Inconsistent spacing between "template" and "<"

[sdutoit]

From an email to cxxeditor@:

template<typename T>

or

template <typename T>

The standard as-is has 1494 instances of 'template<' that I can find trivially, and 2053 instances of 'template <'.

[sdutoit]

I intend to use template <typename T> throughout - if only because there are more instances of that style existing, and the choice is arbitrary.

[lichray]

I +1 for using typename in place of class in template parameter list wherever possible.

[PeterSommerlad]

My +100 template is so "last millenium" :-)
On 29.08.2013, at 23:42, Zhihao Yuan notifications@github.com wrote:

I +1 for using typename in place of class in template parameter list wherever possible.

—
Reply to this email directly or view it on GitHub.

[PeterSommerlad]

Or were you only talking about the space between template and < ???

[lichray]

The thread itself is talking about the spacing. I seldom see <typename T>'s usage in the standard, so I suddenly felt too excited...

[zygoloid]

Editorial meeting consensus: use template< not template < (41% vs 59%, but we prefer the former).
Use class not typename unless the example is demonstrating that typename can be used, since class is 2-3 orders of magnitude more common.

[AlisdairM]

Can we confirm the preference is:

template<class T, template class<class U>>
and not
template <typename T,template class <typename U> >

(space and keyword choice)

I will perform a sweep of all library clauses after the Oulu papers land, and apply the preferred rule.
Understood that this may miss the CD, but should be easily landed before ballot resolutions arrive.

My preference would be template <typename T, template typename<typename U>> so it is a good job that I don't get to pick ;)

[tkoeppe]

@AlisdairM: My understanding is that "space after comma" and "closing >>" are uncontroversial and desirable. On the other hand, both openings template< and template < appear in roughly equal numbers and should not be changed at this point.

It's probably best to wait until all the motions have been merged unless you're happy to rebase.

[AlisdairM]

I want to do whatever is required to close the issue - the legwork seems easy, but repetitive :)
Yes, I definitely plan to wait till all the motions have landed, so that I am looking at the full text.

I would prefer guidance to resolve one way or the other on the 'template <' vs. 'template<', but if the conclusion is that both are acceptable, I will look for local consistency in each clause, and leave it at that.

[FrankHB]

I don't like the idea to choose a style merely according to the amount of instances, unless there are no other reasons.

Personally I prefer template< as we usually don't have any whitespace between identifier and lparen of function parameter list in a function declaration. Spaces are also rarely occurred before ( or < of function/template argument lists. This results a simpler rule: just avoid any space before the first token of function/template parameter/argument lists. The current preference of space fits to my flavor:)

BTW, I've ever searched for some tools to format code in this style, but failed. Notably, clang-format had implemented the needed functionality about whitespace after template, but eventually rejected. The reviewer also mentioned the style in the standard. Perhaps they should reconsider it this time.
See also http://stackoverflow.com/questions/34423615/cpp-template-spacing-using-clang-format-3-6-not-working-as-expected.

Note I still prefer typename to ease some refactoring work, as template<typename T, template<typename> class C>. Some more tips:

• Order is significant. Please note the phrase template class is not well-formed here...
• I prefer >> because it looks tightly.
• I don't keep optional parameter names in any parameter list, unless they are necessary (esp. for documentation). Proper naming is difficult.
• I use class only when it should be a "class". For a template template-parameter, it must be a "class" rather than "typename" since ordinary types have no way to be higher-rank polymorphic in C++. Choosing permissive requirements of grammar is an effort to simplify the parser of a conforming implementation rather than direct encouragement to prefer class over typename.

[jwakely]

I have a weak preference for template< without a space, just because it's slightly more compact and so gives us one more character spare. That way we might be able to fit something on one line that needed breaking (or ran into the margin) otherwise.