[PATCH] clang-format: options for spacing template argument lists

[Daniel Jasper]

On Mon, Oct 28, 2013 at 5:29 PM, Christopher Olsen <chrisaolsen at gmail.com>wrote:

> Thanks for the feedback.  I've updated the patch:
>
> Added the flag SpacesInAngles - A<int> vs A< int >
> Note that LanguageStandard=Cpp03 overrides SpacesInAngles=false in the
> case of '>>' as in A<A<int> >
>
> While I tend to agree with you on spacing in template arguments I have
> 5+ million lines of code developed over 12 years in strict "pad
> everything" style to contend with.  Honestly, I'm hoping clang-format
> will eventually allow us to re-evaluate certain style choices if it
> gets adopted on our project.
>

As I said, I can perfectly understand the necessity for SpacesInAngles
(both because there are large codebase where consistency matters and that
it is a strong matter of what one is used to). So this option is totally
reasonable. Sorry if I was unclear.

The patch looks fine except that there doesn't seem to be a test for the
parsing of the configuration option. That should also be done in
unittests/Format/FormatTests.cpp along with the tests for parsing all the
other options. Do you have commit access or should I submit this patch for
you?

Cheers,
Daniel

Chris
>
> On Sun, Oct 27, 2013 at 4:32 AM, Daniel Jasper <djasper at google.com> wrote:
> > The patch itself looks fine. However, I am also slightly skeptical that
> it
> > is a good general direction.
> >
> > Specifically:
> > - SpaceInEmptyAngles: Quite frankly, it doesn't matter enough. Nobody
> really
> > cares (or should care about this). I'd make a decision and get on with
> life.
> > If you need that space for your codebase, make it dependent on
> > SpacesInAngles.
> > - SpaceAfterTemplateKeyword: Same as above, it doesn't matter enough. We
> > actually had a discussion about this very early into the clang-format
> > development. People don't agree, but it really does not matter (no
> version
> > is more or less readable, ..). We then actually went ahead (IIRC) to
> change
> > all instances in the C++11 standard to do exactly what clang-format does
> > now. I'd rather have clang-format force people into a consistent behavior
> > for such non-issues than add additional technical complexity to it.
> > - SpacesInAngles: I would personally be appalled by having to work in a
> > codebase that uses this. Constructs like "template < class T..." simply
> look
> > too much like comparisons to me. However, I can see that, if you are
> used to
> > it, switching might be hard. The overlapping of the language standard is
> > unfortunate, but I don't really see what we can do here.
> >
> > So, I am more or less fine with introducing SpacesInAngles (goes well
> with
> > SpacesInParentheses), but I'd rather not introduce the other two (I know
> > that we have SpaceInEmptyParentheses - I also don't like that one, but
> "()"
> > is way more common than "<>", so it matters a bit more).
> >
> > I also agree that, at the very least, we need some grouping of the
> options
> > soon. I have not found a way to specify style options more intuitively.
> And
> > to some extend, I think people having to carefully look through the
> options
> > or accept a default is reasonable.
> >
> > Cheers,
> > Daniel
> >
> >
> > On Fri, Oct 25, 2013 at 11:26 PM, Sean Silva <silvas at purdue.edu> wrote:
> >>
> >>
> >>
> >>
> >> On Fri, Oct 25, 2013 at 4:00 PM, Christopher Olsen <
> chrisaolsen at gmail.com>
> >> wrote:
> >>>
> >>> I am evaluating clang-format and need control over spacing in template
> >>> argument lists to conform to our coding standards.
> >>>
> >>> I've attached a patch that adds new flags to control this spacing.
> >>> Unittests are also included.
> >>>
> >>> - SpacesInAngles - A<int> vs A< int >
> >>> - SpaceInEmptyAngles - template <> vs template < >
> >>> - SpaceAfterTemplateKeyword - template<typename T> vs template
> <typename
> >>> T>
> >>>
> >>> Note that LanguageStandard=Cpp03 overrides SpacesInAngles=false in the
> >>> case of '>>' as in A<A<int> >
> >>>
> >>> Please let me know if there is anything I need to fix for submission.
> >>
> >>
> >> I foresee (and have already run into multiple real use cases) where the
> >> spacing before, after, in between, etc. of many different kinds of
> tokens
> >> needs to be tweaked. For example:
> >>
> >> * multiplicative operators don't have spaces, but additive operators do
> `a
> >> + b*c`
> >> * spaces before parenthesized lists in function calls `foo (bar)`
> >> * spaces inside the parentheses of an `if`: `if ( cond ) {`
> >> * "function-like" return: `return(3)`
> >>
> >> I think we should try (though it may not be realistic) to integrate this
> >> functionality in a way that covers the different use cases and makes
> them
> >> interact in a consistent and understandable way; otherwise we will just
> end
> >> up growing a forest of not-easy-to-discover options that don't have very
> >> good coverage of the configuration space for the next project. For
> example,
> >> adapting clang-format to the OP's project coding standards requires
> adding 3
> >> new options; is there a realistic upper bound on the number of such
> options
> >> that we will need in order for clang-format to support, say, 1000
> different
> >> projects from 50 different companies/open-source communities?
> >>
> >> One possibility that I can imagine (although I don't know how feasible
> it
> >> is) is to ship another tool (or more likely keep it under an option to
> >> clang-format) which does a "one time" analysis to determine a set of
> >> parameters that will conform with a given sample source file (or files)
> and
> >> emits a configuration file. This analysis could work with a larger (but
> very
> >> consistent and well-defined) "plumbing" configuration space that
> essentially
> >> parameterizes the "guts" of clang-format (such as `spaceRequiredBefore`,
> >> `spaceRequiredBetween`, `splitPenalty`, etc.) in a data-driven way.
> >>
> >> On the other hand, I think it has been put out there before is that one
> of
> >> the benefits of clang-format is to help "standardize" to some extent on
> a
> >> common subset of style options, and so providing too-fine-grained
> support
> >> for "tweaking" the output might be undesirable from such a perspective.
> On
> >> the other hand, if clang-format wants to "dominate the world", it can't
> >> impose arbitrary changes on a project's coding style. A poignant
> question is
> >> "is it a goal for clang-format be able to conform with
> >> more-or-less-arbitrary styles without requiring requiring the users to
> get
> >> involved with clang-format development?", but I can't speak as to the
> >> answer.
> >>
> >> Daniel, what do you think?
> >>
> >> -- Sean Silva
> >>
> >>>
> >>>
> >>> Thanks!
> >>> Chris
> >>>
> >>>
> >>> _______________________________________________
> >>> cfe-commits mailing list
> >>> cfe-commits at cs.uiuc.edu
> >>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
> >>>
> >>
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20131028/da80eba2/attachment.html>