microseconds formatting requires unicode (again)

[pclouds]

Not sure if I should just reopen #4088 instead. But I wanted to show the ascii version of "us" but setting FMT_UNICODE to zero does nothing (and it makes sense based on fmt/base.h). Perhaps it used to work when utf-8 was disabled by default and the switch breaks it?

[vitaut]

The choice between Unicode and non-Unicode version is done based on whether the literal encoding is UTF-8 which should be consistent with the logic in the standard. You can disable Unicode support on Windows by setting the FMT_UNICODE CMake variable (not macro) to 0.

[pclouds]

Sorry for a very late response. Somehow I missed your reply. This is Linux, not Windows, so I don't think CMake really matters. The cmake comment says

if (NOT MSVC)
  # Unicode is always supported on compilers other than MSVC.

but the compiler support should be a prerequiste, not the decider here. We need to store traces (from FMT) in ascii format and would be ok with disabling unicode support completely. But it seems not supported anymore?

[vitaut]

There is no way to change this globally.

[pclouds]

#4088 mentions 1db2274 which allowed to control this (globally) with FMT_UNICODE. I'm not seeing why FMT_UNICODE is ignored on non-Windows platforms. Perhaps you could elaborate a bit?

I think it's a valid use case to let people decide if unicode is wanted or not, even if the compiler supports it. And if this cannot be supported, it may be less confusing if -DFMT_UNICODE=OFF is rejected on non-Windows.

[vitaut]

This macro is limited to Windows to support legacy code that relies on codepages. Even on Windows it is of little value because some formatters will produce Unicode anyway.