41 +0000,
    >> Dave <[...]@southminster-branch-line.org.uk> wrote:

    >
    >>> Just spent half an hour on the latest release with Sun Studio 11
    >>> compiling it fine, then to get:
    >>>
    >>> gcc -I../include -I. -O3 -fPIC -c MakeDescAux.c
    >>> sh: gcc: not found
    >>> *** Error code 1
    >>>
    >>> So now it decides it *must* have gcc, whereas I'd made sure gcc was not
    >>> in the path.

    >>
    >>
    >> So, someone's makefiles are broken and hardcore the compiler name, flags
    >> and other options. This is definitely not Sun's or Solaris' fault

    >
    > Of course I am aware it is not Sun's fault, but Sun could increase the
    > likelihood of Solaris being an "easy" system to build programs on by
    > making some machines available for test purposes.


    Another nice thing Sun could do would be to provide a pre-built Solaris 10
    x86 virtual machine for VMware player. Some Linuxes are already there:



    Also there are already some Solaris boxes (SPARC and x86) in sourceforge's
    compile farm.

    > Compaq (then HP) have done it for several years and I expect they would
    > discontinue it should they feel the cost outweighed the benefits.


    I think HP have to try hard to get *anyone* to build on their OS. Solaris is
    *much* nicer, so Sun don't need to try so hard.

    Cheers,

    Chris


  4. Re: A suggestion for Sun to take up.

    Chris Ridd wrote:

    > Also there are already some Solaris boxes (SPARC and x86) in sourceforge's
    > compile farm.


    My experience of them was that they were not particularly fast, which
    made them not too attractive.

    They are only for Sourceforge projects, which excludes quite a few.

    Given the amount of popups and c!$p on the web pages on sourceforge, I
    may not bother using it again. Sourceforge site seems to have gone down
    hill (as we say in England) to me.

    >>Compaq (then HP) have done it for several years and I expect they would
    >>discontinue it should they feel the cost outweighed the benefits.

    >
    >
    > I think HP have to try hard to get *anyone* to build on their OS. Solaris is
    > *much* nicer, so Sun don't need to try so hard.


    May be so.

    But some of the HP hardware they had was blindingly quick. Although they
    said it was not to be used for benchmarking, since the machines were
    always likely to have multiple users, I found the Itaniums really quick
    on floating point, especially the quad processor one with my parallel code.

    Something that took well over an hour on a very old Cray was done in 19s
    on the HP. That said, my code was not written for the Cray's vector
    processor, but was written to use the multiple CPUs on the HP. The HP
    was eating my quad processor Ultra 80 alive!

    That said, I must admit there was never that many users on them. It was
    not too unusual for me to find myself the only one on a really quick box.

    But I have my own HP-UX and Tru64 boxes now, so there is not the
    attraction for testing on the HP site, so have not used the site for > 1
    year.

    I think we do with some Suns machines *without* gcc.

    --
    Dave K

    http://www.southminster-branch-line.org.uk/

    Please note my email address changes periodically to avoid spam.
    It is always of the form: month-year@domain. Hitting reply will work
    for a couple of months only. Later set it manually. The month is
    always written in 3 letters (e.g. Jan, not January etc)

+ Reply to Thread