on goto
in [ASM]
|
From: Juha Nieminen on 6 May 2010 13:02 In comp.lang.c++ Nathan <nathancbaker(a)gmail.com> wrote: > Well, 'y < x' must evaluate as 'true' in order for something to be > classified as simpler. So, because *one loop* < *three loops*, my > version certainly looks simpler. [e.g. less opcodes in the generated > binary ] "Simpler" does not mean "produces a smaller executable file" or "has less loops". "Simpler" means "easier to understand". Your version may have less loops, but you have replaced them with complex conditional statements, which are unusual and useless (because they don't do anything the loops weren't doing already). (Ironically, your version is also less efficient because it performs all three index range checks at each iteration, while the nested loop version does only one index range check for the innermost iteration.) >> More importantly, it can dereference index [0] before checking size(), >> so produces undefined behaviour. �Anyway, IMHO it's far less clear >> (=self-evidently correct & efficient as well as maintainable) than >> Juha's code. >> > > More maintainable by the original coder... or more maintainable by > someone 'new' to the code??? If this function were many screens in > length, and someone 'new' decides to have it perform an extra task > 'just' before returning, wouldn't that person have a "devil of a time" > debugging the program if he were not aware of the hidden alternative > 'return' route tucked-away in that nest? Even if you had an understandability problem with a "hidden return", the correct solution is not to make the whole thing an unusual and hard to understand construct which is probably even less efficient. > Yes, rather than waste one's time bickering about the various esteemed > implementation alternatives of a particular algo, why not consider an > algo that renders those salient points mute? Because the whole point of this thread is "how to exit nested loops most cleanly?", not "what is the best way of doing task X?"
From: Nick on 6 May 2010 14:06 Juha Nieminen <nospam(a)thanks.invalid> writes: > In comp.lang.c++ Nathan <nathancbaker(a)gmail.com> wrote: >> Well, 'y < x' must evaluate as 'true' in order for something to be >> classified as simpler. So, because *one loop* < *three loops*, my >> version certainly looks simpler. [e.g. less opcodes in the generated >> binary ] > > "Simpler" does not mean "produces a smaller executable file" or "has > less loops". "Simpler" means "easier to understand". > > Your version may have less loops, but you have replaced them with > complex conditional statements, which are unusual and useless (because > they don't do anything the loops weren't doing already). > > (Ironically, your version is also less efficient because it performs > all three index range checks at each iteration, while the nested loop > version does only one index range check for the innermost iteration.) Particularly if that "size()" function involves a counting loop rather than a value lookup. -- Online waterways route planner | http://canalplan.eu Plan trips, see photos, check facilities | http://canalplan.org.uk
From: Daniel T. on 7 May 2010 00:18 Keith Thompson <kst-u(a)mib.org> wrote: > Nathan <nathancbaker(a)gmail.com> writes: > > > ,--- > > result = value > > xInd = 0 > > yInd = 0 > > zInd = 0 > > while ( data[xInd][yInd][zInd] != value && result != 0) > > { > > ++zInd; > > if (( zInd < data[xInd][yInd].size() ) != true ) { zInd = 0; + > > +yInd}; > > if (( yInd < data[xInd].size() ) != true ) { yInd = 0; ++xInd}; > > if (( xInd < data.size() ) != true ) { result = 0 }; > > } > > return result > > `--- > [...] > > A style point: comparisons to true and false are almost always > superfluous. Rather than > if (( xInd < data.size() ) != true ) > just write > if ( xInd <= data.size() ) > > As for the overall structure of your proposed replacement, you've > replaced three nested loops with one loop and three if statements. > Furthermore, the original version only tests zInd on most iterations; > your version tests zInd, yInd, and xInd on every iteration. However, his version could easily be modified so it didn't make the extra tests... As you probably know. > The original problem is to traverse a 3-dimensional array. A triple > nested loop is the most obvious way to do that. There might be > some advantages in converting it to a single loop, but clarity > isn't one of them, at least in this case. I think there is some confusion here, the original problem is to return the first element in the container that equals a particular value. Using three loops to iterate through one container is unnecessary. i.e., iterator result = data.begin(); while (result != data.end() && *result != value) ++result; return result; A solution very much like the above could even be done in C on an actual three dimensional array...
