|
Kevin McFarlane wrote: There can be some cases in which such a pattern is fine, though usually you should be catching a specific exception, but you should always explain why/what you're doing in such cases.
I wish .net included a few more 'tryDoSomething' methods. For example, if one thread has just changed some information that should be displayed be a control, a sensible thing to do is to BeginInvoke the control's update handler (preferably using a 'Threading.Interlocked.CompareExchange'ed flag to prevent excessive 'BeginInvoke's). Unfortunately, I know of no good way to avoid the risk of the control being disposed before the BeginInvoke. In a situation like that, a tryBeginInvoke would be extremely handy; if it works, great. If the control's been disposed or otherwise lacks a windowing context, there's no need to update it and the call may be safely ignored.
If a try block contains a single BeginInvoke and the catch block is empty, is that not sufficiently clear "Try this, and if it works great--if not, meh."
|
|
|
|
|
David Kentley wrote: I checked and validated results as I went, using if/elses.
I'm a fan of doing some error trapping this way. It's quick and you can log a very specific error message that says exactly what's wrong, like an object doesn't exist. That's good for the developer that has to fix the issue. Besides, having an exception thrown and being caught is expensive from a processor time point of view.
I've also seen large blocks of code wrapped in a try catch block where the catch block logged an "unhandled" exception. Might as well have said "s&*t happened"!
|
|
|
|
|
That reminds me of a great piece of code and comment in Quake:
void(entity targ, entity attacker) ClientObituary =
{
local float rnum;
local string deathstring, deathstring2;
rnum = random();
...
if (targ.classname == "player")
{
if (attacker.classname == "player")
{
if (targ == attacker)
{
attacker.frags = attacker.frags - 1;
bprint (targ.netname);
if (targ.weapon == 64 && targ.waterlevel > 1)
{
bprint (" discharges into the water.\n");
return;
}
if (targ.weapon == IT_GRENADE_LAUNCHER)
bprint (" tries to put the pin back in\n");
else
bprint (" becomes bored with life\n");
return;
}
else
{
attacker.frags = attacker.frags + 1;
rnum = attacker.weapon;
if (rnum == IT_AXE)
{
deathstring = " was ax-murdered by ";
deathstring2 = "\n";
}
...
bprint (targ.netname);
bprint (deathstring);
bprint (attacker.netname);
bprint (deathstring2);
}
return;
}
else
{
targ.frags = targ.frags - 1;
bprint (targ.netname);
if (attacker.flags & FL_MONSTER)
{
if (attacker.classname == "monster_army")
bprint (" was shot by a Grunt\n");
if (attacker.classname == "monster_demon1")
bprint (" was eviscerated by a Fiend\n");
...
return;
}
if (attacker.classname == "explo_box")
{
bprint (" blew up\n");
return;
}
if (attacker.solid == SOLID_BSP && attacker != world)
{
bprint (" was squished\n");
return;
}
...
if (targ.deathtype == "falling")
{
targ.deathtype = "";
bprint (" fell to his death\n");
return;
}
bprint (" died\n");
The whole routine is about 250 lines long, but for whatever reason I particularly like the last comment.
|
|
|
|
|
Recently, I stumbled across this little gem. I don't have the exact code handy, but the gist of it is:
nErrorCode = cFtpConn.SetHost(HOST);
if (nErrorCode == 0)
{
nErrorCode = cFtpConn.SetUser(USERNAME);
if (nErrorCode == 0)
{
nErrorCode = cFtpConn.SetPassword(PASSWORD);
if (nErrorCode == 0)
{
nErrorCode = cFtpConn.SetPath(PATH);
if (nErrorCode == 0)
{
nErrorCode = cFtpConn.SetFilename(FILENAME);
if (nErrorCode == 0)
{
}
else
{
Log("Error setting filename");
}
}
else
{
Log("Error setting path");
}
}
else
{
Log("Error setting password");
}
}
else
{
Log("Error setting username");
}
}
else
{
Log("Error setting host");
}
|
|
|
|
|
C/C++ or C#?
If it's C#, the methods should probably throw Exceptions.
If it's C/C++, I see no real problem with it. Had I written it, it would be:
if ((nErrorCode = cFtpConn.SetHost(HOST)) == 0)
and the Log messages would include the value of nErrorCode.
How would you improve it?
|
|
|
|
|
Let's assume it's C++.
I consider sth. like the code above generally bad coding style. There is far to much nesting here. Supposed that most of the programmers (at least the ones I know, including myself) make an indentation of four spaces (not only two as in the 'sample'), you would quickly run out of monitor space...
I would suggest a kind of 'waterfall style' coding here:
if ((nErrorCode = cFtpConn.SetHost(HOST)) != 0)
{
Log(...);
return;
}
if ((nErrorCode = ...
{
Log(...);
return;
}
...
This is also not perfect since it introduces many returns, but it improves the readability of the code and the return conditions are trivial and repetitive.
PIEBALDconsult wrote: If it's C#, the methods should probably throw Exceptions.
Agreed. In a perfect world, C# - Methods would always throw exceptions and never signal an error by means of a return value. (As long as this is affordable in terms of performance).
