[clang-format] restore indent in conditionals when AlignOperands is DontAlign

[krasimir]

Summary

After D50078, we're experiencing unexpected un-indent using a style combining AlignOperands: DontAlign with BreakBeforeTernaryOperators: false, such as Google's JavaScript style:

% bin/clang-format -style=google ~/test.js
aaaaaaaaaaa = bbbbbbbb ? cccccccccccccccccc() :
dddddddddd             ? eeeeeeeeeeeeee :
                         fffff;

The issue lies with the interaction of AlignOperands: DontAlign and the edited code section in ContinuationIndenter.cpp, which de-dents the intent by Style.ContinuationIndentWidth. From the documentation of AlignOperands: DontAlign:

The wrapped lines are indented ContinuationIndentWidth spaces from the start of the line.

So the de-dent effectively erases the necessary ContinuationIndentWidth in that case.

This patch restores the AlignOperands: DontAlign behavior, producing:

% bin/clang-format -style=google ~/test.js
aaaaaaaaaaa = bbbbbbbb ? cccccccccccccccccc() :
    dddddddddd         ? eeeeeeeeeeeeee :
                         fffff;

[sammccall]

This may not be the right long-term solution, but the current behavior is pretty broken and this is cheap.

clang/unittests/Format/FormatTest.cpp

6287

aligning the question marks here is a bit weird (given DontAlign) but that's another patch.

If we disable the question-column behavior with dontalign, this patch will be completely dead, right?
May want to add a FIXME to remove in that case.

[krasimir]

• add a FIXME

[krasimir]

• refresh Phabricator diff

[krasimir]

• try to update Phabricator diff

[Typz]

clang/unittests/Format/FormatTest.cpp

6287

I don't think this is so weird: even with DontAlign, there are other cases when some alignment is performed: for exemple when formatting tables in column.

As I see it, ternary operator formatting is a similar case of 2D formatting, and while it needs indeed to respect the "general" line wrapping/indent mode (as per AlignOperands), it is OK to keep the alignment of the ternary operator themselves : otherwise, the whole "mode" for ternary operators need to be disabled and makes no sense.

But anyway this is a different patch as you mentioned, and maybe some user of DontAlign can come up with a better approach for formatting ternary ops in that case.

[krasimir]

clang/unittests/Format/FormatTest.cpp

6287

Added a FIXME. I think disabling question-column alignment with dontalign indeed can make this patch obsolete.

6287

Agree, this is more of a policy / preference on how we want these snippets to look like with DontAlign. We should discuss it in more detail whenever we attempt to refine this later. In a way, this touches back on djasper's comment on D50078, which I believe was left unaddressed:

I don't think the alignment of "?" and ":" (in the WhitespaceManager) should be always on. I think we'd need a flag for that part

clang/unittests/Format/FormatTest.cpp

verifyFormat("return aaaaaaaaaaaaaaaaaaaaaaaaa\n"
                "             : cccccccccccccccc ? dddddddddddddddddd\n"
                "                                : eeeeeeeeeeeeeeeeee\n"
                "       : bbbbbbbbbbbbbbbbbbbbbbb ? 2222222222222222\n"
                "                                 : 3333333333333333;",
                Style);
 
   Style.AlignOperands = FormatStyle::OAS_DontAlign;
   Style.BreakBeforeTernaryOperators = false;
+  // FIXME: Aligning the question marks is weird given DontAlign.
+  // Consider disabling this alignment in this case. Also check whether this
+  // will render the adjustment from https://reviews.llvm.org/D82199
+  // unnecessary.
   verifyFormat("int x = aaaaaaaaaaaaaaa ? aaaaaaaaaaaaaaaaaa :\n"
                "    bbbb                ? cccccccccccccccccc :\n"
                "                          ddddd;\n",
                Style);
 }
 
 TEST_F(FormatTest, BreaksConditionalExpressionsAfterOperator) {
   FormatStyle Style = getLLVMStyle();