strftime sets errno to 22

strftime sets errno to 22

[Alba Pompeo]

Hello musl people.

I was reading some commits messages from a project and ended up seeing
something that might be interesting to you.

https://github.com/izabera/inutility/commit/1db3ba948ad1d4cbfa8c0164fa475439c99c06bc

I don't think it's been reported yet.

Ciao,
Alba Pompeo

Re: strftime sets errno to 22

[Rich Felker]

On Sat, Jul 23, 2016 at 10:17:30PM -0300, Alba Pompeo wrote:
> Hello musl people.
> 
> I was reading some commits messages from a project and ended up seeing
> something that might be interesting to you.
> 
> https://github.com/izabera/inutility/commit/1db3ba948ad1d4cbfa8c0164fa475439c99c06bc
> 
> I don't think it's been reported yet.

Except for a few special functions with more restrictive
specification, you can only expect errno to contain something
meaningful immediately after a function returns with a return value
indicating an error. The code above was assuming that errno being set
indicated an error even when strftime succeeded.

With that said, at the time this was reported on IRC I looked for the
cause of errno being set to EINVAL and didn't see it. It might be nice
to know why it's happening, just in case some unintended code paths
with nontrivial cost are getting pulled in.

Rich

Re: strftime sets errno to 22

[Alexander Monakov]

On Sat, 23 Jul 2016, Rich Felker wrote:
> With that said, at the time this was reported on IRC I looked for the
> cause of errno being set to EINVAL and didn't see it. It might be nice
> to know why it's happening, just in case some unintended code paths
> with nontrivial cost are getting pulled in.

nsz mentioned on irc it's from strtoul call (when no digits are seen).

Alexander

Re: strftime sets errno to 22

[Rich Felker]

On Sun, Jul 24, 2016 at 09:59:26AM +0300, Alexander Monakov wrote:
> On Sat, 23 Jul 2016, Rich Felker wrote:
> > With that said, at the time this was reported on IRC I looked for the
> > cause of errno being set to EINVAL and didn't see it. It might be nice
> > to know why it's happening, just in case some unintended code paths
> > with nontrivial cost are getting pulled in.
> 
> nsz mentioned on irc it's from strtoul call (when no digits are seen).

It looks like this is something that should be fixed, just because
calling strtoul without checking isdigit allows invalid formats to be
accepted (strtoul consumes leading whitespace and signs, both of which
are unwatned). Probably we want:

-		width = strtoul(f, &p, 10);
+		width = isdigit(*f) ? strtoul(f, &p, 10) : 0;

Thoughts?

Rich