Unfortunately, it turns out that upstream has changed the ABI and API in a 
backwards-incompatible manner.  In particular, certain symbols from 
sge_tt_text.h are gone.  The difference is very small, however.

Upstream has introduced the .0 soname, but Debian was already using the .0 
soname for the old version.  Trouble.

I ignored symbol changes with C++ mangled names, because the C++ interface 
wasn't really exported before.  The missing symbols with C linkage are, 
according to nm:

delete_char
fast_update
insert_char
keyrepeat
nice_update

I believe that if these symbols (and the corresponding header declarations) 
were reintroduced, the ABI and API would be backwards compatible.

In any case, I packaged the new upstream and updated the Debian packaging to 
match, under the (newly found false) assumption that they were backwards 
compatible.  The new Debian diff is attached to this message (that seemed the 
simplest way, rather than messing around with multiple interdiffs; most of 
the Debian diff to the upstream package is now obsolete).

It's a successful, tested package apart from that; I ran supertransball2 with 
it and it appears to work just fine.

You might actually be able to get away with just doing this, since neither of 
the reverse depends in Debian (libsdl-ruby1.8 and supertransball2) uses the 
missing symbols.  So you won't actually break any package in Debian if you 
cheat on the ABI this time.

It's not really best practice, though.  Normally you should either (a) 
reintroduce the missing symbols, or (b) do a library package rename (with 
appropriate Conflicts/Replaces), and warn all the reverse
depends about the API change.

In any case, I hope that my attachment helps save you some effort.  If you 
don't use it unmodified, it should at least provide a good starting point, so 
all you need to deal with is the API/ABI change thing.

-- 
Nathanael Nerode  <[EMAIL PROTECTED]>

Make sure your vote will count.
http://www.verifiedvoting.org/

Attachment: libsdl-sge_030809-0.1.diff.gz
Description: GNU Zip compressed data

Reply via email to