Re: Surprising "parse error in template argument list"

[Jonathan Wakely <jwakely dot gcc at gmail dot com>]

On 22 January 2016 at 17:11, John Marshall wrote:
> The following code:
>
>         template<class Tp, int Nm> inline Tp* end(Tp (&arr)[Nm]) { return arr + Nm; }
>
>         struct S { int end; };
>         bool negative(const S& s) { return s.end < 0; }
>
>         template <typename T> bool tnegative(const T& t) { return t.end < 0; }
>
> produces a parse error misrecognising the final line's "end < 0" as a template:
>
> end3.cpp: In function âbool tnegative(const T&)â:
> end3.cpp:6:61: error: parse error in template argument list
>  template <typename T> bool tnegative(const T& t) { return t.end < 0; }
>                                                              ^~~~~~~
>
> This message is from current trunk GCC, but similar error messages are produced by GCC 4.4 and 5.1.
>
> It seems clear to me that "t.end < 0" isn't supposed to be any kind of template invocation, so I was surprised by this message.  Is there some standardese that I've missed that requires this to be parsed like this in a template function?  Or is this a G++ parser bug?  (I did a brief bugzilla search, but nothing similar leapt out at me.)

This is a G++ bug, I'll report it to Bugzilla. Thanks for the minimal testcase.

If t.end< was supposed to be the start of a template then it would be
necessary to say "t.template end<" instead, to tell the parser to
expect a template. In the absence of the "template" keyword G++ should
assume otherwise, and parse it successfully as a less-than expression.

>
> (This is distilled down from https://bugs.debian.org/812294 in which code has an "end" field and is "using namespace std", and it appears that with GCC 6's C++14-by-default the field is colliding with some recentish std::end().  So various workarounds are available, probably starting with not using namespace std!)

A workaround would be to change it to return (t.end) < 0;. The
parentheses would stop the parser from doing the wrong thing.