and there's no intermediate sequence point! We actually hit this on
sh3 with -O2 where gcc4 caches tlst in a register prior to recursive
call to inptype() and if you are unlucky the recursive call needs to
realloc tlst.
Introduce a temp variable to force a sequence point.
INTPTR_IS_LONG, not INTPTR_IS_ULONG -- the latter is unused in
other parts of lint's code. This stops vax's lint from complaining
about conversion of integer constants to 'unsigned long' in function
argument lists, via a proper define of INT_RSIZE in common/inittyp.c.
sh3 defined this to 0, so was actually not affected, but better to
eradicate the typo there as well.
Remove the apparently always true "styp(nt) != SHORT" part of the
innermost test. Allow atomatic conversion of literals as long as
they fit into the target type.
This should fix some of the lint issues in proplib on some of our
platforms.
Approved by christos.
targets integer data type value ranges. For now we just use the
hosts uint64_t for parsing & storing integers constants, and test
against the targets limits and assign appropriately, instead of
sometimes (inappropriately) going via the hosts u_long type. As
long as none of our architectures have target long or quad data
types strictly larger than 64 bits, we should be fine with this
fix.
Furthermore, as they stand at the moment, we can't use the current
TARG_INT_MAX and TARG_LONG_MAX constants in C preprocessor expressions,
so remove the conditional on them being equal. Yes, this will
produce dead code for some targets.
This allows an ilp32 host to lint for an lp64 target which uses
e.g. the targets ULONG_MAX constant without triggering an "integer
constant out of range" warning.
OK'ed by christos.
Since type_t is different between lint1.h and lint2.h include the
appropriate file depending on the pass. Make the argument mismatch
error print the type names of the types involved. Now that we have
a tyname() function we can fix the rest of the pass2 warnings to be
more explanatory, but not now.
Stops warnings about pre-processor constructs like #elif - which there is
no point detecting now that we've changed much of the code to require an
ANSI C compiler.
_NETBSD_SOURCE as this makes cross building from older/newer versions of
NetBSD harder, not easier (and also makes the resulting tools 'different')
Wrap all required code with the inclusion of nbtool_config.h, attempt to
only use POSIX code in all places (or when reasonable test w. configure and
provide definitions: ala u_int, etc).
Reviewed by lukem. Tested on FreeBSD 4.9, Redhat Linux ES3, NetBSD 1.6.2 x86
NetBSD current (x86 and amd64) and Solaris 9.
Fixes PR's: PR#17762 PR#25944
* Rename "config.h" to "nbtool_config.h" and
HAVE_CONFIG_H to HAVE_NBTOOL_CONFIG_H.
This makes in more obvious in the source when we're using
tools/compat/config.h versus "standard autoconf" config.h
* Consistently move the inclusion of nbtool_config.h to before
<sys/cdefs.h> so that the former can provide __RCSID() (et al),
and there's no need to protect those macros any more.
These changes should make it easier to "tool-ify" a program by adding:
#if HAVE_NBTOOL_CONFIG_H
#include "nbtool_config.h"
#endif
to the top of the source files (for the general case).
* Don't bother prefixing commands with a line of ${_MKCMD}\
and instead rely upon "make -s". This is less intrusive on
all the Makefiles than the former. Idea from David Laight.
* Rename the variables use to print messages. The scheme now is:
_MKMSG_FOO Run _MKMSG 'foo'
_MKTARGET_FOO Run _MKMSG_FOO ${.TARGET}
From discussion with Alistair Crooks.
* DPSRCS contains extra dependencies, but is _NOT_ added to CLEANFILES.
This is a change of behaviour. If a Makefile wants the clean semantics
it must specifically append to CLEANFILES.
Resolves PR toolchain/5204.
* To recap: .d (depend) files are generated for all files in SRCS and DPSRCS
that have a suffix of: .c .m .s .S .C .cc .cpp .cxx
* If YHEADER is set, automatically add the .y->.h to DPSRCS & CLEANFILES
* Ensure that ${OBJS} ${POBJS} ${LOBJS} ${SOBJS} *.d depend upon ${DPSRCS}
* Deprecate the (short lived) DEPENDSRCS
Update the various Makefiles to these new semantics; generally either
adding to CLEANFILES (because DPSRCS doesn't do that anymore), or replacing
specific .o dependencies with DPSRCS entries.
Tested with "make -j 8 distribution" and "make distribution".
- add a new INTPTR_IS_LONG define and use it.
