The review of the Switch library by Steven Watanabe will begin this Saturday, Jan 5th.
Description:
The built in C/C++ switch statement is very efficient. Unfortunately, unlike a chained if/else construct there is no easy way to use it when the number of cases depends on a template parameter. The Switch library addresses this issue.
If you are interested, please consider submitting a review of this library. I will send more detailed instructions on what to include in the review on the 5th.
Stjepan Rajko wrote: > If you are interested, please consider submitting a review of this > library. I will send more detailed instructions on what to include in > the review on the 5th.
Can I jump the gun with with a few questions please?
Why is the fall through behaviour to throw an exception? This is not consistent with standard switch, possibly a motivation section is needed in the docs to explain this type of decision?
I'm not too clear with the documentation what the parameter 'n' can be? It is of type 'Int' which is described as being an Integer in the parameters section of the documentation. What does this mean, any built in integral type, can I switch on a std::string? What are the limitations? I got a bit lost here in the docs here, after admittedly only a quick read. Maybe just some clarification is required?
The parameter section does not specify the arities of the 2 function object parameters, I had to deduce them from the example. Why do these have to be function objects, and not just any callable type, including ordinary functions (using the Boost.ResultOf mechanism to sort out the return type)? I'm not sure how useful that would be for 'f', but for 'default_' a full blow function object seems a bit much typing, can't I just use a function.
Also, why is 'default_' unary, the default in a standard switch statement does not access the switch parameter, and the example does not exploit this feature either. Why not just a nullary function? Again some motivation might be nice.
And finally, as the introduction mentions the efficiency of the standard switch statement, could the docs show some performance stats for the library code for comparison?
dan marsden wrote: > Why is the fall through behaviour to throw an exception?
I too tripped over this one, yesterday.
I think it shouldn't throw. If we want it to, we can use an approporiate function object.
The 'throw' statement (even if unreachable) can influence other optimizations. Further, without throwing the semantics remain the same if exceptions are disabled.
> I'm not too clear with the documentation what the parameter 'n' can be?
"MPL sequence of integer constants" and "equal to" are sorta fuzzy in this context, too.
Maybe stating that the "[...] the public static member constants 'value' [...] are used for the 'case' conditions" and that "'n' is used as the argument to the 'switch' statement" (or so) works to improve the specification.
> The parameter section does not specify the arities of the 2 function > object parameters, I had to deduce them from the example. Why do these > have to be function objects, and not just any callable type, including > ordinary functions (using the Boost.ResultOf mechanism to sort out the > return type)? I'm not sure how useful that would be for 'f', but for > 'default_' a full blow function object seems a bit much typing, can't > I just use a function.
I think "function object" is still the correct term even if you /can/ use a function pointer for the default: A pointer is an object and a function pointer is a function object. A pointer to member function is also Callable -- and we probably wouldn't want to support this case...
Of course there are no pointers to templates, so using a function pointer for anything but the default is pretty pointless. So is trying to handle varying result types -- maybe the result type should be passed in with another template parameter?
> Also, why is 'default_' unary, the default in a standard switch > statement does not access the switch parameter, and the example does > not exploit this feature either. Why not just a nullary function?
I really like this feature a lot:
If you can't react to some case at compile time doesn't mean you can't at runtime...
> Again some motivation might be nice.
> And finally, as the introduction mentions the efficiency of the > standard switch statement, could the docs show some performance > stats for the library code for comparison?
Yes, please!
Couldn't resist to throw some more Euro cents into the pot...
On Jan 2, 2008 2:36 PM, dan marsden <danmars...@yahoo.co.uk> wrote:
> Stjepan Rajko wrote: > > If you are interested, please consider submitting a review of this > > library. I will send more detailed instructions on what to include in > > the review on the 5th.
> Can I jump the gun with with a few questions please?
Of course - thanks for opening up the discussion! I'm just not sure whether Steven will be able to respond until the review officially starts.
> Why is the fall through behaviour to throw an exception? This is > not consistent with standard switch, possibly a motivation section is > needed in the docs to explain this type of decision?
> I'm not too clear with the documentation what the parameter 'n' can > be? It is of type 'Int' which is described as being an Integer in the > parameters section of the documentation. What does this mean, any > built in integral type, can I switch on a std::string? What are the > limitations? I got a bit lost here in the docs here, after admittedly only > a quick read. Maybe just some clarification is required?
> The parameter section does not specify the arities of the 2 function > object parameters, I had to deduce them from the example. Why do > these have to be function objects, and not just any callable type, > including ordinary functions (using the Boost.ResultOf mechanism to > sort out the return type)? I'm not sure how useful that would be for 'f', > but for 'default_' a full blow function object seems a bit much typing, > can't I just use a function.
> Also, why is 'default_' unary, the default in a standard switch > statement does not access the switch parameter, and the example > does not exploit this feature either. Why not just a nullary function? > Again some motivation might be nice.
> And finally, as the introduction mentions the efficiency of the standard > switch statement, could the docs show some performance stats for the > library code for comparison?
> dan marsden wrote: > > Why is the fall through behaviour to throw an exception?
> I too tripped over this one, yesterday.
> <snip>
The other option is an assertion. Is that any better? Or should a void return type be a special case?
> > I'm not too clear with the documentation what the parameter 'n' can be?
> "MPL sequence of integer constants" and "equal to" are sorta fuzzy in > this context, too.
> <snip>
OK. I'll tighten this up.
> > The parameter section does not specify the arities of the 2 function > > object parameters, I had to deduce them from the example.
I'll be more precise.
> > Why do these > > have to be function objects, and not just any callable type, including > > ordinary functions > > <snip>
You can use a plain function for default_.
> I think "function object" is still the correct term even if you /can/ > use a function pointer for the default: A pointer is an object and a > function pointer is a function object. A pointer to member function is > also Callable -- and we probably wouldn't want to support this case...
Even if it's specified as Callable, it's ok because void int::*() e.g. is illegal anyway.
> Of course there are no pointers to templates, so using a function > pointer for anything but the default is pretty pointless. So is trying > to handle varying result types -- maybe the result type should be passed > in with another template parameter?
I'd rather leave it as result_of<F()>::type.
> > Also, why is 'default_' unary, the default in a standard switch > > statement does not access the switch parameter, and the example does > > not exploit this feature either. Why not just a nullary function?
> I really like this feature a lot:
> If you can't react to some case at compile time doesn't mean you can't > at runtime...
Exactly.
> > And finally, as the introduction mentions the efficiency of the > > standard switch statement, could the docs show some performance > > stats for the library code for comparison?
>> dan marsden wrote: >>> Why is the fall through behaviour to throw an exception? >> I too tripped over this one, yesterday.
>> <snip> > The other option is an assertion. Is that any better?
I think the "default default behavior" should be to use the default constructor. This approach works nice with Any, Variant and Optional.
> Or should a void return type be a special case?
If 'Default's result type is 'void' although something else is expected, we should assert it never actually returns at runtime (note that it is possible to detect the result type with an overloaded comma operator and without requiring 'Default' to work with 'result_of' -- might also be faster). If assertions are disabled the behavior should be unspecified in that case.
<snip>
>> Of course there are no pointers to templates, so using a function >> pointer for anything but the default is pretty pointless. So is trying >> to handle varying result types -- maybe the result type should be passed >> in with another template parameter?
> I'd rather leave it as result_of<F()>::type.
Actually 'result_of<F()>::type' determines the result of the nullary call to F. I don't think I like this sort of "result_of abuse"... The correct usage would be 'result_of<F(MPLConstant)>::type' but it's pointless since 'MPLConstant' varies and so may the whole type expression.
So, if you insist on deducing the result type from the function object (instead of having it specified explicitly) my vote is 'F::result_type', however, I still find another template parameter more appropriate for the following good reasons:
o The function object can work fine with result_of in a non-'switch_' context. The cases can return different things as long as they are convertible to the result of 'switch_', and
o there is no way to determine this type with 'result_of.
