[Freeciv-Dev] Re: Question about attributes

[Mike Kaufman <mkaufman@xxxxxxxxxxxxxx>]

On Thu, Jan 10, 2002 at 12:56:31PM +0100, Raimar Falke wrote:
> On Thu, Jan 10, 2002 at 06:37:38AM -0500, Jason Short wrote:
> > Raimar Falke wrote:
> > 
> > > On Thu, Jan 10, 2002 at 05:53:23AM -0500, Jason Short wrote:
> > >>Raimar Falke wrote:
> > >>
> > >>>You see the attributes aren't really used. I agree that a seperate
> > >>>attribute_length is the way to go. Please send a patch.
> > >>>
> > >>What about just malloc'ing the buffer in attribute_get?  This saves the 
> > >>extra function call (which will generally always be necessary), at the 
> > >>small cost of having to free the buffer later.
> > > 
> > > This all depends on the usage patterns. So far we have not enough
> > > users to make a final decision. The agents I have coded so far doesn't
> > > need these attribute_length call at all because they know the size. If
> > > more users like the tile markers are created we may have to rethink
> > > this. Also note that the tile markers in at least one case needs a
> > > slightly larger buffer.
> > 
> > Interesting.  Keep in mind that this data is unverified; if you connect 
> > to a server in place of some former player, and that player has uploaded 
> > invalid attributes, your client should not die.  So it is important that 
> > error-checking be done in all cases.
> 
> I agree. The error-checking which is in place is via asserts.

this is not good enough. This will cause the client to die via this
assert: assert(max_data_length >= length); if the previous player had
larger sized attributes than I (and these are not necessarily "invalid"
attributes, after all, they're client data).

Or you're forcing users not to screw around with "blessed" attributes.

-mike
> 
> > > Another solution: attribute_get will allocate a large-enough buffer if
> > > the supplied data pointer is NULL.
> > 
> > And return a pointer to the buffer?  Or will the user pass in a 
> > pointer-to-pointer?  Either seems like a reasonable solution.
> 
> The return value is currently used for the size of the attribute.
> 
> Let us implement attribute_length and reconsider this thing if we have
> 5 users of the attributes.
> 
>       Raimar
> 
> -- 
>  email: rf13@xxxxxxxxxxxxxxxxx
>  Q:  Do you know what the death rate around here is?
>  A:  One per person.