On Saturday 04 March 2006 11:04, Andrey Borzenkov wrote:
> [moved to workers]
>
> On Thursday 02 March 2006 20:52, Francisco Borges wrote:
> > % typeset -U dirstack
> >
> > and the shell crashed.
>
> The problem is rather non-trivial. dirsgetfn returns array built on-the-fly
> in heap, while typeset -U calls uniqarray() that tries to zfree array
> elements. There are at least two problems here:
>
> - typeset -U is not prepared to deal with "pseudo" parameters at all. It
> assumes a->getfn() returns pointer to real parameter value. So it would
> have not worked for dirstack anyway
>
> - I was about to change typeset -U to pm->gsu.a->setfn(pm,
> pm->gsu.a->getfn(pm)) (basically doing foo=($foo)) and adding uniqarray
> call to dirssetfn() when I realized that it would not help at all in this
> case as dirssetfn() tries to free passed value too; so it would have
> crashed just the same.
>
> Apparently to solve it in general we need one of
>
> - per-parameter type ->uniq function (is it an overkill?) Possibly
> generalized to per-parameter ->setflags function.
>
> - some way to know if passed pointer was allocated from heap or not. I
> guess it should be possible; something like isheap(p)?
>

Assuming the issue with freeing heap memory is resolved, the patch adds 
support for -U to dirstack. It also adds framework for adding uniqueness 
support to any parameter by changing typeset to simply do equiv. of 
'foo=($foo)'. It also uncovered one case of using free() instead of zfree(); 
assuming previous patch is accepted it is necessary to audit code and replace 
all plain free() with zfree().

-andrey