From: Daniel T. on 7 May 2010 00:23 In article <7ec1a55b-4a14-46c3-bb65-dd5de1aa370b(a)n15g2000yqf.googlegroups.com>, tonydee <tony_in_da_uk(a)yahoo.co.uk> wrote: > On May 6, 2:36�pm, Nathan <nathancba...(a)gmail.com> wrote: > > On Apr 28, 1:16�am, Juha Nieminen <nos...(a)thanks.invalid> wrote: > > > � Care to show an actual example of your "simpler, cleaner and easier to > > > follow" version of exiting a nested loop by meddling with the loop > > > conditions instead of using 'return'? For example, modify the following > > > code to conform to your specifications: > > > > > Value_t* MyClass::findValue(const Value_t& value) > > > { > > > � � for(size_t xInd = 0; xInd < data.size(); ++xInd) > > > � � � � for(size_t yInd = 0; yInd < data[xInd].size(); ++yInd) > > > � � � � � � for(size_t zInd = 0; zInd < data[xInd][yInd].size(); ++zInd) > > > � � � � � � { > > > � � � � � � � � if(data[xInd][yInd][zInd] == value) > > > � � � � � � � � � � return &data[xInd][yInd][zInd]; > > > � � � � � � } > > > � � return 0; > > > } > > > > Hi Juha, > > > > I think that you either copied this from a poorly-written beginner C++ > > book or you failed to understand what the author was attempting to > > demonstrate with that kind of code and that you did not 'catch on' > > that it is not (in any way) demonstrative of how one walks a series of > > sequential cells in the real world. > > > > Why do you insist on writing control structures that serve little > > actual purpose? �For instance, it should be obvious that you only > > require ONE loop and a few conditionals to achieve your desired (or > > assumed) goal. �E.G... > > > > ,--- > > result = value > > xInd = 0 > > yInd = 0 > > zInd = 0 > > while ( data[xInd][yInd][zInd] != value && result != 0) > > { > > � � ++zInd; > > � � if (( zInd < data[xInd][yInd].size() ) != true ) { zInd = 0; + > > +yInd}; > > � � if (( yInd < data[xInd].size() ) != true ) { yInd = 0; ++xInd}; > > � � if (( xInd < data.size() ) != true ) { result = 0 };} > > > > return result > > `--- > > Juha did ask for something "simpler, cleaner and easier to follow", > which your single loop is not. Perhaps because of that, your code > needlessly compares yInd and xInd to their respective limits on every > iteration, rather than something like... > > while ( data[xInd][yInd][zInd] != value && result != 0) > { > if {++zInd == data[xInd][yInd].size()) { > zInd = 0; > if (++yInd == data[xInd].size()) { > yInd = 0; > if (++xInd == data.size()) > result = 0; > } > } > ... > > More importantly, it can dereference index [0] before checking size(), > so produces undefined behaviour. Anyway, IMHO it's far less clear > (=self-evidently correct & efficient as well as maintainable) than > Juha's code. > > > Of course, this is a rather useless (maybe even retarded) function to > > begin with, because it only tells you IF the 'value' is located > > "somewhere" within that array -- it gives you absolutely no indication > > of "where" in that array you might be able to access the item which > > matches your search criteria. > > You missed that the original function was returning a pointer to the > matching cell, allowing a change to be made at that location, > considerably more useful than the search-term-else-0-sentinel version > you coded. > > > Please pardon any butchering of C++ syntax in my pseudo snippet above > > -- I totally lack any C++ training. > > Fair enough... no worries. > > >�But we alt.lang.asm folk *do* > > have an inkling of how to actually code our way out of a paper bag... > > I do believe. > > > > Nathan. > > Your central point that a single loop can serve contributes an > interesting alternative, which I'm sure the readers here will > appreciate, if they bother to look past your smug attitude (which is > quite unwarranted given the serious errors in your implementation).... I posted a single loop alternative last week or so. I think it is a much better solution than Juha's original.
From: Daniel T. on 7 May 2010 00:26
Nathan <nathancbaker(a)gmail.com> wrote: > On May 6, 2:11�am, tonydee <tony_in_da...(a)yahoo.co.uk> wrote: > > > > Juha did ask for something "simpler, cleaner and easier to follow", > > which your single loop is not. �Perhaps because of that, your code > > needlessly compares yInd and xInd to their respective limits on every > > iteration, rather than something like... > > > > � while ( data[xInd][yInd][zInd] != value && result != 0) > > � { > > � � � if {++zInd == data[xInd][yInd].size()) { > > � � � � � zInd = 0; > > � � � � � if (++yInd == data[xInd].size()) { > > � � � � � � � yInd = 0; > > � � � � � � � if (++xInd == data.size()) > > � � � � � � � � � result = 0; > > � � � � � } > > � � � } > > � ... > > > > Isn't <something>.size() a method or function call? We'd also want to > eliminate that needless activity from the loop by defining "zMax = > data[xInd][yInd].size()", and etc., before the loop. > > Just because C++ gifts you with 'high-level' abstractions, that > doesn't confer an automatic excuse to create inefficient, > unmaintainable code. It is precisely the fact that Juha's code doesn't use high level abstractions that is the problem. See my earlier posts on this sub-thread. |