Taken from tools/llvmc2/CompilationGraph.cpp:
...
for (typename C::const_iterator B = EdgesContainer.begin(),
E = EdgesContainer.end(); B != E; ++B) {
const Edge* E = B->getPtr();
...
MS C++ compiler (VS 2008) gives:
...
CompilationGraph.cpp
..\..\..\llvm\tools\llvmc2\CompilationGraph.cpp(58) : error C2371:
'E' : redefinition; different basic types
..\..\..\llvm\tools\llvmc2\CompilationGraph.cpp(57) : see
declaration of 'E'
...
While GCC compiles it fine. (I'm assuming it does, 'cause almost
everyone here uses gcc)
Changing the code to
...
for (typename C::const_iterator B = EdgesContainer.begin(),
End = EdgesContainer.end(); B != End; ++B) {
const Edge* E = B->getPtr();
...
makes the code compilable by MS C++. But as a curiosity (and I really
don't know the answer because I can barely read C++): Which compiler
got it right?
gcc is correct. According to the ISO specification, the for-init-statement is supposed to inject any variable names into the same declarative scope as the condition of an equivalent restructuring of the loop in the form of a while statement, which in turn fronts the declaration to an extra scope that surrounds the /entire/ loop construct. VC++ seems to be scoping the variables as if they were /inside/ of the loop and not creating this extra scope. Frowny. -J
This can be changed in the project settings for a VC++ project. I think it was called something like “Enforce for loop conformance” or so…
This is something I always change though.
This can be changed in the project settings for a VC++ project.
I think it was called something like "Enforce for loop conformance" or
so...
This is something I always change though.
Nope. The for loop conformance option fixes a VC++6 "feature" where
identifiers declared in the for loop initialization are visible after the
for loop's body.
VC6:
for(int i = 0; i < 5; ++i)
{
...
}
std::cout << i; //Prints "5"
This is illegal in ISO C++, and by default in VC8 and beyond (maybe earlier,
but I moved from 6 to 8). The VC++ option enable VC6-like behavior for
legacy code. What Jay pointed out is Yet Another For Scope FUBAR.
To the VC team's credit, though, "inline any suitable" is pure awesome.
gcc is correct. According to the ISO specification, the for-init-statement is supposed to inject any variable names into the same declarative scope as the condition of an equivalent restructuring of the loop in the form of a while statement, which in turn fronts the declaration to an extra scope that surrounds the /entire/ loop construct. VC++ seems to be scoping the variables as if they were /inside/ of the loop and not creating this extra scope. Frowny. -J
Actually, gcc is wrong and VC++ got it right.
From the C++ standard, 6.4p3:
A name introduced by a declaration in a condition (either introduced by the type-specifier-seq or the declarator of the
condition) is in scope from its point of declaration until the end of the substatements controlled by the condition. If the
name is re-declared in the outermost block of a substatement controlled by the condition, the declaration that re-declares
the name is ill-formed.
Which gives us:
while (int x=0) {
int x=0; // error: redeclaration, clashes with condition
}
Both gcc and VC++ emit a compilation error for the above.
Then the standard says this, 6.5.3p1:
names declared in the for-init-statement are in the same declarative-region as those declared in the condition
So names inside the 'for' loop clash with both the condition and the for-init-statement:
for (int x=0; {
int x=0; // error: redeclaration, clashes with for-init-statement
}
but gcc, incorrectly, does not emit a compilation error.
And while we are on the subject, gcc is also wrong on this one:
if (int x=0) {
int x=0; // error: redeclaration, but gcc does not emit any errors.
}
Ah, interesting, have not ran across that before (as I always strive to never use the same name as any scope previously), but rather interesting that GCC gets it wrong while VC++ gets it right, kind of a switch.
And yes, that switch I mentioned is for VC6 style in VC7.1 (what I use), you have to switch it to make it conformant, nice to hear VC8 does it correctly by default now.
Those rules only apply to if and switch statements. (Yes, this is insane, but true.) The entire section you are quoting from, 6.4, is titled "Selection statements [stmt.select]", which specifically covers these two cases. A for is an iteration statement, not a selection statement.
So, if you read 6.5.3p1 (which is actually about for statements) it states that a for loop is rewritten as a while loop where names defined by the for-init-statement are declared in the declaration space of the condition of the while statement. Then, if you read 6.5.1p2 it states:
<quote>
-2- When the condition of a while statement is a declaration, the scope of the variable that is declared extends from its point of declaration (basic.scope.pdecl) to the end of the while statement. A while statement of the form
while (T t = x) statement
is equivalent to
label:
{ // start of condition scope
T t = x;
if (t) {
statement
goto label;
}
} // end of condition scope
</quote>
Unlike the case of the for statement, there are no subsequent rules that redefine where this declaration is scoped. This rewrite means that the rules you cite in 6.4 simply do not apply: for loops don't cause conditions in selection statements, and therefore should not be scoped in that manner. They almost are, but the rules for while statements actually force the declaration outside of the condition of the if statement.
You are correct, though, that gcc should be emitting an error in the case of the if statement. Naughty gcc.
Those rules only apply to if and switch statements. (Yes, this is insane, but true.) The entire section you are quoting from, 6.4, is titled "Selection statements [stmt.select]", which specifically covers these two cases. A for is an iteration statement, not a selection statement.
See 6.4p2: "The rules for conditions apply both to selection-statements and to the for and while statements"
So, if you read 6.5.3p1 (which is actually about for statements) it states that a for loop is rewritten as a while loop where names defined by the for-init-statement are declared in the declaration space of the condition of the while statement. Then, if you read 6.5.1p2 it states:
<quote>
-2- When the condition of a while statement is a declaration, the scope of the variable that is declared extends from its point of declaration (basic.scope.pdecl) to the end of the while statement. A while statement of the form
while (T t = x) statement
is equivalent to
label:
{ // start of condition scope
T t = x;
if (t) {
statement
goto label;
}
} // end of condition scope
</quote>
Unlike the case of the for statement, there are no subsequent rules that redefine where this declaration is scoped. This rewrite means that the rules you cite in 6.4 simply do not apply: for loops don't cause conditions in selection statements, and therefore should not be scoped in that manner. They almost are, but the rules for while statements actually force the declaration outside of the condition of the if statement.
Again, see the 6.4p2 reference. 'while' statements should also follow this rule:
If the name is re-declared in the outermost block of a substatement controlled by the condition, the declaration that re-declares the name is ill-formed.
gcc correctly scopes the declaration inside the while statement:
while (int x=0) {
int x=0; // error: redeclaration.
}
{
(Which actually makes me happy as that makes things a lot more consistent and sensible.) -J