Re: Bison patch to use unlocked I/O if available

[Akim Demaille]

>>> "Paul" == Paul Eggert <address@hidden> writes:

 > Akim Demaille <address@hidden> writes:
 >> The changed order, in particular, is a PITA.  Of course
 >> this misfeature comes from C itself.

 > I agree with you about the order and about the misfeature.
 > But the code already had many uses of putc and puts.

OK.

 > There is some argument for doing things uniformly -- that is, either
 > always use putc/puts when possible, or always use printf.  The code
 > wasn't uniform in that sense.  I chose the former as it is slightly
 > more efficient and it gives a signal to the reader when printf's power
 > isn't needed.  But you could talk me into the latter as well, at least
 > for Bison itself.

OK with me.

 > For data/* and lib/*, though, we're talking about code in other
 > applications.  Wouldn't it be better there to use the
 > higher-performance solution if it's as easy as this?

Of course not, you are totally right for gnulib!  And if the printf's
of the parsers generated by Bison were to be used in something else
than debugging, I would also agree to move to fputs.  But bison's
runtime is dominated by m4's, and the parser's outputs are debugging
traces.