 |
 |
Hi, Andy the problem is not with Forte compiler in general, just with Forte version 7 and the way it handles templated classes (by putting separate object files in special cache folders in the directory where the objects get created). This results in lots of problems with libtool, because libtool moves object files around (especially convenience libraries) and thereby the template objects may get lost. Also, I experienced a whole number of issues with multiply defined symbols, deriving from libraries containing the same symbols deriving from templates. In GNU and also in the newer Solaris compilers there are no such problems (these symbols are defined as weak, and the linker is not disturbed by multiple identical weak symbols). There is no problem with the newer Forte compilers and I think the support for Forte7 will run out soon. I had some discussions with the Authors of libtool about this issue and I'm not sure whether you find many packages that are complex and use templates. At least they told me that Solaris is one of the "more difficult" systems. But maybe you can give me some tips, I'm no Solaris expert. But I believe that a nice autotools build is really something valuable, because it is such a standard and it makes package building for many distributions so much easier. (standard make targets, respects $prefix and $DESTDIR, automatic tarball building and sanity check. Possibility to create multiple compiled versions with different options/architectures all referring to the same sources directories, configure script with standard behavior) Cheers, Derek On Monday 26 September 2005 13.52, Andrew Hanushevsky wrote: > Hi Pete, > > No, I think it would be wrong to disallow using the Forte compiler. In > general, that compiler picks up errors that other compilers do not and > produces better code on Solaris/Sparc. Additionally, it provides me with a > good dbx-able debug version that I don't get from gcc. In general, it > would be a royal pain for me if I could not use that compiler. Since > autotolls does work with that cmpiler for other packages (I have used > packages that use autotools and the Solaris compiler), I don't immediately > see what the problem is and, whatever it is, why it can't be circumvented. > > Andy > > On Mon, 26 Sep 2005, Peter Elmer wrote: > > Hi All, > > > > Derek and I talked on Friday about the migration to autotools. The main > > outstanding issue (with libtool) would be solved by dropping support > > for building xrootd with Workshop6 and Forte7 on Solaris. Is there > > anybody out there who builds xrootd themselves with Forte7 on Solaris > > rather than using one of the binary downloads? > > > > [Andy, I see you still have Forte7 as the default compiler on your > > machine at SLAC, but I'm guessing that this is just because BaBar > > constrained this to be the compiler installation on the Sol9 machines. Is > > that correct? (Anybody know which compiler version shipped with Sol9? In > > any case the BaBar SRT overrides the default compiler version and > > actually uses the Studio9 installation from afs in the latest releases.] > > > > I _think_ SLAC and In2p3 were the main sites which had solaris and I > > think that both of them use the binary downloads. ROOT appears to > > support "CC5.5", which must be XXXX8 (where XXXX8 is whatever name Sun > > chose between Forte7 and Studio9, how tedious Sun is...). Derek thought > > that there should be no issues with XXXX8, but was going to verify this. > > > > This said, if nobody sees any urgent reason to keep support for > > building xrootd with Workshop6 and Forte7, I would propose that we will > > do the following: > > > > o Since nobody has reported big problems with xrootd 20050920-0008, > > I can declare that a "production" build (the last one supporting > > Workshop6 and Forte7, effectively) (Wilko?) > > > > o Derek and I can do the autotools migration on top of this coming > > week and produce another version completely equivalent (for the > > existing xrootd code, plus an optional module for the authorization > > for Alice) to 20050920-0008 up to the autotools changes (and no > > Workshop6 or Forte7 support). This would also be the starting point for > > the next round of development. > > > > Comments? > > > > thanks, > > Pete > > > > ------------------------------------------------------------------------- > > Peter Elmer E-mail: [log in to unmask] Phone: +41 (22) 767-4644 > > Address: CERN Division PPE, Bat. 32 2C-14, CH-1211 Geneva 23, Switzerland > > ------------------------------------------------------------------------- -- Dr. Derek Feichtinger Tel: +41 22 767 10 07 LCG/ARDA Group email: [log in to unmask] CERN http://people.web.psi.ch/feichtinger CH-1211 Genève 23