On Dec 23, Lionel Elie Mamane (<email address hidden>) wrote:
> Package: libsp1
> Version: 1.3.4-1.2.1-42
> Severity: grave
>
> Machine one:
>
> ii libsp1 1.3.4-1.2.1-28 Runtime library for James Clark's SP suite
>
> master@tofu:~$ readelf -s /usr/lib/libsp.so.1|grep __ti16OutputByteStream
> 1203: 0020c888 8 OBJECT GLOBAL DEFAULT 20 __ti16OutputByteStream
>
> Machine two:
>
> ii libsp1 1.3.4-1.2.1-42 Runtime library for James Clark's SP suite
>
> lmamane@topaz:/usr/src/backports/pdns/pdns-2.9.16$ readelf -s /usr/lib/libsp.so.1|grep __ti16OutputByteStream
>
> This is an incompatible ABI change, and warrants a soname change. As a
> result, jade doesn't work anymore on machine two, since it was linked
> against an older version of libps1, that had the __ti16OutputByteStream
> symbol.
Jade has had a dependency on libsp1 >= -30 since -30, so how did you get jade
-42 installed with libsp1 -28? You are correct that an ABI change should be
handled with a different soname, however: (1) the only package affected is
jade, which has dependencies to prevent this, (2) it will not affect anyone
upgrading from woody to sarge since -28 was in woody, and the newer versions
of jade will force an upgrade to a version of libsp1 >= -30, and (3) it's darn
close to the sarge release to be making a soname change that doesn't really
affect anyone who installs normally, i.e., in a way that the dependencies
would take care of the problem. Do you really think this should be a release
critical bug? I'd be more inclined to change this if I had reports that there
were other programs that need the old libsp1 ABI as well as programs that need
the new one.
On Dec 23, Lionel Elie Mamane (<email address hidden>) wrote: libsp.so. 1|grep __ti16OutputByt eStream eStream topaz:/ usr/src/ backports/ pdns/pdns- 2.9.16$ readelf -s /usr/lib/ libsp.so. 1|grep __ti16OutputByt eStream eStream
> Package: libsp1
> Version: 1.3.4-1.2.1-42
> Severity: grave
>
> Machine one:
>
> ii libsp1 1.3.4-1.2.1-28 Runtime library for James Clark's SP suite
>
> master@tofu:~$ readelf -s /usr/lib/
> 1203: 0020c888 8 OBJECT GLOBAL DEFAULT 20 __ti16OutputByt
>
> Machine two:
>
> ii libsp1 1.3.4-1.2.1-42 Runtime library for James Clark's SP suite
>
> lmamane@
>
> This is an incompatible ABI change, and warrants a soname change. As a
> result, jade doesn't work anymore on machine two, since it was linked
> against an older version of libps1, that had the __ti16OutputByt
> symbol.
Jade has had a dependency on libsp1 >= -30 since -30, so how did you get jade
-42 installed with libsp1 -28? You are correct that an ABI change should be
handled with a different soname, however: (1) the only package affected is
jade, which has dependencies to prevent this, (2) it will not affect anyone
upgrading from woody to sarge since -28 was in woody, and the newer versions
of jade will force an upgrade to a version of libsp1 >= -30, and (3) it's darn
close to the sarge release to be making a soname change that doesn't really
affect anyone who installs normally, i.e., in a way that the dependencies
would take care of the problem. Do you really think this should be a release
critical bug? I'd be more inclined to change this if I had reports that there
were other programs that need the old libsp1 ABI as well as programs that need
the new one.
--
Neil Roeth