Re: bison 2.5 on sparc solaris 10

[Joel E. Denny]

On Sun, 21 Aug 2011, Joel E. Denny wrote:

> If I also use -pedantic when compiling bison, I see two problems in 
> bison's source.  The first is an assert whose argument is too long for a C 
> string.  That's easy enough to work around.  The second problem comes from 
> gnulib: bison's src/scan-code.l uses gnulib's c_isdigit, which uses gcc's 
> statement expression extension, which also appears in many other places in 
> gnulib.  I find that surprising given that one of the purposes of gnulib is 
> to improve portability.

I misunderstood the gnulib problem.  Statement expressions are only used 
if __GNUC__ is defined, so they do not create a portability problem.

Even if I specify -U__GNUC__ to disable gnulib's use of statement 
expressions, there are more such issues with no workaround that's 
immediately obvious to me.

Until someone reports an actual portability problem with bison's source 
code that -pedantic would have caught, I'm not going to worry about 
routinely checking bison's compilation with -pedantic.  Using -pedantic in 
the test suite is sufficient for the bug reported at the beginning of this 
thread.