I'd recommend you to read thoroughly the examples you have found (more if they are in the kernel code of an operating system.) The situation you describe corresponds to an algorithm that should make decisions at each stage of the execution, and those stages require to undo the previous steps.
- You first allocate some resource
#1, and continue.
- then you allocate another resource (say resource
#2) if that fails, then you have to free resource #1, as it is not longer valid.
- ...
- finally you allocate resource
#N, if that fails you must free resources #1 to #N-1.
The figure you show allows you to write in one line, a set of resource allocations, between which you have to decide if you continue.
In this scenario a policy like this is recommended (for novice C programmers, as it avoids the use of goto but becomes less readable (as it nests as things happen)
if ((res_1 = some_allocation(blablah)) != ERROR_CODE) {
if ((res_2 = some_other_allocation(blablatwo)) != ANOTHER_ERROR_CODE) {
...
if ((res_N = some_N_allocation(blablaN)) != NTH_ERROR_CODE) {
do_what_is_needed();
return_resource_N(res_N); /* free resN */
} else {
do_action_corresponding_to_failed_N(); /* error for failing N */
}
return_resource_N_minus_one(resN_1); /* free resN_1 */
...
} else {
do_action_corresponding_to_failed_2(); /* error for failing #2 */
}
return_resource_1(res1); /* free #1. (A): (see below) */
} else {
do_acttion_corresponding_to_failed_1(); /* error for failing #1 */
}
/* there's nothing to undo here, as we have returned the first resource in (A) above. */
nothing to say about this code, but that it has no gotos, but is incredible far less readable (it's a mess of nested things in which, when you fail for resource N, then you have to return up to N-1 resources.) you can messup the resources deallocated by putting them in the wrong position and it's error prone. But on the other side, it allocates and deallocates the things in just one place and is as compact as the code with gotos.
writing this code with gotos gives this:
if ((res_1 = some_allocation(blablah)) == ERROR_CODE) {
do_acttion_corresponding_to_failed_1(); /* error for failing #1 */
goto end;
}
if ((res_2 = some_other_allocation(blablatwo)) == ANOTHER_ERROR_CODE) {
do_action_corresponding_to_failed_2(); /* error for failing #2 */
goto res1;
}
...
if ((res_N = some_N_allocation(blablaN)) == NTH_ERROR_CODE) {
do_action_corresponding_to_failed_N(); /* error for failing #N */
goto resN1;
}
do_what_is_needed();
return_resource_N(res_N); /* free resN */
resN1: return_resource_N_minus_one(resN_1); /* free resN_1 */
...
res1: return_resource_1(res1); /* free #1. (A): (see below) */
end: /* there's nothing to undo here, as we have returned the first resource in (A) above. */
There's only thing that can be said about the first code that will make it perform better in some architectures. Dealing with goto is a pain for the compiler, as normally it has to make assumptions about all the possible resulting blocks that will end jumping to the same label, and this makes things far more difficult to optimice, resulting in not so optimiced code. (this is clear when you use structured blocks, and only implies one or two places you can come from), and you will get worse performance code (not much worse, but somewhat slower code)
You will agree with me that the equivalent code you post in your code is more readable, probably exactly the same level of correctness.
Other required use of goto constructs is when you have several nested loops and you have to exit more than the closest loop to exit.
for(...) {
for(...) {
...
for (...) {
goto out;
}
...
}
}
out:
this is also C specific, as other languages allow you to label the construct you want to exit from and specify it in the break statement.
E.g. in Java:
external_loop: for(...) {
for(...) {
...
for (...) {
break external_loop;
}
...
}
}
In this case you don't need to jump, as the break knows how many loops we need to exit.
One last thing to say. With just the while() construct, all other language constructs can be simulated, by introducing state variables to allow you to do things (e.g. stepping out of each loop by checking some variable used precisely for that). And even less.... if we allow for recursive function call, even the while() loop can be simulated, and optimicers are capable of guessing a faster implementation without recursion for the simulated block. Why in the schools nobody says never use if sentences, they are evil? This is because there's a frequent fact that newbies tend to learn one struct better than others and then, they get the vice of using it everywhere. This happens frequently with goto and not with others, more difficult to understand but easier to use, once they have been understood.
The use of goto for everything (this is the legacy of languages like assembler and early fortran) and maintaining that code normally ends in what is called spaghetti programming. A programmer just selects at random a place to write his/her code in the main code of a program, opens an editor and inserts it's code there:
Let's say that we have to do several steps, named A to F:
{
code_for_A();
code_for_B();
code_for_C();
code_for_D();
code_for_E();
code_for_F();
}
and later, some steps, named G and H have to be added to be executed at the end. Spaghetti programming can make the code end being something like this:
{
code_for_A();
code_for_B();
code_for_C(); /* programmer opened the editor in this place */
goto A;-------.
|
B:<---------------+-.
code_for_G(); | | /* the code is added in the middle of the file */
code_for_H(); | |
goto C;-------+-+--.
| | |
A:<---------------' | |
code_for_D(); | |
code_for_E(); | |
code_for_F(); | |
goto B; --------' |
|
C:<--------------------'
}
While this code is correct (it executes steps A to H in sequence), it will take a programmer some time to guess how the code flows from A to H, by following back and forward the gotos.
