Re: FYI: by default, elide analyzer-related warning options

[Dmitry V. Levin]

On Sat, Jan 01, 2022 at 03:51:26PM -0800, Paul Eggert wrote:
> On 1/1/22 12:08, Jeffrey Walton wrote:
> > Shouldn't the litmus test be "bug free code"?
> 
> No, as that's not practical when developing real-world applications.
> 
> In practice, -fanalyzer is sooooo slow that it has significant adverse 
> effects on ordinary development. Among other things, this can cause 
> developers to do less testing, which hurts reliability. So it's 
> reasonable to enable -fanalyzer only on request.
> 
> As Jim wrote, the case for -Wanalyzer-double-free and 
> -Wanalyzer-use-after-free would be stronger if these options were 
> finding real bugs in coreutils and grep. In practice, though, they're 
> not doing that.
> 
> PS. If you want more-extensive checking, and even slooooooower builds, 
> try -flto -fanalyzer.

It's not just slow, it's also a memory hog to the extent that
it doesn't fit into virtual memory on 32-bit architectures, see
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103602


-- 
ldv