I also went down the route of using a descriptor class to cache all the FieldInfo instances per type. It was actually using FieldInfo.SetValue()/GetValue() that turned out to be slow for me. I agree that it definitely seems like a more elegant solution, but in my particular case it was not.
-----Original Message----- From: Murphy, James [mailto:[EMAIL PROTECTED]] Sent: Friday, May 10, 2002 9:49 AM To: [EMAIL PROTECTED] Subject: Re: [DOTNET] OT: Relational to Object mapping. The goal is to have all the specificity you want without being invasive. The persistent metadata for the class is orthogonal to the class design. You can definitely get yourself into trouble if you do too much heavy lifting with reflection APIs but I think this can be done once per class type and cached where necessary to offset performance problems. Since its early everything I've done has been single threaded. I'm obviously interested to see what contention issues exist in the reflection classes. :-/ Jim > -----Original Message----- > From: Sinnott, John [mailto:[EMAIL PROTECTED]] > Sent: Friday, May 10, 2002 12:46 PM > To: [EMAIL PROTECTED] > Subject: Re: [DOTNET] OT: Relational to Object mapping. > > > Not to say that your experience (or code) couldn't be better > than mine, but > I found in writing my own O/R mapper for .NET that reflection > seemed fast > enough when benchmarking a single thread, but when testing against a > concurrent system, it became very slow. I switched instead > to defining a > persistent base class that stores its state in a hashtable. > Invasive, yes, > but much, much faster. It also allows for some more > flexibility in what > data gets loaded into an object. > > John > > -----Original Message----- > From: Murphy, James [mailto:[EMAIL PROTECTED]] > Sent: Friday, May 10, 2002 8:25 AM > To: [EMAIL PROTECTED] > Subject: Re: [DOTNET] OT: Relational to Object mapping. > > > 3. Fast and scalable > > > > *** I object - NOT with reflection. > > Oh - sure play the "Reflection is slow" card. You take a hit > the first time > the metadata is accessed but afterwards I've found its very > adequate. Given > the usage patter than a typical application will save and > load a very finite > set of class types the reflection performance problem isn't one. > > You can read messages from the DOTNET archive, unsubscribe > from DOTNET, or > subscribe to other DevelopMentor lists at http://discuss.develop.com. > You can read messages from the DOTNET archive, unsubscribe from DOTNET, or subscribe to other DevelopMentor lists at http://discuss.develop.com. You can read messages from the DOTNET archive, unsubscribe from DOTNET, or subscribe to other DevelopMentor lists at http://discuss.develop.com.
