[PATCH]: PR c/22476 -Wmissing-format-attribute [take 4]

[Mark Mitchell]

Joseph S. Myers wrote:

> The point is that this is a new feature (there's an open PR, but it's 
> clearly a feature request rather than a bug), although a reasonably simple 
> and safe one, and we couldn't locate any statement for 4.1 stage 3 (as 
> opposed to 4.0 stage 3) of what latitude is being given to front-end 
> maintainers for new features 

Let me try to be somewhat more clear.

I think it's appropriate for front-end maintainers and back-end 
maintainers to have the authority to occasionally approve patches that 
are not quite within the guidelines for the current development stage.

However, some caveats:

1. The patch should provide significant value.  For example, fixing a 
bug in back end that affects much real code, even though it is not a 
regression, is OK.  But, working hard to pass a corner-case test in a 
C++ conformance testsuite is not.

2. The patch should be considered approximately as safe as other patches 
that would be accepted at that time.  For example, on a release branch, 
we should not be applying patches for non-regressions that we would 
reject if they were for regressions.  I'm not trying to create a vehicle 
to allow people to do rewrites of major functionality on release 
branches or in Stage 3.

3. Caution used should be in proportion to the potential for harm.  That 
implies that maintainers for more obscure components have more latitude 
than those for less obscure components.  In particular, the C front end 
and the IA32 back end are both incredibly heavily used, so introducing 
new bugs there is very severe, and maintainers of these components 
should exercise proportionate caution.

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304