| 1 |
On Sunday, January 02, 2011 02:31:06 Ryan Hill wrote: |
| 2 |
> On Sun, 2 Jan 2011 00:08:34 -0500 Mike Frysinger wrote: |
| 3 |
> > On Saturday, January 01, 2011 23:09:11 Enrico Weigelt wrote: |
| 4 |
> > > BTW: several blog/maillist postings talked about the problem that |
| 5 |
> > > even on recompile, older library versions could be linked in even |
| 6 |
> > > on recompile. |
| 7 |
> > |
| 8 |
> > you'll need to provide an actual example. i have yet to see one. |
| 9 |
> |
| 10 |
> Not exactly the same thing, but the recent spidermonkey bump changed the |
| 11 |
> lib from libjs.so to libmozjs.so. mediatomb kept finding the old libjs on |
| 12 |
> rebuild. I had to uninstall mediatomb to dump the library and reinstall |
| 13 |
> it. |
| 14 |
> |
| 15 |
> There's not much we can do about stuff like that, but it's not an uncommon |
| 16 |
> occurrence. I usually run into a sticky preserved lib or two every couple |
| 17 |
> months. |
| 18 |
> |
| 19 |
> I still think it's much better than the current portage 2.1 behavior which |
| 20 |
> results in a broken mediatomb. I'll take the rare possibility of a symbol |
| 21 |
> collision over a guaranteed broken package any day. |
| 22 |
|
| 23 |
yes, this is a case where linking would pick up the old library. but that is |
| 24 |
*only* because the SONAME is stupid and is the same thing as the link name. |
| 25 |
imo, these packages are pretty much broken and should get their SONAME fixed. |
| 26 |
a simple change as making the SONAME "libjs.so.0" instead of "libjs.so" would |
| 27 |
make this a non-issue. |
| 28 |
-mike |
Archives