User-defined literals for format and named arguments

[dean0x7d]

This is a bit of fun with C++11 user-defined literals. This brings C++ Format a little closer to Python's str.format for literals:

# Python
"{0}{1}{0}".format("abra", "cad")

// C++11 with user-defined literals
"{0}{1}{0}"_format("abra", "cad");

The literal _format just passes the arguments to fmt::format and is equivalent to fmt::format("{0}{1}{0}", "abra", "cad");

Literals can be used with named arguments as well:

# Python
"Hello, {name}! Your number is {number}".format(name="World", number=1);

// C++11 with user-defined literals
"Hello, {name}! Your number is {number}"_format("name"_a="World", "number"_a=1);

The _a literal just constructs a fmt::arg under the hood.

Compiler support for UDLs is required. Minimum Clang 3.1, GCC 4.7 or VS2015.
To use the literals using namespace fmt::literals; will import just _format and _a. But using namespace fmt; is also a possibility.

[vitaut]

Thanks for the contribution! This looks like a cool even if a bit unconventional use of user-defined literals. I'd like to experiment with this a bit more before merging in, hopefully in the beginning of the next week when I return from the trip I'm currently in.

[vitaut]

@dean0x7d, I get some warnings and errors on GCC 4.8.4 after merging this:

cppformat/test/format-test.cc:1615:427: warning: invalid suffix on literal; C++11 requires a space between literal and identifier [-Wliteral-suffix]
   EXPECT_EQ(format("{first}{second}{first}{third}",
                                                                                                                                                                                                                                                                                                                                                                                                                                           ^
...
cppformat/test/format-test.cc: In member function ‘virtual void LiteralsTest_Format_Test::TestBody()’:
/home/viz/Downloads/3/cppformat/test/format-test.cc:1610:237: error: expected ‘)’ before ‘{’ token
   EXPECT_EQ(format("{}c{}", "ab", 1), "{}c{}"_format("ab", 1));
                                                                                                                                                                                                                                             ^
...

Do you know what might be causing these?

[vitaut]

I've temporarily moved the changes to the udl branch until the above issue is resolved.

[dean0x7d]

This is really strange. I tested this originally with GCC 4.8.4, Clang 3.7 and VS2015. I just tried again with GCC 4.8.4 and I can't reproduce the error. Everything is working fine.

Is it possible that it's some system/library specific preprocessor definition messing with _format?

[vitaut]

Interestingly, the error only occurs in tests. I can compile and run a small standalone example successfully:

#include "format.h"
using namespace fmt::literals;
int main() {
  fmt::print("The answer is {answer}.", "answer"_a=42);
}

And _format is not defined there - I've just checked with an #ifdef and #error.

[vitaut]

Looks like a preprocessor problem, because changing the test to

  std::string str = "{}c{}"_format("ab", 1);
  EXPECT_EQ(format("{}c{}", "ab", 1), str);

compiles.

[dean0x7d]

The gtest macros pass their arguments both as code and as strings using the macro stringize operator (#). It's possible that's where it trips over the literal syntax. But that's still pretty weird. I haven't been able to reproduce the error either in tests or standalone. I tried with GCC 4.8.4 and 5.2.0.

[vitaut]

Yeah, looks like a stringification issue. The following code

#define stringify(x) #x
stringify("{}c{}"_format("ab", 1));

is preprocessed to

""{}c{}"_format(\"ab\", 1)";

Note that the double quotes in "{}c{}" are not escaped.

Here's a relevant bug report: https://gcc.gnu.org/ml/gcc-bugs/2015-04/msg02027.html

I guess a simple solution is to assign the result of "..."_format(...) to a string variable and pass it to EXPECT_EQ as I did in my previous comment.

[dean0x7d]

I made the required changes to the tests. I haven't been able to reproduce the stringification bug on my end, so I can't actually test the workaround but I think this should be fine.

Is there any way to add the relevant commit to this pull request or should I open a new one?

[vitaut]

Normally the PR is updated automatically when you push to correspondent branch, but I guess not for merged PRs. So I just pulled the commit from your branch. As expected the tests compile correctly now (why the issue only reproduces on my machine is a mystery). Thanks!

[vitaut]

@dean0x7d BTW, could you by any chance add some docs describing the new functionality?