> Steven Watanabe wrote: > > Tobias Schwinger <tschwinger <at> isonews2.com> writes: > >> Of course there are no pointers to templates, so using a function > >> pointer for anything but the default is pretty pointless. So is trying > >> to handle varying result types -- maybe the result type should be passed > >> in with another template parameter?
> > I'd rather leave it as result_of<F()>::type.
> Actually 'result_of<F()>::type' determines the result of the nullary > call to F. I don't think I like this sort of "result_of abuse"... The > correct usage would be 'result_of<F(MPLConstant)>::type' but it's > pointless since 'MPLConstant' varies and so may the whole type expression.
> So, if you insist on deducing the result type from the function object > (instead of having it specified explicitly) my vote is 'F::result_type', > however, I still find another template parameter more appropriate for > the following good reasons:
> o The function object can work fine with result_of in a non-'switch_' > context. The cases can return different things as long as they are > convertible to the result of 'switch_', and
> o there is no way to determine this type with 'result_of.
The current implementation seems to use result_type - is it planned to change to use result_of?
I agree that result_of<F()>::type is slightly abusive, since that's not what actually gets called. Would using result_of<F(boost::mpl::front<Cases>::type)> be an option for a non-empty case sequence? As long as the order of the cases doesn't matter (btw, does it?), the user could put the desired type in the front of the Cases sequence if the return type differs for different MPLConstant types.
Stjepan Rajko wrote: > On Jan 6, 2008 7:34 AM, Tobias Schwinger <tschwin...@isonews2.com> wrote: >> Steven Watanabe wrote: >>> Tobias Schwinger <tschwinger <at> isonews2.com> writes: >>>> Of course there are no pointers to templates, so using a function >>>> pointer for anything but the default is pretty pointless. So is trying >>>> to handle varying result types -- maybe the result type should be passed >>>> in with another template parameter? >>> I'd rather leave it as result_of<F()>::type. >> Actually 'result_of<F()>::type' determines the result of the nullary >> call to F. I don't think I like this sort of "result_of abuse"... The >> correct usage would be 'result_of<F(MPLConstant)>::type' but it's >> pointless since 'MPLConstant' varies and so may the whole type expression.
>> So, if you insist on deducing the result type from the function object >> (instead of having it specified explicitly) my vote is 'F::result_type', >> however, I still find another template parameter more appropriate for >> the following good reasons:
>> o The function object can work fine with result_of in a non-'switch_' >> context. The cases can return different things as long as they are >> convertible to the result of 'switch_', and
>> o there is no way to determine this type with 'result_of.
Actually, the latter is more severe...
> The current implementation seems to use result_type - is it planned to > change to use result_of?
> I agree that result_of<F()>::type is slightly abusive, since that's > not what actually gets called. > Would using > result_of<F(boost::mpl::front<Cases>::type)> be an option for a > non-empty case sequence?
I think it's too complicated: We can't use 'result_of' to determine the result of 'switch_', so it should be as simple to determine as possible (ideally without deduction at all).
> As long as the order of the cases doesn't > matter (btw, does it?), the user could put the desired type in the > front of the Cases sequence if the return type differs for different > MPLConstant types.
Further, we still need a special-engineered function object; one of the cases will have a special role. It might work, but it feels inelegant to me: The function object's result type should be convertible to whatever 'switch_' wants to return.
So what will deducing that type from the function object buy us?
The only answer I can currently see is "nothing but trouble" :-). Please tell me if I'm missing something.
On Jan 6, 2008 12:20 PM, Tobias Schwinger <tschwin...@isonews2.com> wrote:
> Stjepan Rajko wrote: > > Would using > > result_of<F(boost::mpl::front<Cases>::type)> be an option for a > > non-empty case sequence?
> I think it's too complicated: We can't use 'result_of' to determine the > result of 'switch_', so it should be as simple to determine as possible > (ideally without deduction at all).
