|
I also subscribe to the struct approach (of course you need to determine the break even point between how many parameters you have before it is worth wrapping them in a struct), but am going to go a little further. I have projects where the function I wish to call may be several layers deeper or I need to add a new request from a client UI and direct a call to a new function on the server - I know you have all been there too and know what I am talking about.
In some of my projects, I have a generic message class that can be XML'ified (, compressed, encrypted) and transmitted between server and client. The class has a member variable, which is a list of name/value pairs and this can be used directly for most parameters. When adding new stuff, only the sender and receiver function need to know about the parameters and all layers in between just work without being updated. For more advanced parameters, such as structs, the class has a string member variable that can be an XML string containing any kind of object that has been serialized to XML.
That whole core is kind of old and would probably benefit from being updated to JSON.
If extreme performance is an issue, I avoid the whole back and forth string conversion and use a different mechanism, but for normal applications the processing time is plenty fast enough and the savings in maintenance when you need to add a new parameter is well worth it.
Soren Madsen
"When you don't know what you're doing it's best to do it quickly" - Jase #DuckDynasty
|
|
|
|
|
In most cases I consider a large number of parameters (let say over 8) as design flaw, however it is possible that you end with such a list, in which case I would use struct/class...
I'm not questioning your powers of observation; I'm merely remarking upon the paradox of asking a masked man who he is. (V)
|
|
|
|
|
In the old days, I was working with a Fortran compilers where we had to push the release due to one (important) customer who had run into a limitation of the previous one: It could take only 99 (ninety-nine) parameters. The new version could handle 128 parameters, which was sufficient for the immediate needs of the customer, but the design allowed expansion to 256 parameters.
If there ever was a case for Fortran COMMON blocks, I would call this a candidate.
(Don't take me wrong - I am not suggesting using COMMON blocks as good programming practice. Nor is 100+ function parameters.)
|
|
|
|
|
|
COMMON is certainly good "programming practice" as long as each COMMON is named and contains a single value. Otherwise I submit that COMMON should be found non-complying in coding standards.
Gus Gustafson
|
|
|
|
|
When Dealing with Microsoft related programming products creating your procedures with the parameters (Sender as object, e as eventArgs) as an extremely recommended practice, and have your parameter class inherit from system.eventArgs.
This way your code naturally interacts with windows runtime components.
|
|
|
|
|
Except that you have created event arguments for use with none events. The sender is probably going to be completely useless.
|
|
|
|
|
You do understand that we are working with event driven programming right? The entire inner working of the object programming model has a point of origin and the event that triggered it.
You should always be making your code generic as it can be understood by everything, and your procedure should be tread safe - which means everything you need should be passed as a parameter.
|
|
|
|
|
Colborne_Greg wrote: You do understand that we are working with event driven programming right?
That's a very narrow point of view, focused on one technology set. At no stage did Marc state he was asking about an event. And while there is a point of origin, there's no guarantee that he is responding to an event - you may have noticed that Marc was talking in a generic sense, hence the reason he mentioned ROR. So, in the case of doing something like processing a batch input on a file, forcing the programming model to use an event signature doesn't really make sense.Colborne_Greg wrote: You should always be making your code generic as it can be understood by everything, and your procedure should be tread safe - which means everything you need should be passed as a parameter.
And that is why my response, at the top of this thread, stated that I would normally wrap things up into an object.
|
|
|
|
|
When the procedure was called is an event
when the processing a batch input on a file BOF reached and EOF reached are events.
Everything is an event, and inheriting from eventargs allows your code to be repurposed everywhere, and when all your delegates have the same signature you wont get held back when you want to dive deeper into generics.
|
|
|
|
|
Seriously, everything isn't an event. Not all environments are event based, and for those that are, creating events for things that don't need to be events is an unnecessary overhead. We might as well stop now because I'm not going to convince you and you aren't going to convince me.
|
|
|
|
|
You are thinking to hard.
Kiss - keep it simple stupid.
Everything is an event whether you are programming with events, responding to events, or looking for events in code via a conditional statement, an event is a English word that gives abstract context to meaning.
Its like arguing that a date/time, or decimal is not a string - well yes those are strings and are processed differently, which is the opposite to generics - inherit from eventargs whether or not you are responding to an actual event so that all your delegates have the same signature, then when you get into the thinking of what an actual event is your code will become cleaner.
|
|
|
|
|
I suppose the difference between you and me is that I have experience in none event driven systems, as well as in none .NET platforms so my first instinct is not to always make things an event. They aren't suitable for every platform. So perhaps you need to think a little harder and get out of the .NET bubble. It'll open your mind.
|
|
|
|
|
-_- I am a cobol programmer, I also started my statement with... when dealing with Microsoft related products which is the dot net bubble...
|
|
|
|
|
Also you can send a class object as the sender for the processing to interact with.
In your example of the processing of the batch input, the result can be given in the same procedure to the sender to be displayed.
|
|
|
|
|
I'm not sure that I've ever needed to do that other than in cases similar to database connection strings -- like a factory of some sort -- and then XML works. 
