Quoting Olly Betts (2025-08-04 21:39:07)
> On Mon, Aug 04, 2025 at 11:11:31AM -0300, David Bremner wrote:
> > > +static inline notmuch_private_status_t
> > > +_notmuch_xapian_errorp (const Xapian::Error &error)
> > > +{
> > > +    const char *type = error.get_type();
> > > +	return (! strcmp (type, "DatabaseModifiedError")) ?
> > > +	    NOTMUCH_PRIVATE_STATUS_OPERATION_INVALIDATED : NOTMUCH_PRIVATE_STATUS_XAPIAN_EXCEPTION;
> > > +}
> > 
> > My first reaction to using strcmp here is ugh, but I guess that's the
> > intended method of introspection. I have CC'ed xapian-users, in case
> > someone there wants to comment. Naively, it seems like a case where
> > typeid would be useful, but I am not enough of a C++ expert to
> > understand why Xapian doesn't do that.
> 
> In C++ the natural approach is to use multiple catch clauses to handle
> some error subclasses differently, i.e. something like:
> 
>     try {
>         do_something_with_xapian();
>     } catch (const Xapian::DatabaseModifiedError&) {
>         do_something_special();
>     } catch (const Xapian::Error& e) {
>         report_error(e);
>     }
> 
> get_type() is more intended to provide a way to report the type for
> logging exceptions, etc.  It should work as above though, and it's
> an exception case so performance shouldn't really matter (but I'd
> guess it actually would be fairly efficient as string comparisons
> should be well optimised).

The reason I went with get_type() was to avoid duplication, as we want
to perform the same cleanup for all Xapian::Error subclasses. The only
thing that differs is the error code we return at the end.

-- 
Anton Khirnov
_______________________________________________
notmuch mailing list -- notmuch@notmuchmail.org
To unsubscribe send an email to notmuch-leave@notmuchmail.org