Sure - although the library could offer something along the lines of result_of::switch.
> > As long as the order of the cases doesn't > > matter (btw, does it?), the user could put the desired type in the > > front of the Cases sequence if the return type differs for different > > MPLConstant types.
> Further, we still need a special-engineered function object; one of the > cases will have a special role. It might work, but it feels inelegant to > me: The function object's result type should be convertible to whatever > 'switch_' wants to return.
I agree that being able to specify the return type explicitly would be very useful, precisely because you could use function objects that are not result_of/result_type compatible, or override the return type even if one is specified by the function object. I can see overriding it to boost::any or some other "common denominator" as a frequent use case.
> So what will deducing that type from the function object buy us?
> The only answer I can currently see is "nothing but trouble" :-). Please > tell me if I'm missing something.
>From what I can see, it buys simplicity when the use case is not
complicated (the return type is available through result_type or result_of and does not change), or when you really want to leave it to the function object to specify what the return type should be. Granted, all of the proposed solutions for return type deduction seem slightly imperfect / inelegant, but as long as the behavior is clearly explained in the documentation they could be useful.
Tobias Schwinger wrote: > Stjepan Rajko wrote: >>> o there is no way to determine this type with 'result_of.
> Actually, the latter is more severe...
>> The current implementation seems to use result_type - is it planned to >> change to use result_of?
>> I agree that result_of<F()>::type is slightly abusive, since that's >> not what actually gets called.
>> Would using >> result_of<F(boost::mpl::front<Cases>::type)> be an option for a >> non-empty case sequence?
> I think it's too complicated: We can't use 'result_of' to determine the > result of 'switch_', so it should be as simple to determine as possible > (ideally without deduction at all).
>> As long as the order of the cases doesn't >> matter (btw, does it?), the user could put the desired type in the >> front of the Cases sequence if the return type differs for different >> MPLConstant types.
> Further, we still need a special-engineered function object; one of the > cases will have a special role. It might work, but it feels inelegant to > me: The function object's result type should be convertible to whatever > 'switch_' wants to return.
> So what will deducing that type from the function object buy us?
> The only answer I can currently see is "nothing but trouble" :-). Please > tell me if I'm missing something.
I haven't been following the review (yet, though I'll be in the next few days) and I haven't read the docs yet. But from intuition, having implemented at least 2 switch implementations (spirit and phoenix), I'm guessing that there are N functions with N return types, right? If so, it follows (to my mind) that the result should be a boost::variant of all the possible return types given the arguments (through application of result_of to all the functions).
Joel de Guzman wrote: > Tobias Schwinger wrote: >> Stjepan Rajko wrote:
>>>> o there is no way to determine this type with 'result_of.
>> Actually, the latter is more severe...
>>> The current implementation seems to use result_type - is it planned to >>> change to use result_of?
>>> I agree that result_of<F()>::type is slightly abusive, since that's >>> not what actually gets called. >>> Would using >>> result_of<F(boost::mpl::front<Cases>::type)> be an option for a >>> non-empty case sequence? >> I think it's too complicated: We can't use 'result_of' to determine the >> result of 'switch_', so it should be as simple to determine as possible >> (ideally without deduction at all).
>>> As long as the order of the cases doesn't >>> matter (btw, does it?), the user could put the desired type in the >>> front of the Cases sequence if the return type differs for different >>> MPLConstant types. >> Further, we still need a special-engineered function object; one of the >> cases will have a special role. It might work, but it feels inelegant to >> me: The function object's result type should be convertible to whatever >> 'switch_' wants to return.
>> So what will deducing that type from the function object buy us?
>> The only answer I can currently see is "nothing but trouble" :-). Please >> tell me if I'm missing something.
> I haven't been following the review (yet, though I'll be in the next > few days) and I haven't read the docs yet. But from intuition, having > implemented at least 2 switch implementations (spirit and phoenix), > I'm guessing that there are N functions with N return types, right? > If so, it follows (to my mind) that the result should be a boost::variant > of all the possible return types given the arguments (through > application of result_of to all the functions).
