clang-format and single-line statements
2024-12-1 16:0:0 Author: maskray.me(查看原文) 阅读量:0 收藏

The Google C++ Style is widely adopted by projects. It contains a brace omission guideline in Looping and branching statements:

For historical reasons, we allow one exception to the above rules: the curly braces for the controlled statement or the line breaks inside the curly braces may be omitted if as a result the entire statement appears on either a single line (in which case there is a space between the closing parenthesis and the controlled statement) or on two lines (in which case there is a line break after the closing parenthesis and there are no braces).

1
2
3
4
5
6
7
8
9

if (x == kFoo) { return new Foo(); }


if (x == kFoo) return new Foo();


if (x == kBar)
  Bar(arg1, arg2, arg3);

In clang-format's predefined Google style for C++, there are two related style options:

1
2
3
% clang-format --dump-config --style=Google | grep -E 'AllowShort(If|Loop)'
AllowShortIfStatementsOnASingleLine: WithoutElse
AllowShortLoopsOnASingleLine: true

The two options cause clang-format to aggressively join lines for the following code:

1
2
3
4
5
6
7
8
for (int x : a)
  foo(x);

while (cond())
  foo(x);

if (x)
  foo(x);

As a heavy debugger user, I find this behavior cumbersome.

1
2
3
4
5
6
7

#include <vector>
void foo(int v) {}
int main() {
  std::vector<int> a{1, 2, 3};
  for (int x : a) foo(x); 
}

When GDB stops at the for loop, how can I step into the loop body? Unfortunately, it's not simple.

If I run step, GDB will dive into the implementation detail of the range-based for loop. It will stop at the std::vector::begin function. Stepping out and executing step again will stop at the std::vector::end function. Stepping out and executing step another time will stop at the operator!= function of the iterator type. Here is an interaction example with GDB:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
(gdb) n
5         for (int x : a) foo(v);
(gdb) s
std::vector<int, std::allocator<int> >::begin (this=0x7fffffffdcc0) at /usr/include/c++/14.2.1/bits/stl_vector.h:873
873           begin() _GLIBCXX_NOEXCEPT
(gdb) fin
Run till exit from #0  std::vector<int, std::allocator<int> >::begin (this=0x7fffffffdcc0) at /usr/include/c++/14.2.1/bits/stl_vector.h:873
0x00005555555561d5 in main () at a.cc:5
5         for (int x : a) foo(v);
Value returned is $1 = 1
(gdb) s
std::vector<int, std::allocator<int> >::end (this=0x7fffffffdcc0) at /usr/include/c++/14.2.1/bits/stl_vector.h:893
893           end() _GLIBCXX_NOEXCEPT
(gdb) fin
Run till exit from #0  std::vector<int, std::allocator<int> >::end (this=0x7fffffffdcc0) at /usr/include/c++/14.2.1/bits/stl_vector.h:893
0x00005555555561e5 in main () at a.cc:5
5         for (int x : a) foo(v);
Value returned is $2 = 0
(gdb) s
__gnu_cxx::operator!=<int*, std::vector<int, std::allocator<int> > > (__lhs=1, __rhs=0) at /usr/include/c++/14.2.1/bits/stl_iterator.h:1235
1235        { return __lhs.base() != __rhs.base(); }
(gdb) fin
Run till exit from #0  __gnu_cxx::operator!=<int*, std::vector<int, std::allocator<int> > > (__lhs=1, __rhs=0) at /usr/include/c++/14.2.1/bits/stl_iterator.h:1235
0x0000555555556225 in main () at a.cc:5
5         for (int x : a) foo(v);
Value returned is $3 = true
(gdb) s
__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >::operator* (this=0x7fffffffdca0) at /usr/include/c++/14.2.1/bits/stl_iterator.h:1091
1091          { return *_M_current; }
(gdb) fin
Run till exit from #0  __gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >::operator* (this=0x7fffffffdca0) at /usr/include/c++/14.2.1/bits/stl_iterator.h:1091
0x00005555555561f7 in main () at a.cc:5
5         for (int x : a) foo(v);
Value returned is $4 = (int &) @0x55555556b2b0: 1

You can see that this can significantly hinder the debugging process, as it forces the user to delve into uninteresting function calls of the range-based for loop.

In contrast, when the loop body is on the next line, we can just run next to skip the three uninteresting function calls:

1
2
for (int x : a) 
  foo(x);

The AllowShortIfStatementsOnASingleLine style option is similar. While convenient for simple scenarios, it can sometimes hinder debuggability.

For the following code, it's not easy to skip the c() and d() function calls if you just want to step into foo(v).

1
if (c() && d()) foo(v);

Many developers, mindful of potential goto fail-like issues, often opt to include braces in their code. clang-format's default style can further reinforce this practice.

1
2
3
4
5
6
7

if (v) {
  foo(v);
}
for (int x : a) {
  foo(x);
}

Other predefined styles

clang-format's Chromium style is a variant of the Google style and does not have the aforementioned problem. The LLVM style, and many styles derived from it, do not have the problem either.

1
2
3
4
5
6
% clang-format --dump-config --style=Chromium | grep -E 'AllowShort(If|Loop)'
AllowShortIfStatementsOnASingleLine: Never
AllowShortLoopsOnASingleLine: false
% clang-format --dump-config --style=LLVM | grep -E 'AllowShort(If|Loop)'
AllowShortIfStatementsOnASingleLine: Never
AllowShortLoopsOnASingleLine: false

A comparative look at other Languages

Go, Odin, and Rust require {} for if statements but omit (), striking a balance between clarity and conciseness. C/C++'s required ()` makes opt-in braces feel a bit verbose.

C3 and Jai, similar to C++, make {} optional.


文章来源: https://maskray.me/blog/2024-12-01-clang-format-and-single-line-statements
如有侵权请联系:admin#unsafe.sh