Re: Warnings, standards and flags

[Tim Prince <tprince at myrealbox dot com>]

Terry Frankcombe wrote:
> Hi folks
>
> Two (and a half?) specific questions, and a more general discussion
> point:
>
> Is there any good reason (or at least a rationale) that -std does not
> imply -pedantic?
>
> Why does -Wtabs work with -std=gnu, but -std=f95 not allow the
> over-ride?
>
>
> More generally, gfortran---and indeed all of gcc---seems to like to make
> it really hard for a user to do some fairly obvious things without using
> 36 dozen flags.  Is this considered to be a good thing?  Why are "meta
> flags" not given higher priority?  (While this is somewhat
> understandable given the pedigree of gcc, it's frustrating for users
> coming to gcc after becoming used to broad-brush behaviour from other
> specific compilers.  Of course sensibility is in the eye of the
> beholder, but surely the above are two examples where the intent of the
> user is reasonably obvious, but apparently not really considered by
> developers?)
>
>   
The "fairly obvious things" which commercial compilers target with 
meta-flags don't extend much beyond SPEC benchmarking.  Supposing we 
compare typical options:

ifort [10.0]: -xW -assume protect_parens,byterecl,buffered_io -O2 -g

gfortran -funroll-loops -ftree-vectorize -O2 -g

I don't see much difference in space consumed, except that I don't have 
an option to set abrupt underflow in gfortran (other than -ffast-math).  
In practice, I would also want an option to fuse loops more aggressively 
in ifort; I don't know of such an option for gfortran.
While ifort defaults to -O2 in the absence of -g, depending on that 
default has not proven consistently practical.