libvirt has several uses of xmlNodeGetContent() (from libxml2) added at
different times over the years. Some of those uses report an Out of
Memory error when xmlNodeGetContent() returns NULL, and some of them
ignore a NULL return (treating it as if it were ""), and some just
assume that the return will never be NULL, but always at least a pointer
to "".
I ran across this when I noticed a usage of the latter type - it wasn't
checking for NULL at all. A lack of check seemed troubling, so I looked
at other uses within libvirt and found the hodge-podge described above,
so no help there in determining the right thing to do. I then looked at
the libxml2 documentation for xmlNodeGetContent(), which says:
Returns: a new #xmlChar * or NULL if no content is available.
To an uninformed outsider, this sounds like the function could return
NULL simply if the node was empty (e.g. "<wwn/>"). But when we look at
the return from xmlNodeGetContent() for this example, it says that the
content is "", not NULL.
In the meantime, since libxml doesn't abort on OOM errors (as libvirt
does), it could also be possible that it's returning NULL due to OOM. So
using anecdotal evidence acquired so far, one *could* surmise that any
time libvirt gets a NULL return from xmlNodeGetContent(), it is indeed
an OOM error.
The purist in me thinks that isn't right, though - I took a quick look
at the libxml code and saw cases where it returns NULL that don't seem
related to OOM, but rather to the type of node or something. But being
an outsider and not wanting to learn any more than necessary about the
internals of libxml, I'm not sure if any of those cases even apply to
libvirt's simple use of xmlNodeGetContent().
So, in the end I just want to modify libvirt's dozen or so calls to
xmlNodeGetContent() to consistently do the right thing, but first I want
to learn the true answers to these questions:
1) Keeping in mind that we've already successfully parsed the XML, will
calls to xmlNodeGetContent() in the simple cases as when libvirt calls
it only return NULL for OOM, but not for any other reason?
2) If not, is the proper way to distinguish OOM in this case to call
xmlGetLasterror(), and check if the domain is XML_FROM_MEMORY?
3) Aside from returning NULL in the case of errors, would it ever be
possible for correct XML to return NULL as valid "node content", or is
it always an error of some kind?
Since libvirt now aborts on OOM, an OOM error could be handled in one
place by a wrapper function around xmlNodeGetContent() (we already have
such a function, currently a one-liner passthrough, and not called by
everyone). But if there is any chance that any other libxml error could
be encountered, then I suppose we really should be reporting those
without aborting, and then still checking for NULL on return from the
wrapper function (presumably by just logging the contents of "message"
from the xmlErrorPtr returned from xmlGetLastError().