@"<options option1='true' option2='foobar' option3='42' />"
Obviously, the problem with "but you have absolutely no clue what these optional parameters are unless you look up the online documentation (if it exists)" is taken to a whole new level. 
You'll never get very far if all you do is follow instructions.
|
|
|
|
|
I think I have some internal, arbitrary rule that says;
If the parameters are basically unrelated except for being passed to that function, then leave them as individual arguments, well defined by their names.
If the parameters are related, then wrap them in some structure or other.
In other words, I wouldn't wrap parameters in a class or struct just to make the function signature smaller or more legible, but I would do so if it made sense for the parameters to be combined;
The danger of wrapping them is that you replace a long function call with a complex struct creation, to no real advantage - reading a method signature with meaningful argument names is self-documentation; the same method with a single argument called "necessaryInfoToDoTheJob is less descriptive!
|
|
|
|
|
|
You preferred way is also the Windows API one: on MSDN you may find a neat, short page about the function, the big mess being on the passed struct.
Veni, vidi, vici.
|
|
|
|
|
We have been using the struct practice for years, and I am personally strongly in favor of it. I would like to add a few comments/modifications, though:
First, a set of options is not a function declaration issue, but a system data design issue. Two or more functions referring to, say, page layout properties should not declare "their own" parameter structs (possibly overlooking some essential parameter). The set of options affecting page layout, say, is one well defined set for the entirre application (or even more).
Second, you should never slump together completely unrelated options in one struct, even if one function (or even several) inspects them all. One option struct defines page layout parameters, another one typographical characteristics, a third one the current user. So you might end up with "several" (i.e. a few) option parameters, but not a hundred of them. (For all practical purposes, this second point is also a requirement for my first point.)
Third, as software develops, new options will be added. For an exported library function, it must be prepared to handle calls from applications both newer (supplying a larger struct with added and unknown fields) and older (supplying structs with missing fields) applications, and be able to handle them both. So the struct must identify the version. So one of the fields in the option struct is "This is format 3". An alternative is "This struct contains 44 valid bytes of parameters. The very best is to include both: This struct contains 44 bytes of parameters of format 3" - then you can add parameters in format 3 as long as the extensions are fully compatible, and bump the format code only when an incompatible extension is introduced. So the caller must fill in two extra fields, but then again, the same declaration can be used twenty years later. (Don't expect anyone below thirty to see the value of that...)
Another alternativ is of course the Win32 API way: Start with MyFunction. Then, when one option is added, call it MyFunctionEx. After the second extension, make it MyFunctionEx. After extension eight, it is MyFunctionExExExExExExExEx, and so it continues 
|
|
|
|
|
I think it's important that there's an easy way to find which options are defined. I've had issues with Javascript libraries that take an object as a parameter, and expect you to put the right properties in there (the XMLHttpRequest AJAX handler does this iirc). In a statically typed language like C# or Java (or even C, Delphi, C++ etc), it makes sense to use a struct or class (using those words in their C# meaning) for groups of parameters with values, and a flagwise enum for those without. If you don't have enums then the old school approach of constants set to the appropriate bit values and using | to combine them is fine too.
The grouping is important, it shouldn't just be public struct StuffUsedByThisFunction, it should have some semantic meaning even outwith the context of the function call.
In dynamic languages the convention seems to be dictionaries with lax validation. I agree that this is often unhelpful, but it's encouraged by the object model in those languages which doesn't really do static declaration of valid members. I've not really worked in those languages enough to have developed a good solution, though.
|
|
|
|
|
|
Into a function? I don't even see how that makes sense.
Generally though I hate dependency injection. It makes it really really hard to find out what type of object you're actually dealing with when you're looking at the code.
|
|
|
|
|
Parameters?
Why not just use Global Variables???
Wait for it...
ROTFLMAO... I did flag this as a JOKE!
That is a HORRIBLE IDEA, for the record...
And for the record, we used a structure, and it worked beautifully.
The tricky part is when you call a function that needs 1 of those parameters... Do you pass the structure, or struct.val3 ? (we based it on the completeness of the function. If we felt it wasn't changing, then we sent as little data as possible, but more than one field meant the structure).
|
|
|
|
|
If the function has "too many" arguments, accept that it's ugly and live with it. Putting them in a struct is just hiding the ugliness and moving the complexity elsewhere, possibly remote from the function call itself. In my mind, it just makes for extra work to understand and maintain.
The optional argument example is a perfect time to use an overload. That way you have a place to document the unique behavior of that variation of the function, and have documented for the next developer which arguments need to be specified as a set.
The struct solution isn't a bad way to hide the problem, although a polymorphic class might be better depending on the needs.
We can program with only 1's, but if all you've got are zeros, you've got nothing.
|
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
![Beer | [beer]](https://www.codeproject.com/script/Forums/Images/beer.gif)
on me.
