On Tue, 19 Jun 2012, Dominique Leuenberger a.k.a DimStar wrote:
Quoting Richard Guenther
: On Tue, 19 Jun 2012, Stephan Kulow wrote:
On 19.06.2012 14:15, Richard Guenther wrote:
I don't see why "manual" is bad. In fact we should be perfectly aware about an ABI/API change _before_ we check things in. Being ignorant and relying on automatic rebuilds "fixing" things is even worse than a concious "manual" rebuild.
Aehm, who is perfectly aware of ABI changes?
I invite you to review e.g. https://build.opensuse.org/request/show/125038
ABI change or not?
No ABI change by definition as the SONAME does not change.
That's assuming that upstream knows what they do... and NEVER miss out on anything.
Yes, and you only detect if that is true (and thus detect bugs) if you actually _not_ rebuild $world after Qt 4.8.2.
Do we WANT to run into that? knowing that this results in random crashes? Yes, of course, SONAME should be bumped by upstreams if that happens.
Well, while we are able to "fix" issues in the packages we build ourselves (only if you stay inside a single repository!) there are people who build their own stuff - and you'd break their binaries. So YES! We DO want to run into this.
On the other hand: you see SO many upstreams bumping SONAME only if the 'break' existing ABI... ADDITION of it seems to be done without changes.. (thus usually resulting in 'requiring libfoo.so.1 not being enough', as you need libfoo.so.1 >= version it was built against)
Well, more people need to be teached about symbol versioning.
Richard.
--
Richard Guenther