C program flow control - Very Basic Question.
in [Embedded]
|
Prev: ready to run 32bit controller
Next: ARM7
From: D Yuniskis on 2 Jun 2010 16:15 Hi Bob, Bob wrote: > Back before C++ exceptions, I wrapped setjump/longjump in a struct w/ > heap cleanup func pointers to handler errors that just wouldn't > reasonably work any other way (floating point and floppy disk errors). I > treat C++ exceptions just as warily. If the problem can be handled with > error returns, then you don't need exceptions/longjump. I've only used setjmp/longjmp to implement simulators and coroutines. Other than that, I think your "error returns" comment applies best! (i.e., make the problem *visible*)
From: Dennis on 2 Jun 2010 17:24 Tim Wescott wrote: > But used carefully and sparingly, "goto" in C can provide a very good > mechanism for exception handling. I use it occasionally, and it does > well for what it is. Were I to see a "goto" in a code review I would > pin the author of the code against the wall to make sure he was using it > correctly; were I to enter some of my code with a "goto" in a code > review I would schedule extra time to be pinned to the wall and justify > myself (I'd probably have careful comments in the code, too). > In our code reviews a goto had to be either a usage that the team had agreed on (very few - usually related to exception cases) or you had to defend the usage. > Exception handling in object-oriented programming is specifically > designed to address the problems in C++ that C's "goto" and long branch > (long goto? I can't remember what it's called) solve in C. > When we went to C++ most of the error case gotos went away with well designed destructors (to release resources) and throw() of an exception.
From: Dennis on 2 Jun 2010 17:37 David Brown wrote: > There are times when "goto" is the best way to structure your code - > I've used it a few times. You have to be careful with it, as it's easy > to write spaghetti code and lose track of things like variable > initialisations (C++ is fussier about when you are allowed to use it). > But setjmp() and longjmp() are far worse - if you think you need them, > you are almost certainly wrong and should re-structure your code. (If > you /know/ you need them, it's a different matter - few rules in C > programming are absolute.) One of the incentives to move to C++ was the nested setjmp() and longjmp() in the existing C code (yes setjmp() and longjmp() was the cleanest solution in C). With matching constructors and destructors in C++ it was much easier to manage resources automatically on a throw() in the exception case. Of course C++ takes some careful design up front.
From: Tim Wescott on 2 Jun 2010 19:01 On 06/02/2010 10:01 AM, D Yuniskis wrote: > Hi Tim, > > Tim Wescott wrote: >> Exception handling in object-oriented programming is specifically >> designed to address the problems in C++ that C's "goto" and long >> branch (long goto? I can't remember what it's called) solve in C. > > setjmp()/longjmp()? That's it. The more you start doing "C++ish" things the more dangerous it gets -- but it's kewl for (very, very carefully constructed) library code-like things. -- Tim Wescott Control system and signal processing consulting www.wescottdesign.com
From: Tim Wescott on 2 Jun 2010 19:03
On 06/02/2010 02:24 PM, Dennis wrote: > Tim Wescott wrote: > >> But used carefully and sparingly, "goto" in C can provide a very good >> mechanism for exception handling. I use it occasionally, and it does >> well for what it is. Were I to see a "goto" in a code review I would >> pin the author of the code against the wall to make sure he was using >> it correctly; were I to enter some of my code with a "goto" in a code >> review I would schedule extra time to be pinned to the wall and >> justify myself (I'd probably have careful comments in the code, too). >> > In our code reviews a goto had to be either a usage that the team had > agreed on (very few - usually related to exception cases) or you had to > defend the usage. Yup. And if you're smart you take your code to one or two of the folks you know are going to _really_ care and you run it by them _first_. >> Exception handling in object-oriented programming is specifically >> designed to address the problems in C++ that C's "goto" and long >> branch (long goto? I can't remember what it's called) solve in C. >> > When we went to C++ most of the error case gotos went away with well > designed destructors (to release resources) and throw() of an exception. And yup. -- Tim Wescott Control system and signal processing consulting www.wescottdesign.com |