Em qui., 23 de abr. de 2020 às 16:27, Peter Geoghegan <pg@bowt.ie> escreveu:
On Thu, Apr 23, 2020 at 11:41 AM Ranier Vilela <ranier.vf@gmail.com> wrote: > And if I already propose a solution even if it is not the best, it is much better than being silent and missing the opportunity to fix a bug.
The problem with that theory is that you're not creating any value over simply running Coverity directly. Your patches don't seem to be based on any real analysis beyond what makes Coverity stop complaining, which is not helpful.
In some cases, this may be true. But in fact I already fixed some bugs with this technique. Even you already used a patch of mine to provide a fix. Wasn't that helpful?
For example, the nbtree.c/btvacuumpage() issue you reported yesterday involved a NULL pointer dereference, but if the code path in question ever dereferenced the NULL pointer then it would be fundamentally wrong in many other ways, probably leading to data corruption. The fix that you posted obviously completely missed the point. Even when Coverity identifies a serious issue, it usually needs to be carefully interpreted.
I disagree. In case of nbtree.c/btvacuumpag().
If you are validating "opaque" pointer, in three different ways to proceed with cleaning, nothing more correct than validating the most important first, if the pointer is really valid. And that is what the patch does.
Anybody can run Coverity. Many of us do. Maybe the approach you've taken would have had a noticeable benefit if you were not dealing with a codebase that has already been subject to lots of triage of Coverity issues.
Sorry, but the plsql-bugs list, has many reports of segmentation faults,
that shouldn't exist, if everyone uses Coverity or other tools, after writing the code.
