I’m definitely a fan of keeping software architecture as simple and as minimal as possible… KISS, YAGNI etc. In line with this I try to keep use of third party tools to an absolute minimum and stick with out of the box stuff as much as possible.
One tool I’ve used in the past based on its popularity and proposed benefits is AutoMapper. It’s used to automatically assign values from one object to another based on conventions such as matching property names. It’s used quite a lot to map entity classes to viewModel/EditModel classes and despite its authors protestations is often used to map viewModel/EditModel classes back to entity classes too. Does AutoMapper save time?
AutoMapper doesn’t actually save much time
AutoMapper is supposed to save the time developers would spending typing out all the manual and so called tedious left-right assignments in code. I’ve used AutoMapper on three projects and each time so much configuration was required or so much confusion about why something wasn’t working as expected was experienced by the developers that after all was said and done it would have just been quicker to type the code out manually. And besides left-right assignment code is pretty much the easiest code any developer is ever going to have to write, everyone understands it (including the BAU team who did not write the app but who will support it) and it’s very rare to lose significant time on software projects because of the sheer volume of code we have to type. It’s always the research, whiteboarding, collaboration, brainstorming required to figure out what code to write, not actually writing it that costs the most time.
In the projects I worked on which used AutoMapper tasks which the team lost time on included debugging cryptic error messages, writing custom rules when something non trival is inevitably needed, explaining to junior developers how it all works/is supposed to work and funnily enough discussing the merits of using it or not. All these tasks added up to more than the time saved by not typing mappings manually so I wouldn’t consider including AutoMapper in a build again unless the system was very large and the vast majority of mappings could be done without custom mappers. Other reasons I’d choose manual mappings over AutoMapper include…
Run time errors as no compile time errors
Mappings are computed at runtime, so if your mapping object A to B and rename a property on one but not the other you won’t know until runtime. AssertConfigurationIsValid can however be called at startup or from a unit test to ensure you do indeed catch these issues at runtime.
No static analysis
Find all references won’t find references for many properties as they not explicitly used.
Long term support?
This isn’t specific to AutoMapper, but AutoMapper is mainly written by a talented guy called Jimmy Bogard. Although he has a bit of help, its still a pretty small project which may not be around or supported in a couple of years. Given that software in the industry I’m working in now (banking) often has a life time of 20 years this is a factor I needed to consider too, although again this is not specific to AM.
Recommendations for using AutoMapper
For me I no longer see the benefits of using it and I’d just rather not create a dependency on someone elses kit without it having validated benefits, but each to their own so if you intend on using it keep your life simple by:
Not using custom mappers as you end up writing almost as much code as you would to do things manually. Wrap standard Mapper.Map / source class -> destination code in an extension method (for example) and then do plain left-right assignment for anything that doesn’t map out of the box, this means your using less of the complex features of AutoMapper but still getting the majority of your properties mapped:
OrderDto dto = Mapper.Map<OrderDto>(order);
dto.CustomerFullName = order.FirstName + ‘ ‘ + order.LastName;
Call AssertConfigurationIsValid at startup when in development mode. You can’t get compile time errors but you can check for valid mappings immediately upon runtime. Keep an eye on how long this takes however.
Am guilty of doing this in the past myself but avoid using AutoMapper to go from ViewModels to entities as there is a lot of black box stuff going on which quite often doesn’t sit well with entity framework and causes unintended consequences such as triggering inserts rather than updates. Simply put your entities are too important to use AutoMapper on, do them manually…
FYI, not using AutoMapper is fine
In the not too distant past during some refactor time on an almost done project an external architect recommended to our team that we retrofit AutoMapper in after we had manually done all the left-right assignments without issue. Recommendation was based on the fact we were not using AutoMapper rather than on the need for a solution to a problem we were having (we were having none). There would have been absolutely no benefit only cost to using AutoMapper at that point. Point is with AutoMapper and other third party tools you should resist the urge to use them just because they are popular or the big books mention them, consider that everything has a cost and the choice to use something or not should depend on your project particulars not what is flavour of the month on StackOverflow or other developer hangouts online.
One thought on “AutoMapper doesn’t actually save much time… and other reasons not to use it”
I’ve just started a new job and am having to set something up using automapper. It feels like a whole load of extra work just to write some simple class mappings myself. I’m glad this article exists.