Re: [boost] [config] Rethinking feature macros?

Subject: Re: [boost] [config] Rethinking feature macros?
From: Peter Dimov (lists_at_[hidden])
Date: 2017-11-06 21:55:34

Andrey Semashev wrote:

> Dead compilers are not the problem.

Okay, I guess. This was my first argument in favor of positive macros, that 
old compilers would require no maintenance.

My second argument in favor of positive macros was that they degrade more 
gracefully when one is using an older Boost.Config. This happens in several 
scenarios; first, in the course of normal development, when your library 
needs a macro but Boost.Config doesn't have it yet; second, when you merge 
to master but Config hasn't been merged yet; third, when a user upgrades 
your library in an older Boost release to take advantage of bugfixes.

Since your typical code is something like

    // old things

    #if (new feature is available)

    // new things

    #endif

with a positive macro and an older Config the new things don't get included 
even when the compiler supports the new feature, which leaves the user not 
worse than before.

With a negative macro, the new portion is compiled and errors out when the 
compiler does not support the new feature.

Steven correctly points out that this is not necessarily a benefit as it 
allows one to forget to include Config (or perhaps allows new code to pick 
up an old Config via a system include instead of the correct one?)

But from where I sit, this seems an acceptable tradeoff.

Either way, judging by the reaction and the arguments offered, a switch to 
positive macros would be unlikely at this time. :-)