[clang-format] Use tabs on GNU style

[owenpan]

Summary

On default configuration of GNU Indent, it uses 8-width tabs.
Because GNU Indent's indentation works like UT_ForContinuationAndIndentation in clang-format, this commit replaces UseTab value from UT_Never to UT_ForContinuationAndIndentation.

Quoting Clang-Format Style Options (15.0.0git):

BasedOnStyle (String)
...
GNU A style complying with the GNU coding standards

Quoting GNU Coding Standards:

The rest of this section gives our recommendations for other aspects of C formatting style, which is also the default style of the indent program in version 1.2 and newer. It corresponds to the options

-nbad -bap -nbc -bbo -bl -bli2 -bls -ncdb -nce -cp1 -cs -di2
-ndj -nfc1 -nfca -hnl -i2 -ip5 -lp -pcs -psl -nsc -nsob

We don’t think of these recommendations as requirements, because it causes no problems for users if two different programs have different formatting styles.

And GNU Indent Manual:

-ut
--use-tabs

Use tabs. This is the default.
See Indentation.

-nut
--no-tabs

Use spaces instead of tabs.
See Indentation.

Because neither -nut nor -ut are specified in the GNU Coding Standards' default style, it's -ut (use tabs) by default.
Although using tabs are not requirements (as specified in the GNU Coding Standards), using only spaces is definitely not the default GNU-style indentation.

This kind of indent roughly corresponds to clang-format's UT_ForContinuationAndIndentation. Quoting GNU Indent's source code:

if (settings.use_tabs && (settings.tabsize > 1))
{

...

    offset = settings.tabsize - (cur_col - 1) % settings.tabsize;
    while (cur_col + offset <= align_target)
    {
        putc(TAB, output);
        cur_col += offset;
        offset = settings.tabsize;
    }
}

while (cur_col < target_column)
{
    putc(' ', output);
    cur_col++;
}

This portion from pad_output function first tries to fill tabs for alignment and fine-adjusts with spaces.

[curdeius]

It seems like a breaking change that may be painful for users of GNU style. @MyDeveloperDay, wdyt?

[MyDeveloperDay]

Oh gosh! It hard to make such changes and someone else not call them a regression, I’m not sure how I feel. We do kind of have to be able to fix bugs and if this gets closer to gnu style then I guess it’s better

[owenpan]

In D127484#3574992, @MyDeveloperDay wrote:

Oh gosh! It hard to make such changes and someone else not call them a regression, I’m not sure how I feel. We do kind of have to be able to fix bugs and if this gets closer to gnu style then I guess it’s better

+1.

clang/unittests/Format/FormatTest.cpp

23954

To simplify the test case a little bit.

[a4lg]

Thanks, suggestion by @owenpan is applied.

[owenpan]

In D127484#3574803, @curdeius wrote:

It seems like a breaking change that may be painful for users of GNU style.

Yeah. There are tens of thousands of lines starting with 8 spaces but none starting with a tab in all .c files under git://git.sv.gnu.org/coreutils.

[MyDeveloperDay]

We get heavily criticized for changing defaults between versions, I think if someone want to use GNU style, and they choose to now have this option set it should be done so in their own .clang-format file

On reflection, I think its wrong of us to change that now in case people are doing

BasedOnStyle: GNU

I sort of feel this is a style we have to leave alone. But what do GNU users think? are they complaining? What led you to this change in the first place (I saw the twitter thread)

[a4lg]

Interesting.

I measured some development branch (targeting *.c, *.cc, *.cpp and *.h):

Program	prefix: 8 spaces	prefix: 1 tab	8sp percentage
Coreutils	23494	461	98.08%
Indent	5113	5908	46.39%
Binutils	128781	745992	14.72%
GCC	155080	1156554	11.82%
Glibc	18680	176524	9.57%

That's completely different from Coreutils (and that's why I submitted this patch because I'm a Binutils contributor).

Yes, there's definitely a reason to keep this style in clang-format. In this case, following .clang-format file will work:

BasedOnStyle: GNU
UseTab: ForContinuationAndIndentation

or following style setting on Visual Studio Code (not as JSON but a string):

{ BasedOnStyle: GNU, UseTab: ForContinuationAndIndentation }

[owenpan]

In D127484#3576061, @a4lg wrote:

Interesting.

I measured some development branch (targeting *.c, *.cc, *.cpp and *.h):

Program	prefix: 8 spaces	prefix: 1 tab	8sp percentage
Coreutils	23494	461	98.08%

All of the 461 lines with a tab prefix are in a single header file: coreutils/src/longlong.h. So probably by accident.

[MyDeveloperDay]

@a4lg If we have decided NOT to do this can we Abandon the review so as to remove it from the reviewers "clang-format" lists?

clang/lib/Format/Format.cpp

   Style.FixNamespaceComments = false;
   Style.SpaceBeforeParens = FormatStyle::SBPO_Always;
   Style.Standard = FormatStyle::LS_Cpp03;
+  Style.UseTab = FormatStyle::UT_ForContinuationAndIndentation;
   return Style;
 }
 

clang/unittests/Format/FormatTest.cpp

                "  int foo ();\n"
                "}",
                Style);
+  verifyFormat("void\n"
+               "foo ()\n"
+               "{\n"
+               "  for (int i = 0; i < 5; i++)\n"
+               "    {\n"
+               "      for (int j = 0; j < 5; j++)\n"
+               "\t{\n"
+               "\t  bar ();\n"
+               "\t}\n"
+               "    }\n"
+               "}",
+               Style);
 }
 TEST_F(FormatTest, MozillaDefaultStyle) {
   FormatStyle Style = getMozillaStyle();