Re: [xml] sgi regression failure

[Martin Cole <mjc sci utah edu>]

So this is broken on machines without the file /etc/xml/catalog

The new catalog is added fine in the default catalog, its child gets set
correctly, but the catalog->type gets set to XML_CATA_BROKEN_CATALOG
because the default catalog file doesnt exist.

Later when we try to look up the ID, we skip it because of its type.

starting at line:2042 in catalog.c
    while (catal != NULL) {
        if (catal->type == XML_CATA_CATALOG) {
            if (catal->children == NULL) {
                xmlFetchXMLCatalogFile(catal);
            }
            if (catal->children != NULL) {
                ret = xmlCatalogXMLResolve(catal->children, pubID, sysID);
                if (ret != NULL) {
                    if (normid != NULL)
                        xmlFree(normid);
                    return(ret);
                }
            }
        }
        catal = catal->next;
    }

So if I add:

    if (doregister) {
        catal->type = XML_CATA_CATALOG; //<<<--- newly added line
        cur = xmlHashLookup(xmlCatalogXMLFiles, catal->URL);
        if (cur != NULL)
            cur->children = catal->children;
    }

at line: 1549 of catalog.c it works.

I just don't know if this is too heavy handed, or if there is some other
way to set the default catalog. Any thoughts?

--mc


On Fri, 2006-02-24 at 10:20 -0500, Daniel Veillard wrote:
> On Fri, Feb 24, 2006 at 07:55:23AM -0700, Martin Cole wrote:
> > I am attaching the code for a standalone.  As I mentioned in the
> > original mail, yes this code works, it has been tested on linux and mac
> > as well. I guess I should have just given you the code to start with. I
>
>   Oh sorry, I missed that.
>
> Then the simplest way is to run the code in a debugger in parallel on
> both machines and see where the behaviour start to diverge. Of course
> make sure that both are updated to the latest released version.
> You can also export the catalog debug environment variable and see how
> both diverge:
>   http://xmlsoft.org/catalog.html#validate
> This code really isn't platform specific in any way, so I have no
> idea why it would fail to work on a given OS, and without being able
> to debug a live session I doubt it's gonna be easy.
>
> Daniel
>