Re: looking at fmtmsg-is dprintf oom-safe?

[Rich Felker <dalias@libc.org>]

On Thu, Apr 24, 2014 at 09:57:44AM -0700, Isaac Dunham wrote:
> Hello,
> I've been poking at fmtmsg(), since if I added that and used sdbm for
> dbm.h, I'd have XSI coverage.
> Due to the use case for fmtmsg (reporting errors, including terminal
> errors), I expect it needs to work even when malloc() fails.
> My understanding is that this would exclude using fopen().
> 
> I could use several calls to write() (I currently have a preliminary
> version that does this), but dprintf() would be more optimal.
> As far as I can tell, dprintf() should still work when malloc() fails,
> but I'd like to confirm this.

Yes, dprintf runs in O(1) memory (stack) and is async-signal-safe. The
stack usage is tiny unless you're printing floating point, and in the
fp case, something like 512 bytes to 8k depending on arch the size of
long double.

Rich