Re: vasnprintf.c: "out_of_memory", -Wanalyzer-free-of-non-heap, -Wanalyzer-malloc-leak

[Paul Eggert]

On 4/30/22 17:24, Bruno Haible wrote:
> These dependencies save an 'errno = ENOMEM;' assignment in one or
> two places, but can cause integration problems; I am especially
> thinking at the use in GNU libintl and libasprintf.

Ah, I didn't know about the integration problems. I was worried about 
the case where malloc fails with some errno value other than ENOMEM and 
that errno value should be reflected to the user; but that's less 
important than getting integration right. (malloc can fail with EAGAIN 
on GNU/Linux and I assume other errno values are also possible.)

> It would be worth to eliminate the false positive reports by GCC.

Yes, though the weird thing is I'm also using GCC 11.3.1 and am not 
getting the false positives.

> We could
> use
>    assume (result==resultbuf)
> for one part.

I am surprised GCC doesn't deduce that itself; it's part of that same 
weird thing.

If this happens only with unusual configuration settings I expect we 
don't need to worry about it. It's just a warning....

> buf_malloced is NULL 99% of the time; here I prefer the code
> that saves a function call.

Good point; hadn't noticed that. I suppose this can also help make 
branch prediction more accurate.