What are we going to do with that Variant? Switch again and go back where we're coming from? It might but isn't necessarily what we want.
Now in order to get our Variant we have to o check result_of<Fn(arg)>::type for every function, o figure out unique types, and o make a variant from them, possibly handling special cases like using Boost.Optional or the bare type where more appropriate.
It's surely useful for certain use cases but takes a lot of template instantiations, so I think it should be an extra metafunction.
Tobias Schwinger wrote: > Joel de Guzman wrote: >> Tobias Schwinger wrote: >>> Stjepan Rajko wrote: >>>>> o there is no way to determine this type with 'result_of.
>>> Actually, the latter is more severe...
>>>> The current implementation seems to use result_type - is it planned to >>>> change to use result_of?
>>>> I agree that result_of<F()>::type is slightly abusive, since that's >>>> not what actually gets called. >>>> Would using >>>> result_of<F(boost::mpl::front<Cases>::type)> be an option for a >>>> non-empty case sequence? >>> I think it's too complicated: We can't use 'result_of' to determine the >>> result of 'switch_', so it should be as simple to determine as possible >>> (ideally without deduction at all).
>>>> As long as the order of the cases doesn't >>>> matter (btw, does it?), the user could put the desired type in the >>>> front of the Cases sequence if the return type differs for different >>>> MPLConstant types. >>> Further, we still need a special-engineered function object; one of the >>> cases will have a special role. It might work, but it feels inelegant to >>> me: The function object's result type should be convertible to whatever >>> 'switch_' wants to return.
>>> So what will deducing that type from the function object buy us?
>>> The only answer I can currently see is "nothing but trouble" :-). Please >>> tell me if I'm missing something. >> I haven't been following the review (yet, though I'll be in the next >> few days) and I haven't read the docs yet. But from intuition, having >> implemented at least 2 switch implementations (spirit and phoenix), >> I'm guessing that there are N functions with N return types, right? >> If so, it follows (to my mind) that the result should be a boost::variant >> of all the possible return types given the arguments (through >> application of result_of to all the functions).
> What are we going to do with that Variant? Switch again and go back > where we're coming from?
Use the variant visitor. That may amount to the same thing, but in this case, we have a value, not a function that returns a value.
> It might but isn't necessarily what we want.
Why not?
> Now in order to get our Variant we have to > o check result_of<Fn(arg)>::type for every function, > o figure out unique types, and > o make a variant from them, possibly handling special cases like using > Boost.Optional or the bare type where more appropriate.
> It's surely useful for certain use cases but takes a lot of template > instantiations, so I think it should be an extra metafunction.
Sorry, perhaps I missed the "extra metafunction" solution. What is it again?
Anyway, you have good points. Yes, an explicitly specified return type would be ok too. In the standpoint of optimization, yes, a variant return is suboptimal at both compile time and at runtime.
Joel de Guzman wrote: > Tobias Schwinger wrote: >> Joel de Guzman wrote: >>> Tobias Schwinger wrote: >>>> Stjepan Rajko wrote: >>>>>> o there is no way to determine this type with 'result_of.
>>>> Actually, the latter is more severe...
>>>>> The current implementation seems to use result_type - is it planned to >>>>> change to use result_of?
>>>>> I agree that result_of<F()>::type is slightly abusive, since that's >>>>> not what actually gets called. >>>>> Would using >>>>> result_of<F(boost::mpl::front<Cases>::type)> be an option for a >>>>> non-empty case sequence? >>>> I think it's too complicated: We can't use 'result_of' to determine the >>>> result of 'switch_', so it should be as simple to determine as possible >>>> (ideally without deduction at all).
