[Development] DBus signals for (dlclose'd) style plugin causing crash during application exit?

Thiago Macieira thiago.macieira at intel.com
Tue May 2 13:37:54 CEST 2017

On Tuesday, 2 May 2017 02:48:15 CDT René J. V. Bertin wrote:
> I'm aware of the direct reason. Apparently you're sure enough that the CI
> failure is without consequence to put the burden and responsibility of
> incorporating the patch on end-users. That's what I mean with getting an
> internal act together: what do you tell a commercial client who runs into
> this issue? That you're waiting for data from the wild to address a
> potential side- effect of a fix that is clearly necessary?

The commercial client will complain to paid support, support will investigate 
the issue and figure out what the regression is.

> I'm sorry but I cannot find other words than "man up".
> No one can reproduce the CI failure on a regular set-up? Fine, maybe the CI
> is flawed somewhere? Incorporate the patch with whatever workaround you can
> find for use on the CI (a configure flag to deactivate the patch, an extra
> runtime check that avoids the fatal operation even if the condition should
> never occur, anything that works).

Yes, none of the regular QtDBus developers can reproduce the issue nor anyone 
who has investigated the issue could. (That set of people has a count of 1)

The problem in 5.6 is an autotest that fails. In 5.8, the build deadlocks, so 
we can't even just disable a test. We'd have to disable all of the code that 
uses qdbusxml2cpp, like dbusmenu and dbustray. That's not acceptable.

> If the regression is not a false alarm you'll end up getting a bug report
> with the missing information because everyone will be using the patch.
> You'll finally address a known bug with confirmed sightings in the wild,
> POSSIBLY introducing another bug which might never be triggered.

It's not a false alarm. IT's reliably reproduceable in the CI.

Just not on the developers' computers. (again: population of 1)

> Not just the (un)happy few who are aware of a patch that you cannot even
> obtain easily. Digging up TWO codereviews and getting the patches in usable
> form is not what I have in mind with "easily", and after that you still
> need to build your own Qt copy.

Yeah, hoping that someone annoyed enough will reproduce the regression and 
submit a fix to my patch.

Thiago Macieira - thiago.macieira (AT) intel.com
  Software Architect - Intel Open Source Technology Center

More information about the Development mailing list