dolfin team mailing list archive

[Barry Smith <petsc-maint@xxxxxxxxxxx>]

> As far as I can see, the information that is being hardwired into the PETSc 
> library files is the installation          location of the PETSc library files 
> themselves. Correct me if I'm wrong. 

  In the PETSc shared libraries are hardwired the locations of the PETSc dynamic
libraries (the dynamic libraries contain all the actual implementations while 
the shared libraries (what you link against) contain the user API functions).
If we do not hardwire, we have no way of finding the dynamic libraries.

> If the idea is so that the linker can find the library at runtime, that is
> what standard locations like /usr/lib are for, and search paths for the linker
> can also be set in /etc/ld.so.conf.

  /usr/lib reminds me about the old days of Windows where you dump all your
dlls into System32 (even Microsoft finally figured out this was a stupid idea).
And changing a file like /etc/ld.so.conf is insane.

  Now we could require the user to set PETSC_DIR or something similar at RUNTIME
to locate the dynamic libraries, but I dislike that.

  Given all this new information I'm going to have to rethink the entire process.

   Barry


On Tue, 5 Jul 2005, Faheem Mitha wrote:

> 
> 
> On Tue, 5 Jul 2005, Barry Smith wrote:
> 
> > 
> >  Well knock me over with a sledge hammer; the one constant in my world
> > is gone! From
> > 
> > http://www.gnu.org/software/autoconf/manual/autoconf-2.57/html_mono/autoconf.html#SEC34
> > 
> > By default, configure sets the prefix for files it installs to
> > `/usr/local'. The user of configure can select a different prefix
> > using the `--prefix' and `--exec-prefix' options. There are two ways
> > to change the default: when creating configure, and when running it.
> 
> Yes.
> 
> > > Now, if the location of the binaries is hardwired into them at compile
> > > time
> > > using --prefix as the location, which seems to be what is going on, then
> > > this
> > > would explain my problems.
> > 
> >  We are hardwiring into the libraries the location of the dynamic libraries
> > that need to be loaded at runtime. The location is determined by --prefix
> > as I understood --prefix to mean.
> 
> I'm not clear on what you mean. By 'the libraries' you mean the PETSc library
> files, I presume? Then, what do you mean by the 'dynamic libraries'?
> 
> As far as I can see, the information that is being hardwired into the PETSc
> library files is the installation location of the PETSc library files
> themselves. Correct me if I'm wrong.
> 
> In general, hardwiring things is a bad idea because it makes things break when
> you want to move files around, which you invariably do when packaging.
> 
> If the idea is so that the linker can find the library at runtime, that is
> what standard locations like /usr/lib are for, and search paths for the linker
> can also be set in /etc/ld.so.conf.
> 
> There is nothing in the GNU standards about hardwiring things as far as I
> know, though I could be wrong.
> 
> >  How should we get the information to compile into the libraries? If we
> > cannot use --prefix do we have to make up our own option name
> > --with-install-directory=...?
> 
> I dunno. Why not ask around where technical people hang out? The various
> Debian lists are pretty helpful, and there might be some appropriate GNU list.
> 
> I don't claim to be an expert on any of this. I'm just trying to make a
> package of PETSc. I'm sure there are other people far better qualified to give
> advice.
> 
> >  Barry
> > 
> > And I use to think that Windows sucked.
> 
> Oh, come now. That's uncalled for.                              Faheem.
> 
>