Regards
Thomas
modified on Monday, November 3, 2008 4:55 AM
|
|
|
|
|
If you can't throw exceptions or the return type is already taken, there is another approach:
if ((nErrorCode = cFtpConn.SetHost(HOST)) != 0)
{
Log("Error setting host");
}
if (nErrorCode == 0 && (nErrorCode = cFtpConn.SetUser(USERNAME)) !=0)
{
Log("Error setting username");
}
if (nErrorCode == 0 && (nErrorCode = cFtpConn.SetPassword(PASSWORD)) != 0)
{
Log("Error setting password");
}
Once nErrorCode is non-zero the assignment and comparision are never done.
Sorted!
Panic, Chaos, Destruction.
My work here is done.
|
|
|
|
|
|
In fact, I think your approach is more problematic. What if resources need to be released before returning from the function? The if branch will grow bigger and bigger as you are nearing the end of the function, and most of it will be a copy-paste code.
|
|
|
|
|
Nemanja Trifunovic wrote: What if resources need to be released before returning from the function? The if branch will grow bigger and bigger as you are nearing the end of the function
True, but if there are some actions to be taken that are not part of the functional code (e.g. releasing resources), I'd consider to take a try/finally approach or implementing sort of Dispose pattern - depending on the problem to solve and whether it is C/C++ or C#.
I would definitely not end up with a bunch of if s in this case, but the 'sample code' does not give any hint in that direction - and I think this is not the point here...
Regards
Thomas
|
|
|
|
|
Thomas Weller wrote: True, but if there are some actions to be taken that are not part of the functional code (e.g. releasing resources), I'd consider to take a try/finally approach or implementing sort of Dispose pattern - depending on the problem to solve and whether it is C/C++ or C#.
Exceptions coupled with automatic release of resources are the best approach. But if this approach is not available, the nested if s still work well and are reasonably readable: at least the error paths are somewhat separated from the "normal" flow
Typical well written COM code often looks like:
ISomeInterface* pInter(NULL);
HRESULT hr = E_FAIL;
if (SUCCEEDED(SomeFactory->CreateSomeObject(&pInter)))
{
if (SUCCEEDED(pInter->Operation1()))
{
if (SUCCEEDED(pInter->Operation2()))
{
DoSomething(pInter);
hr = S_OK;
}
else
hr = E_WHATEVER2;
else
hr = E_WHATEVER1;
}
else
pInter->Release();
}
return hr;
|
|
|
|
|
Nemanja Trifunovic wrote: Exceptions coupled with automatic release of resources are the best approach.
This is a quite good definition of what Dispose pattern is in C#...
In my opinion, there are two problems with code consisting of many nested if paths:
- It is hard to follow if it gets lengthy. Error probability increases dramatically with every level of nesting - especially when it comes to maintenance.
- This sort of coding simply does not well with monitor space. Lines are indented for every nesting level - and soon you have to scroll horizontally only for reading source code!
That's why the many if s are a coding horror in my opinion: Readability and maintainability issues.
Regards
Thomas
|
|
|
|
|
Thomas Weller wrote: This is a quite good definition of what Dispose pattern is in C#...
It even better describes the RAII idiom in C++
Thomas Weller wrote: It is hard to follow if it gets lengthy
It does, but at least it is cleanly separated: the "normal path" is in the if part, and the error handling in the else part. With the "pipe" model, both code paths interrupt each other and thats really messy and error prone.
Thomas Weller wrote: Error probability increases dramatically with every level of nesting - especially when it comes to maintenance.
How come? There is no copy-paste code and if something needs to be changed, it needs to be changed in one place. With the "pipe" model, if you add a new resource allocation, you need to make sure that it is released in each return path.
Thomas Weller wrote: This sort of coding simply does not well with monitor space. Lines are indented for every nesting level - and soon you have to scroll horizontally only for reading source code!
No argument here, except that most editors have this secret little feature called "line wrapping"
Thomas Weller wrote: Readability and maintainability issues.
Exactly the same arguments I have for the opposite argument - don't you love programming discussions?
|
|
|
|
|
|
Thomas Weller wrote: What the heck is the RAII idiom in C++
Resource Acquisition is Initialization[^] (a horrible name, but a very useful idiom)
Thomas Weller wrote: How can line wrapping help with horizontal scrolling?
So what does it help with then? Try turning on line wrapping in Notepad and start typing - no matter what you do, there will be no horizontal scroll bars
|
|
|
|
|
Nemanja Trifunovic wrote: So what does it help with then?
Sorry, I was confusing line wrapping with the expand/collapse region feature. If you said word wrapping instead it would have been clear to me what you mean. I think this is because I am a German and not used to the exact idioms you are using in the U.S.
This is a good example for a misunderstanding that would have been resolved within seconds in a face-to-face situation...
Regards
Thomas
|
|
|
|
|
if ()
{
if ()
{
if ()
{
if ()
{
if ()
{
I Fail
to see
why
this
is any
less a
horror
becaus
e it
was
line
wrappe
d
automa
ticall
y.
