On 2024-10-19 09:14, Alejandro Colomar wrote: ...
On Sat, Oct 19, 2024 at 10:25:52AM GMT, Bruno Haible wrote: ...What I would therefore suggest — if we want to change Gnulib at all regarding realloc — is to make realloc (ptr, 0) with ptr != NULL abort.That would be bad. Programs that _need_ realloc(p,0) to work will have to workaround it by rolling their own realloc() wrapper, which will most likely contain bugs.
Perhaps some package maintainers will prefer Bruno's suggestion, while some will prefer realloc(p,0) to have the traditional/BSD semantics you're proposing. So we could have two new Gnulib modules, say realloc-ptr-0-abort for Bruno's suggestion, and realloc-ptr-0-nonnull for the traditional/BSD semantics.
I can work on developing realloc-ptr-0-nonnull and using it in some of the packages I help maintain. That would help us move forward. I hope a realloc-ptr-0-abort module isn't needed except perhaps briefly for debugging.
If the tunable is added to glibc as Siddhesh is proposing, I suppose the realloc-gnu module should be changed to behave more like realloc-ptr-0-nonnull, as it's more convenient for package maintainers to assume the realloc-ptr-0-nonnull behavior than to write code that can work with either current or future (or ancient :-) glibc behavior.
