On Tue, Apr 05, 2005 at 01:22:50PM -0400, Joel Reed wrote:
I'm reviewing xmlFileOpen_real and xmlGzfileOpen_real in xmlIO.c and have two quick questions/possible improvements (i'll fire off a patch if no dissenters). 1) xmlFileOpen_real does a if (!xmlCheckFilename(path)) return(NULL); before the fopen(path, "r") call. This quick-out prevents library users from ever seeing a message in their structure error handler like "No such file or directory." By deleting this check, we can let fopen call fail and set errno, then let normal libxml error handling bubble the error back up to library users. Any reason why not to do this?
after poking around alot in xmlIO.c it seems like far too much work to report "No such file or directory" errors to the structured or generic error handlers - i'm pretty sure its not even possible to do correctly without impacting the ABI, such that the existing places in libxml2 which rely on such errors not being passed to the error handlers continue to work as-is while everyone else can process such errors. while i was in there i did look at the stat() call reduction, as DV suggested. i've attached a patch that reduces the number of stat calls by 1 on platforms with HAVE_ERRNO_H.
3) xmlGzfileOpen_real does not have anything like:
if (fd == NULL) xmlIOErr(0, path);
after the gzopen(path, "rb") call, unlike xmlFileOpen_real. this
minor inconsistency could be addressed easily enough. Any reason not to?
the 2nd hunk in the patch corrects xmlIO.c xmlGzfileOpen_real to act like xmlFileOpen_real in reporting IO errors to the handles on a NULL fd - EXCEPT when its due to ENOENT - a non existant file. jr
Attachment:
stat-less-2.6.19.patch
Description: Text document