}
else
{
}
}
else
{
}
}
else
{
}
}
else
{
}
}
else
{
}
Today's lesson is brought to you by the word "niggardly". Remember kids, don't attribute to racism what can be explained by Scandinavian language roots.
-- Robert Royall
|
|
|
|
|
Exactly my point, unless I did not have the idea of putting it that way.
Regards
Thomas
|
|
|
|
|
And the alternative:
Acquire1();
if (!Works1())
{
Release1();
return;
}
Acquire2();
if (!Works2())
{
Release1();
Release2();
return;
}
Acquire3();
if (!Works3())
{
Release1();
Release2();
Release3();
return;
}
...
AcquireN();
if (!WorksN())
{
Release1();
Release2();
Release3();
...
ReleaseN();
return;
}
Forget to copy one of the Release functions and you have a nice resource leak
|
|
|
|
|
In C# you can make 1-N classes, put the release code in the destructors and have it cleaned up automatically. Alternately you could have a finally block with a series of if (Thing1.Aquired) Thing1.Release() statements.
Today's lesson is brought to you by the word "niggardly". Remember kids, don't attribute to racism what can be explained by Scandinavian language roots.
-- Robert Royall
|
|
|
|
|
As I said, ideally you would use exceptions and RAII (in C# that would be using ) and then the code would simply look like:
{
Resource1 r1;
r1.DoSomething1();
Resource2 r2;
r2.DoSomething2();
...
}
My point is that if we need to stick to the C-style error handling it is much safer to write structured code with one entry and one exit, and no copy-paste blocks.
BTW, the number of if s is the same in both styles - the difference is how you structure them.
|
|
|
|
|
This is also not perfect since it introduces many returns, but it improves the readability of the code and the return conditions are trivial and repetitive.
How about:
if ((err = action1()) != 0)
log_error1();
else if ((err = action2()) != 0)
log_error2();
else if ((err = action3()) != 0)
log_error3();
or
do
{
if ((err = action1()) != 0)
{log_error1(); break;}
if ((err = action2()) != 0)
{log_error2(); break;}
prepare_for_action3();
if ((err = action3()) != 0)
{log_error3(); break;}
...
} while(0);
The former style is nicer if each action is a single function. If stuff is required between the actions, the second approach may be helpful.
|
|
|
|
|
Hmm, I fear I'm not very happy with that either.
The first example is just not working because due to else the branches below the first one are simply unreachable, no matter what the outcome of action1() may be.
The second alternative is just replacing return; with break; . Furthermore, it introduces a hardcoded boolean expression which always evaluates to the same value. This in my opinion is not very desirable in itself.
Sure, in the example things are very easy to understand, but imagine a real life example where things can become much more complicated...
Regards
Thomas
Programming today is a race between software engineers striving to build bigger and better idiot-proof programs, and the Universe trying to produce bigger and better idiots. So far, the Universe is winning.
Programmer - an organism that turns coffee into software.
|
|
|
|
|
Thomas Weller wrote: The first example is just not working
Sorry for that - of course it is working. My brain must be on vacation or something. Thus this indeed is a viable alternative.
Regards
Thomas
Programming today is a race between software engineers striving to build bigger and better idiot-proof programs, and the Universe trying to produce bigger and better idiots. So far, the Universe is winning.
Programmer - an organism that turns coffee into software.
|
|
|
|
|
Thomas Weller wrote: The second alternative is just replacing return; with break;. Furthermore, it introduces a hardcoded boolean expression which always evaluates to the same value. This in my opinion is not very desirable in itself.
Replacing a return with a break may be useful if the code has to do something besides totally exit a function. If the routine opened a file at the beginning, for example, I would consider doing a break and then closing the file after the 'while' to be much cleaner than doing "close(theFile); return;" in each failure case.
It's a little irksome having a hard-coded boolean constant like that, but C does not provide any other block structure whose semantics are "run once, but be able to jump to the beginning or end." I would consider "do ... while(0);" and "do ... while(1);" to be cleaner than a "goto", at least in cases where the enclosing block does not contain any case labels.
If a certain amount of code will be common to several case handlers, it would be far better to do something like:
switch(foo)
{
case 0:
code_0_special();
COMMON:
common_to_code_0_1_3();
break;
case 1:
code_1_special();
goto COMMON;
case 2:
code_2_special();
break;
case 3:
code_3_special();
goto COMMON;
default:
handle_default();
}
than
switch(foo)
{
case 0:
code_0_special();
do {
common_to_code_0_1_3();
break;
case 1:
code_1_special();
continue;
case 2:
code_2_special();
break;
case 3:
code_3_special();
continue;
default:
handle_default();
break;
} while(1);
}
or
switch(foo)
{
do
{
case 0:
code_0_special();
break;
case 1:
code_1_special();
break;
case 3:
code_3_special();
break;
} while(0);
common_to_code_0_1_3();
break;
case 2:
code_2_special();
break;
default:
handle_default();
break;
} while(1);
}
The former would IMHO be an appropriate use of "goto"; the second is just plain horrible. The third isn't quite so bad, but is IMHO less clear than the goto.
|
|
|
|