[Libreoffice-commits] update string copy semantics to be undefined in a non-crashing way

[Bjoern Michaelsen]

On Thu, Oct 04, 2012 at 01:48:15PM +0100, Michael Meeks wrote:
> 	The arguably bit was near the crux of the discussion - IIRC it was
> decided that aborting was not a great idea - we have enough odd
> corner-case bugs crawling out of the woodwork just now.

Thats not how I remember it, I remember you stating that as your opinion and
then leaving for the next call. The remaining folks came to conclude:
- Many of our open and hard to fix bugs are because of not aborting early and
  end on a totally corrupted program state
- we dont have many aborts on illegal progam state in bugzilla at all, so either:
  - Aborting early on an illegal program state is easy an quickly fixed
  - or we arent aborting at all on illegla program state in relevant numbers.

Given that, I would suggest to carefully add more aborts on illegal bug states
esp. on master. If by beta/rc time we find one abort to fire a lot without
being able to fix it, we can still demote the abort to a 

> The SAL_WARN thing sounds sensible - I'd really appreciate it if someone that
> understands the full set of all the warn / log variant macros, and also likes
> using stream operators could work out how to print something more meaningful
> about the error :-) preferably un-conditionally - so no environment variable
> is needed etc. [ is that a SAL_WARN? ].

How about a SAL_ERROR that always prints its message unconditionally and aborts
on dbgutl builds (as we have a tinderbox running subsequentcheck on dbgutl,
right?)?

Best,

Bjoern