User-defined string literals not split

[Meinersbur]

clang-format is currently unable to break user-defined string literals. For instance,

"A very very very very very very very very very very very very very long string literal";

is split into

"A very very very very very very very very very very very very very long "
"string literal";

to fit into the 80 characters column limit.

But

unsigned operator ""_s(const char*);
"A very very very very very very very very very very very very very long string literal"_s;

remains the same (Godbolt). clang-format is allowed to break the string into

"A very very very very very very very very very very very very very long "
"string literal"_s;

as explained by cppreference.com:

Implementation-wise, this is because clang-format only recognizes a hardcoded set of string literal encodings to be breakable:

llvm-project/clang/lib/Format/ContinuationIndenter.cpp

Lines 2520 to 2527 in 9d1b6ee

	if ((Text.ends_with(Postfix = "\"") &&
	(Text.starts_with(Prefix = "@\"") || Text.starts_with(Prefix = "\"") ||
	Text.starts_with(Prefix = "u\"") ||
	Text.starts_with(Prefix = "U\"") ||
	Text.starts_with(Prefix = "u8\"") ||
	Text.starts_with(Prefix = "L\""))) ||
	(Text.starts_with(Prefix = "_T(\"") &&
	Text.ends_with(Postfix = "\")"))) {

[llvmbot]

@llvm/issue-subscribers-clang-format

Author: Michael Kruse (Meinersbur)

clang-format is currently unable to break user-defined string literals. For instance, ``` ```

https://godbolt.org/z/5zhde1M7E

[shivrm]

Hi, I'd like to work on this. Could this be assigned to me?

[shivrm]

I've added regex-based detection for user-defined suffixes. However, because of how BreakableStringLiteral works, the suffix currently appears on each broken line. For example:

"A very very very very very very very very very very very very very long string literal"_s;

would be split as

"A very very very very very very very very very very very very very long "_s
"string literal"_s;

(both lines have the _s suffix)

If this is fine, I'll go ahead and make a pull request.

[Meinersbur]

All suffixes being the same is permitted by the language specification (https://godbolt.org/z/Kb9WMec5x), but I would find having it appear only once to be preferable. Some suffixes might be too long to be repeated on every line. Maybe add an option to choose?

[shivrm]

I've modified BreakableStringLiteral to make prefixes and suffixes appear only once.

So

L"A very very very very very very very very very very very very very long string literal"_s;

would be split as

L"A very very very very very very very very very very very very very long"
" string literal"_s;

(both L and _s appear only once)

Is this fine?

This does break an existing test case though: https://github.com/llvm/llvm-project/blob/311d115ed809724855f821616ed7aabab32896d4/clang/unittests/Format/FormatTest.cpp#L15816-L15846

How would you suggest I proceed? Should I change the test case?

[Meinersbur]

I would keep the existing behaviour for prefixes: There might have been some rationale to repeat them. Also my argument of potentially long uds-suffixes does not apply.

Also, this # only targeting user-defined string literals. Changing other behaviour could cause concistency problems.

[shivrm]

Sure, I'll keep the existing behavior and open a pull request soon :)