Versions of required software

[Sven N. Thommesen]

I have had conversations with Marcus and Paul about the various support 
software that's required to run Swarm 2.0.1. I quote some of the questions 
and answers below, in the hopes that this will be helpful for some. I'll 
have Paul add this info to the FAQ as well.

-Sven


--- the compiler ---

ST>
My understanding is that gcc 2.7.2 and above are now ok, and
that 2.95 is no longer required. Correct?

MD>
I don't really know if 2.7.2 works.  I pretty much use gcc 2.95
across the board, but occasionally use the older egcs 1.1.2
distributed with Redhat 6.

ST>
... but that if you use gcc 2.95, it needs a patch?

MD>
gcc 2.95 has no new bug.  That patch that keeps getting mentioned
suppresses a warning and adds the long long and long double stuff you wanted.

--> NOTE: this means that if your simulation uses long longs or long 
doubles AND you use serialization, you'd better be using the patched 2.95 
compiler!

(...)

MD>
I don't really know if 2.7.2 works.  I pretty much use gcc 2.95
across the board, but occasionally use the older egcs 1.1.2
distributed with Redhat 6.

ST>
Well, the docs say that 2.7.2 and above is ok, so I guess someone
ought to test it? The docs also say egcs 1.0.x is ok, and I can
report I had no trouble compiling Swarm using egcs 1.0.3 on RedHat
5.1.

MD>
Ok, all I'm saying is that right now I don't know of a reason that
2.7.2 wouldn't work but that I haven't confirmed that it does.  At
various points between 1.4.1 and 2.0 I'm sure I did 2.7.2 builds for
one reason or another.  That and because it once worked make me think
it is reasonable to let someone else show that to be false.  Fwiw, I
think this is a fairly common approach and expectation in the
free software community..

(...)

ST>
And 2.95 with your patch is (at the moment) the *only* version of
gcc/egcs that contains the long-long/long-double support?

MD>
IIRC it was just long double.  The runtime needed some other tweaks.


--- system libraries ---

ST> Does Swarm itself care whether you have glibc 2.0.x or 2.1.x?

MD> No.

ST> Does it work with libc 5.x ?

MD>
Unknown.  The avcall and mframe stuff may be broken.  There have been
indications on the list that is the case on swarm-support.  libc5 is
ancient history as far as I'm concerned.


--- the debugger ---

ST> gdb may need a patch?

MD>
I sent in a bug fix to gdb, but don't have it handy.  I just go
straight to the snapshots.  gdb 4.18's name mangling crashes when it
sees phase-separation symbols that Swarm adds.  That was a new bug
that 4.17 didn't have.

PJ> in RH6, gdb is no good. period.


--- avcall / libffi / ffcall ---

ST>
libffi / ffcall are still listed as required -- does avcall now do
the necessary tricks (and if not, on what platforms are libffi or
ffcall needed?)

MD>
libffi is needed on Irix and PowerPC.   ffcall as an external package
is no longer needed, although there might be platforms where newer
ffcall releases would work on platforms yet untested with Swarm.


--- tcl/tk ---

ST>
The latest release of SuSE comes with tcl7.6/tk4.2. Will this
definitely not work, or is it a "don't know" case?

MD>
ScheduleItem uses a Tcl/Tk 8.0 feature.  If you don't use
ScheduleItem, then tcl7.6/tk4.2 should work.  Tcl/Tk 8.0
should be considered the oldest supported version.

--- blt ---

ST>
BLT 2.4i requires a patch for d-n-d to work: does this mean
2.4g or 2.4h don't need the patch?

MD>
I only saw the problem with Tcl/Tk 8.2.  It was not a new bug in BLT,
just an untested configuration.  I see there's a new BLT2.4j.  The BLT
guy acked my patch and said he'd installed it, but I don't see it.
However, he also suggested there was a deeper problem (i.e. the
adaptation I suggested pointed to an invariant that shouldn't have
been violated in the first place).  I haven't tried it, but it did
sound like he'd investigated, so it is probably fixed.

(...)

MD> BLT2.4j seems to work fine, unmodified, with Tcl/Tk 8.2.


--- java: jikes ---

ST>
On the IBM web site, jikes is up to 1.0.4 ("stable") and 1.0.6
("current"). These both are buggy? (I notice that version 0.47
isn't even available on their web site!)

MD>
AFAIK, anything newer than 0.47 can't be used for compiling
models.  However, these versions can be used for building the Swarm
itself.

--- java: jdk ---

ST>
Note that Blackdown has separate tarballs
of jdk1.2pre2 for glibc 2.0 and glibc 2.1 on their web site.

--- java: emacs ---

ST>
emacs 20.x is needed only if one wants to compile the swarm
java support stuff, not if the binary rpms are used?

MD> Right.


--- hdf5 ---

ST>
Is the hdf5 supplied the same as that available from NCSA, or
has that also been modified?

MD>
It (if you mean the Debian and Solaris packages) is the same as the
latest beta release from NCSA.  The source tarball in needed-software
is a copy.

PJ>
I never did figure what sfi had for hdf5, but mine is the
1.2beta1release2 without any tampering.


--- "R" ---

ST>
The rpms for R are named R-with-hdf5-*. Are these different
from the R found on CRAN?

PJ> Yes, different because they have hdf5 support, the standard ones do not.




                 ==================================
  Swarm-Support is for discussion of the technical details of the day
  to day usage of Swarm.  For list administration needs (esp.
  [un]subscribing), please send a message to <address@hidden>
  with "help" in the body of the message.