>>>>> As long as the order of the cases doesn't >>>>> matter (btw, does it?), the user could put the desired type in the >>>>> front of the Cases sequence if the return type differs for different >>>>> MPLConstant types. >>>> Further, we still need a special-engineered function object; one of the >>>> cases will have a special role. It might work, but it feels inelegant to >>>> me: The function object's result type should be convertible to whatever >>>> 'switch_' wants to return.
>>>> So what will deducing that type from the function object buy us?
>>>> The only answer I can currently see is "nothing but trouble" :-). Please >>>> tell me if I'm missing something. >>> I haven't been following the review (yet, though I'll be in the next >>> few days) and I haven't read the docs yet. But from intuition, having >>> implemented at least 2 switch implementations (spirit and phoenix), >>> I'm guessing that there are N functions with N return types, right? >>> If so, it follows (to my mind) that the result should be a boost::variant >>> of all the possible return types given the arguments (through >>> application of result_of to all the functions). >> What are we going to do with that Variant? Switch again and go back >> where we're coming from?
> Use the variant visitor. That may amount to the same thing, but > in this case, we have a value, not a function that returns a value.
>> It might but isn't necessarily what we want.
> Why not?
Because we're just coming from inside the switch and if we wanted to do something conditionally we could have done it there. Variant, Any, Optional or some non-union type can all make sense, depending on what we're up to.
>> Now in order to get our Variant we have to >> o check result_of<Fn(arg)>::type for every function, >> o figure out unique types, and >> o make a variant from them, possibly handling special cases like using >> Boost.Optional or the bare type where more appropriate.
>> It's surely useful for certain use cases but takes a lot of template >> instantiations, so I think it should be an extra metafunction.
> Sorry, perhaps I missed the "extra metafunction" solution. > What is it again?
Ooops -- the English word "extra" might not work quite the way I expected (see below for clarification).
> Anyway, you have good points. Yes, an explicitly specified return > type would be ok too. In the standpoint of optimization, yes, a > variant return is suboptimal at both compile time and at runtime.
Exactly. The mechanism you propose could still be useful and provided by the library as a utilty metafunction (which is what I was trying to say with "should be an extra metafunction" :-)).
Tobias Schwinger wrote: > Joel de Guzman wrote: >> Anyway, you have good points. Yes, an explicitly specified return >> type would be ok too. In the standpoint of optimization, yes, a >> variant return is suboptimal at both compile time and at runtime.
> Exactly. The mechanism you propose could still be useful and provided by > the library as a utilty metafunction (which is what I was trying to say > with "should be an extra metafunction" :-)).
Tobias Schwinger wrote: > Joel de Guzman wrote: >>> What are we going to do with that Variant? Switch again and go back >>> where we're coming from? >> Use the variant visitor. That may amount to the same thing, but >> in this case, we have a value, not a function that returns a value.
>>> It might but isn't necessarily what we want. >> Why not?
> Because we're just coming from inside the switch and if we wanted to do > something conditionally we could have done it there. Variant, Any, > Optional or some non-union type can all make sense, depending on what > we're up to.
This is not always the case. Many times, you don't supply the functions or have control over them. That's specially true if you are a library writer, you know that ;-)
Anyway, I agree with you, not in the standpoint of what is a better API, but in the standpoint of practicality. Many languages do switch the /variant/ way. But in our case, the variant will indeed be a burden that might not be needed. So, I am for "you pay only for what you need".
Joel de Guzman wrote: > Anyway, I agree with you, not in the standpoint of what is > a better API, but in the standpoint of practicality.
Yep. I can follow your perspective: The only logical outcome of a switch operation is in fact a union type.
Besides the efficiency aspect, the devil is in the details: The default constructor's semantics of Variant, the trade-off between 'variant' and 'any', and syntactic inconsistency vs. clutter in case there's just one type...
> Many languages do switch the /variant/ way.
In a lot of programming languages (especially dynamically typed ones) variants are a much more of a natural thing than they are in C++. What you call "the variant way" (or should I say the "optional variant way") is often just "the way variables work".
Perl did not even have a switch statement until Version 5.10 -- you'd simply "bend the syntax to switch":
|
