On Thu, 20 Dec 2001, Marcus Crafter wrote:
> Hi All,
>
> Hope all is well.
>
> On avalon-dev there's been a short discussion about the
> Parameters object that is used in the generated sitemap
> class. I thought I'd bring it up here so we can flesh out a
> solution.
>
> Have a look at this snippet from a generated sitemap:
>
> ----------------------
> public class sitemap_xmap extends AbstractSitemap {
> static final String LOCATION = "org.apache.cocoon.www.sitemap_xmap";
>
> static {
> dateCreated = 1007984808622L;
> }
>
> /** An empty <code>Parameter</code> used to pass to the sitemap components */
> private Parameters emptyParam = new Parameters();
>
> <snip>...
>
> if (isSelected("userstatus", "permitted", listOfMaps,
> emptyParam, objectModel)) {
> ----------------------
>
> The problem is 2 fold, and has to do with the 'emptyParam'
> reference when it's passed into the invocation of selectors and
> actions, etc:
>
> 1. The Parameters class itself is not thread safe. Concurrent
> updates to the one object may corrupt data.
>
> 2. There is only one parameter object across the whole sitemap,
> accessed by all threads, and potentially many pipelines.
>
> The second point is the killer if a developer decides to set a
> parameter in a selector, or action, etc, and read it back elsewhere.
> The problem is that the one object is shared among all threads - if 2
> threads update the object (regardless of whether it's threadsafe
> or not), results can be undefined if they use the same key.
>
> As far as I can see there's no real need to be able to set
> parameters in this object in selectors and actions, etc, as the Request
> object can handle sharing of data between pipeline components via
> set/getAttribute (or is there some other use for it ?).
>
> If this is the case, we could make the parameters object read-only
> before it's passed to selectors, etc, which would prevent this problem
> from occuring (for example by wrapping it, or toggling its
> read-only flag).
If we can vote +1 here I can commit an already patched sitemap.xsl which
makes the emptyParam read-only.
Giacomo
>
> Alternatively, we could change the generated sitemap code to
> create a new Parameters object per thread. I'm not sure how much
> this influences other uses of this object though. Perhaps
> someone else a bit closer to that part of the code might be able to
> comment better ?
>
> I'm wondering what everyone else's thoughts are on a solution
> for this as it seems 'emptyParam' is used almost everywhere in
> the generated sitemap class, and could be a touchy thing to
> change.
>
> Any comments/thoughts ?
>
> Cheers,
>
> Marcus
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, email: [EMAIL PROTECTED]
