Re: [PATCH] itoa

[Corinna Vinschen <vinschen at redhat dot com>]

Hi Jon,

On Dec  8 18:13, Jon Beniston wrote:
> Hi Corinna,
> 
> > What other platforms?
> 
> Various Windows compilers and a variety of embedded targets, but no,
> not generally unices (although you often see people asking for it in
> various forums).

Ok.

> > The API is rather unsafe, given that the size of the buffer is not
> > checked (no size parameter).  
> 
> The API is that the buffer should be sizeof(int)*8+1 to handle the
> worst case.  E.g: http://www.cplusplus.com/reference/cstdlib/itoa/
> I've added this to the DESCRIPTION.

Given that this is non-standard anyway, what about creating a safer
alternative instead, along the lines of Craig's/Freddie's itoa_s/np?

Also, shouldn't the functions start with an underscore so as not to
pollute the standard namespace?  itoa and utoa are very likely not
uncommon as function names in existing projects.

> > But what I'm really missing are checks for the base parameter being
> > valid.  It could be > 36.  Or negative.
> 
> According to the docs, base must be between 2 and 36. I didn't add
> this check initially, as I don't see much parameter range checking in
> newlib generally, however, I've now added a check and it terminates
> the string and returns NULL if base is out of range. (It seems MSVC
> doesn't check the base, as it divides by zero if you use 0).
> 
> I've also modified it as per Freddie's suggestion.

Did you test it?  AFAICS, the itoa function never returns the sign:

  if (val < 0)
    val = -val;
    *str++ = '-';
  return utoa (val, str, base);

this returns a pointer to str + 1 in case val is < 0.


Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

Attachment: pgpW1igfNRU31.pgp
Description: PGP signature