Re: pedantic

Re: pedantic

[Mike Stump]

> Date: Sun, 8 Nov 1998 11:04:39 -0800
> To: egcs@cygnus.com
> From: Jason Merrill <jason@cygnus.com>

> I'd like to be able to set the pedantic_errors flag without setting
> pedantic, so that code that is using gratuitously erroneous code that we
> still know how to handle can be an error, but strict standards
> conformance is not enforced.  Thoughts?

I worry about too many different cases.  Without an example of the
case you're thinking about, it is hard to say, but for `gratuitously
erroneous code', an error 100% of the time might not be too bad.  If
you think it might be reasonable to compile it, then, I don't see any
harm in compiling with a warning.  I guess I'd need an example to say
anything worthwhile on the matter.

Re: pedantic

[Jason Merrill]

>>>>> Mike Stump <mrs@wrs.com> writes:

 > I worry about too many different cases.  Without an example of the
 > case you're thinking about, it is hard to say, but for `gratuitously
 > erroneous code', an error 100% of the time might not be too bad.  If
 > you think it might be reasonable to compile it, then, I don't see any
 > harm in compiling with a warning.

The difference is between compiling it because we can and compiling it
because it's reasonable.  I want to offer modes where we compile everything
we can (for people building stuff they got off the net and are not
interested in fixing) and where we only compile reasonable code (for
developers).

Jason

Re: pedantic

[Joern Rennecke]

> Perhaps -permissive, to match the form of -pedantic?

Why, the logical choice is -no-antsy, in accordance with the gcc
future.options file.

Re: pedantic

[Mike Stump]

> From: Jason Merrill <jason@cygnus.com>
> Date: 09 Nov 1998 23:30:41 -0800

> Perhaps -permissive, to match the form of -pedantic?

Nice name.

> The simplest thing is to have it affect flag_pedantic_errors, so
> that -pedantic-errors becomes redundant.  So people wanting pedantic
> warnings would need to say -pedantic -permissive.

> Or -pedantic by itself could turn off flag_pedantic_errors.  That would be
> more consistent with the current behavior.

Sounds good to me.

Re: pedantic

[Jason Merrill]

>>>>> Mike Stump <mrs@wrs.com> writes:

 > What default do you prefer?  I assume that you want to tighten the
 > compiler by default, and have the people compiling random old code use
 > -floose?

Yes.

 > Explained that way, what you said makes perfect sense even without an
 > example, the only thing left is the default mode and the name of the
 > option and individual discussions on what we should put into what
 > class.

Perhaps -permissive, to match the form of -pedantic?

The simplest thing is to have it affect flag_pedantic_errors, so that
-pedantic-errors becomes redundant.  So people wanting pedantic warnings
would need to say -pedantic -permissive.

Or -pedantic by itself could turn off flag_pedantic_errors.  That would be
more consistent with the current behavior.

Jason

Re: pedantic

[Mike Stump]

> From: Jason Merrill <jason@cygnus.com>
> Date: 09 Nov 1998 15:09:14 -0800

> The difference is between compiling it because we can and compiling it
> because it's reasonable.  I want to offer modes where we compile everything
> we can (for people building stuff they got off the net and are not
> interested in fixing) and where we only compile reasonable code (for
> developers).

What default do you prefer?  I assume that you want to tighten the
compiler by default, and have the people compiling random old code use
-floose?  I know some people would welcome the tightening.  g++ is
loose at times (more than it needs to be) for backwards compatibility
and to cater to broken code.

Explained that way, what you said makes perfect sense even without an
example, the only thing left is the default mode and the name of the
option and individual discussions on what we should put into what
class.

Sounds fine to me.

Re: pedantic

[Jason Merrill]

>>>>> Joe Buck <jbuck@Synopsys.COM> writes:

 >> I'd like to be able to set the pedantic_errors flag without setting
 >> pedantic, so that code that is using gratuitously erroneous code that we
 >> still know how to handle can be an error, but strict standards
 >> conformance is not enforced.  Thoughts?

 > I'm not sure I understand: what would the effect be?  I presume certain
 > warnings would become errors, but which ones?  How would this differ
 > from -Werror?

The effect would be to turn pedwarns into errors.  Other warnings would not
be affected, and pedwarns that we only give with -pedantic would also not
be affected.

Jason

Re: pedantic

[David L Nicol]

Mark Mitchell wrote:
> 
> >>>>> "Jason" == Jason Merrill <jason@cygnus.com> writes:
> 
>     Jason> I'd like to be able to set the pedantic_errors flag without
>     Jason> setting pedantic, so that code that is using gratuitously
>     Jason> erroneous code that we still know how to handle can be an
>     Jason> error, but strict standards conformance is not enforced.
>     Jason> Thoughts?

-Wall doesn't do it?  There are identifiable things to warn for that
-Wall cathces not but -pedantic will bail out on?  Sounds reasonable
to this armchair consultant; how about -Wpedantic?


______________________________________________________________________
 David Nicol 816.235.1187 UMKC Network Operations david@news.umkc.edu
  "they cannot truly imagine anyone creating anything genuinely new."

Re: pedantic

[Joe Buck]

> I'd like to be able to set the pedantic_errors flag without setting
> pedantic, so that code that is using gratuitously erroneous code that we
> still know how to handle can be an error, but strict standards
> conformance is not enforced.  Thoughts?

I'm not sure I understand: what would the effect be?  I presume certain
warnings would become errors, but which ones?  How would this differ
from -Werror?

Re: pedantic

[Mark Mitchell]

>>>>> "Jason" == Jason Merrill <jason@cygnus.com> writes:

    Jason> I'd like to be able to set the pedantic_errors flag without
    Jason> setting pedantic, so that code that is using gratuitously
    Jason> erroneous code that we still know how to handle can be an
    Jason> error, but strict standards conformance is not enforced.
    Jason> Thoughts?

I don't really see the point.  We already have several dialects
(-ansi, -pedantic, and the -std= variants for C), plus minor
variations like -fwritable-strings.  I'd like to see us move towards
relatively strict standards conformance, especially for C++, and
gradually remove the backwards compatibility code that allows
non-conforming code to compile.

Perhaps if you give an example motivating your proposal that would
help.

-- 
Mark Mitchell 			mark@markmitchell.com
Mark Mitchell Consulting	http://www.markmitchell.com

pedantic

[Jason Merrill]

I'd like to be able to set the pedantic_errors flag without setting
pedantic, so that code that is using gratuitously erroneous code that we
still know how to handle can be an error, but strict standards
conformance is not enforced.  Thoughts?

Jason