Re: [PATCH v2 0/8] object_id part 4

[Jeff King <peff@peff.net>]

On Sun, Jun 19, 2016 at 10:50:14AM +0200, Johannes Sixt wrote:

> Am 19.06.2016 um 00:13 schrieb brian m. carlson:
> > * Adjust the Coccinelle patches to transform plain structs before
> >    pointers to structs to avoid misconversions.  This addresses the issue
> >    that Peff caught originally.
> 
> To avoid future mistakes, can you write down how "transform plain structs
> before pointers to structs" looks like? Is it a particular order of
> Coccinelle rules? Which part of the interdiff between the previous round and
> this round makes the difference?

Yeah, I'd also like a better understanding of what went wrong in the
original (just to be able to better evaluate the tool).

Also, for the record, the issue was noticed by Johannes originally, not
me, as indicated above (I just found a similar case after seeing his
report).

> On a tangent, I wondered recently, why we need oidcpy() and oidclr(). After
> all, in place of, e.g.,
> 
> 	oidcpy(&pair->two->oid, &p->oid);
> 	oidclr(&one->oid);
> 
> we can write
> 
> 	pair->two->oid = p->oid;
> 	one->oid = null_oid;
> 
> Is there a particular reason *not* to make this transition? I find the
> latter less cluttered with equal clarity.

I think traditionally we've avoided struct assignment because some
ancient compilers didn't do it. But it's in C89, and I suspect it's
crept into the code base anyway over the years without anyone
complaining.

-Peff