- XXX: the PTRDIFF, SIZEOF, INTPTR defines really make lint more relaxed
in some platforms than others. We should really be looking for the
particular tokens to enable this kind of checking. I.e.
now:
char *p;
int foo = (int)p;
does not produce a warning on INTPTR_IS_LONG == 0 platformas.
In reality it should only elide the warning if:
char *p;
int foo = (intptr_t)p;
but it is not that smart (yet).
expressions.
Details:
lint did fail on constructs like
struct foo *x;
x = ({ struct foo *y; /* do stuff */; y;});
since it gave the whole ({ }) the same structure representing the
type as y, but that structure is reclaimed when y goes out of scope.
2) inline is acceptable in c99 -- create a new c99 keyword class.
XXX The handling of sflag and Sflag is utterly bogus throughout this
pass. I think I have to make some adjustments.
C99 support.
- turn lerror() into a macro so that the filename and the line number of the
error are printed before we abort.
- recurse in type printing to provide the proper type name.
(generally it's 20k). Adjust mblklen temporarily to the size of the block
required and allocate one. This avoids coredumps when mapping in identifiers
that have huge values. (In my example it was a char[] for a 640k pixmap).
round has been tested on Solaris/x86 and Linux hosts.
* Add host tools cap_mkdb, ctags, m4, uudecode.
* Protect __RCSID() and __COPYRIGHT() better.
* Reduce the number of places that need to include "config.h", to keep
sources closer to their "vanilla" versions.
* Add more compat #defines and autoconf-checked functions.
models of type sizes that we currently support, and include
the appropriate one in each arch's targparam.h.
* Use the type size constants provided by targparam.h in the
type table, rather than using "sizeof(type) * CHAR_BIT" (which
would get the host's type size, not the target's). XXX Not
yet done for floating point types.
* Add a new BITFIELDTYPE lint comment that suppresses illegal
bitfield type errors if the type is an integer type (e.g.
long, long long), and also suppresses non-portable bitfield
type warnings.
overriding FD_SETSIZE. Not overriding it makes it stomp all over memory
(which caused the debug outputs we've seen lately).
It used to work, but toolification of lint broke it.
XXX Note that the overflow code for many cases seems to be buggy. I've
only fixed one bug that was bothering me. A set of regression tests
and extensive testing are needed.
* Rewrite src/tools Make logic to work like the rest of the tree wrt
"dependall" and "install". The old "make build" hack is gone.
* Remove the MKTOOLS logic. This was linked to the "make build" hack,
and was only needed because TOOLDIR originally had no writable default.
* Redo the GNU configure/make logic to make it fit reasonably in a
BSD make wrapper. Use new ${.ALLTARGETS} variable to scan for
targets in $(srcdir), and mark them with .MADE: to prevent rebuilding.
* Only build cross tools in src/tools; remove some messy logic in
src/usr.* and src/gnu/usr.* that would do target filename rewriting
(improves consistency and readability).
* Add the ability to build cross gdb at tool build time by setting
MKCROSSGDB (default no) to "yes" in mk.conf.
* Add src/tools/groff and set up paths to work with this cross groff.
pull the correct one in based on the MACHINE_CPU variable.
MACHINE_CPU will be set according to the target system we are
building for by <bsd.own.mk>.
One component of addressing bin/14280.
- convert to ANSI KNF
- remove trailing whitespace
- translate some comments from german into english
code compiles and runs clean, and tested by running "make lint" against
xlint source using previous and this lint produces same results.
foo((sockaddr *(void *))0);
This fix is imperfect, because right now we just check the subtype
chains for NULL and we return to the caller when the loop ends, leaving
the upper layers to cope with the syntax error. Ideally we should:
a.) return an error to the upper layer, or
b.) not call the type analysis routines in the presence of a syntax
error.
That would require a significant re-write which would take much more time
than I have...
__attribute__ and __extension__ workarounds.
Our invariant is: No gcc extensions if __GNUC__ is not defined, so lint
should not be playing around trying to pretend it is gcc.
at the lexical analysis stage not at the syntax parsing stage. The upshot
is that:
if (expr) {
stmt1;
}
/* LINTED */
stmt2;
stmt2 is the look-ahead token for the parser to choose between and if-then-else
statement as opposed to an if-then statement. Unfortunately the side effect
is that the LINTED directive gets reset before stmt2 gets parsed. We fix this,
by remembering the the linted directive during the if statement parsing and
restoring it at the appropriate time.
place" (e.g., when DESTDIR is set). This causes the lint driver to
pass -nostdinc -idirafter <dir> to cpp, causing it to ignore
/usr/include and look somewhere else instead..
