[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Possible library versioning approach -- (evaluation requested)

On Mon, Dec 23, 2002 at 09:00:19PM -0600, Rob Browning wrote:
> Daniel Jacobowitz <dan@debian.org> writes:
> > I'm dubious.  "A lot longer" is a bit of an understatement, if you
> > look at the number of shared libraries some programs use... there is
> > no substitute for actually paying attention to compatibility.
> My problem is that it doesn't seem like "paying attention" is
> sufficient.  If I don't put the relevant sub-library name/versions
> into libgwrap-glib's soname, then which of the other approaches would
> you favor, the configure-time integer soname lookup table, the hybrid
> name and table approach where you would have a name of libgwrap-glib-1
> and the soname would be chosen from the sub-lib-collection lookup
> table, the placement of the sub-lib names/versions into
> libgwrap-glib's name, or something else?
> Don't forget the complication that gwrap (and I presume other libs)
> needs to be able to build safely against guile 1.3.4, 1.4, or 1.6, and
> *should* be able to build against either glib 1.2 or 2.0.

Proper management of symbol versions and a decently thought-out ABI
mean that this problem never arises.  Consider the way glibc does it;
yes, there are periodic problems based on the ridiculous complexity
level, but it's kept the same SONAME and changed pretty substantially.

I'm not talking about "paying attention" in the libgwrap-glib area but
in the libraries it depends upon.  Anything that doesn't start at the
bottom will not gain enough robustness to be worthwhile, IMVHO.

Yes, I know that versioning has portability issues.  They can be

Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Reply to: