This is the mail archive of the
mailing list for the glibc project.
Re: -Werror policy
- From: Roland McGrath <roland at hack dot frob dot com>
- To: Joseph Myers <joseph at codesourcery dot com>
- Cc: <libc-alpha at sourceware dot org>
- Date: Mon, 24 Nov 2014 14:18:10 -0800 (PST)
- Subject: Re: -Werror policy
- Authentication-results: sourceware.org; auth=none
- References: <alpine dot DEB dot 2 dot 10 dot 1411132331520 dot 5050 at digraph dot polyomino dot org dot uk> <20141114001831 dot 8C7CF2C3B16 at topped-with-meat dot com> <alpine dot DEB dot 2 dot 10 dot 1411140028450 dot 5050 at digraph dot polyomino dot org dot uk>
> On Thu, 13 Nov 2014, Roland McGrath wrote:
> > However, I think we should probably write some internal macro using _Pragma
> > that becomes the idiom rather than the whole push/pop boilerplate. We
> > might even make the "mandatory comment" items be (ignored) string arguments
> > to the macro, so it is quite hard to accidentally fail to supply the
> > information.
> I think string arguments would likely be cumbersome when you have
> explanations that may be more than one line long.
Well, it was just the first thought. The point is to have something that
serves as a reminder to write enough commentary. It doesn't really matter
how much of the commentary is in an ignored string and how much is in
nearby comments. The main thought I was having there is that something
syntactically mandatory would be more helpful than a mere policy about
having comments, in making sure we don't let something through review
absentmindedly without requiring adequate comments to justify the warning
> > We should never use per-file -Wno-* options in makefiles. If there is a
> > file-wide issue, we can just put:
> > #pragma GCC diagnostic ignored "-Wfoo"
> > near the top of the file (but usually after all its #include lines so it
> > does not affect code from headers). In the same vein, but separate from
> Not in the case of files imported verbatim from elsewhere (though I don't
> know if the timezone/ files actually still need -Wno-strict-prototypes,
> and -fwrapv in one case).
If we have cases where we really want a file verbatim from a noncooperating
elsewhere (i.e. not just gnulib or some other such place where they are
happy to insert #ifdef _LIBC stuff), then IMHO little wrapper files that
#include the verbatim file are still better than makefile-only magic
implicitly tied to the contents of the source file.
> > -Werror issues, we should have a long-term goal of eliminating all per-file
> > flags settings in favor of pragmas and/or attributes in the source files.
> I'm not sure the pragmas / attributes work for things such as -fexceptions
> (at least, they are only intended for specific cases such as diagnostics
> and optimization options, not for completely arbitrary options).
We can figure out what our options are in all the cases when we get around
to attempting the cleanup.