if compiling with -fformat-extensions). Gcc's format checker never
actually supported %q length specifiers. It treats %q as an alias
for %ll, which is correct if quad_t is long long (e.g., on i386's)
and broken otherwise (e.g., on alphas).
quad_t's currently should be printed in the same way that they
already need to be printed to avoid compiler warnings on all
supported systems: cast them to a standard type that is at least
as large (long or long long) and use the length specifier for that
(%l or %ll). This is problematic since long long isn't standard
yet. C9x's intmax_t should be implemented soon.
Don't accept %L length specifiers in the kernel either. The only
legitimate ones are for long doubles, but the kernel doesn't even
support plain doubles. (gcc bogusly accepts %Ld as an alias for
%lld, and it sometimes prints "q" in error messages about "ll" and
"L" length specifiers, becauses it represents all these specifiers
as 'q'.)
Submitted by: bde
- plain %r and %z were disallowed. The hard NULs in the warnings were
hopefully caused by disallowing of plain formats being nonsense.
- new formats for shortening to a byte were allowed, but even the libc
printf doesn't support them.
- old %hr and %hz formats were allowed, but the kernel printf doesn't
support them. The kernel doesn't support %hd either, but this is
harder to fix.
Submitted by: bde
This is enabled by the undocumented option -fformat-extensions.
This option should be named better and/or give more control over
the extensions.
Fixed a message - don't warn about the field width when it's the
precision that has the wrong type. Didn't fix excessive checking
for the precision relative to the type - ANSI requires both to be
ints, but gcc permits the field width to be either int or unsigned
int.
attribute. It is like the existing "printf" archetype, except that
it doesn't complain if the format string is a null pointer. See
the node "Function Attributes" in the GCC info pages if you don't
know what this is all about.
This change will allow us to add format string checking for the
err(3) family of functions.
non-i386, non-unix, and generatable files have been trimmed, but can easily
be added in later if needed.
gcc-2.7.2.1 will follow shortly, it's a very small delta to this and it's
handy to have both available for reference for such little cost.
The freebsd-specific changes will then be committed, and once the dust has
settled, the bmakefiles will be committed to use this code.
