It’s important to note that, differently from what we’ve seen for
the serial test harness (see Parallel Test Harness), the
AM_TESTS_ENVIRONMENT and TESTS_ENVIRONMENT variables cannot be use
to define a custom test runner; the LOG_COMPILER and LOG_FLAGS (or
their extension-specific counterparts) should be used instead:
## This is WRONG!
AM_TESTS_ENVIRONMENT = PERL5LIB='$(srcdir)/lib' $(PERL) -Mstrict -w
## Do this instead.
AM_TESTS_ENVIRONMENT = PERL5LIB='$(srcdir)/lib'; export PERL5LIB;
LOG_COMPILER = $(PERL)
AM_LOG_FLAGS = -Mstrict -w
(http://www.gnu.org/software/automake/manual/html_node/Parallel-Test-Harness.html)
"As with the serial harness above, by default one status line is printed
per completed test, and a short summary after the suite has completed.
However, standard output and standard error of the test are redirected
to a per-test log file, so that parallel execution does not produce
intermingled output. The output from failed tests is collected in the
test-suite.log file. If the variable ‘VERBOSE’ is set, this file is
output after the summary."
(http://www.gnu.org/software/automake/manual/html_node/Parallel-Test-Harness.html)
Makefile.am: add valgrind to the AM_TESTS_ENVIRONMENT
configure.ac: enable valgrind testing option for ./configure
m4/valgrind-tests.m4: enable tracing children for libtool wrapper
script compatibility, but ignore standard binaries in /usr or /bin
it turned out to be a bad idea to not export the default allocator's
methods via the ProtobufCAllocator's function pointers. there is at
least one user (protobuf-c-rpc) that makes allocations by directly
invoking those methods.
according to microsoft's platform evangelist, "we recommend that you
consider using a different compiler such as Intel or gcc" if you need a
conforming C compiler. since there's already a project that maintains
the stdint.h / inttypes.h headers for microsoft compilers
(https://code.google.com/p/msinttypes/) there's not much point in
maintaining this ourselves.
there's not much point to having the "private" definitions split out
into a separate header file. they're still in the namespace and there's
nothing that can be done to prevent "unauthorized" uses. just integrate
the definitions into the main header file but put them in the bottom and
note that they're "private".
this makes it very slightly easier to copy the protobuf-c support
library into another project wholesale, since one less file is required.
i'm not quite sure what this thing is used for or how useful it could
possibly be. it's not used anywhere in protobuf-c or protobuf-c-rpc, and
it wasn't in protobuf-c 0.14 or earlier. just delete it; if anything
actually needs this kind of logic, it's easy enough to replicate it.
libraries should never generate output on their own to stdout/stderr.
remove the PRINT_UNPACK_ERRORS macro and rename UNPACK_ERROR to
PROTOBUF_C_UNPACK_ERROR.
the error strings are left in but compiled out by default. they could
theoretically be re-enabled for a debugging session by changing the
PROTOBUF_C_UNPACK_ERROR macro to something like:
#define PROTOBUF_C_UNPACK_ERROR(...) do { fprintf(stderr, __VA_ARGS__); fputc('\n', stderr); } while (0)
some of these headers aren't used in the protobuf-c code base any more,
and in any case the results of these checks (the HAVE_*_H defines in
config.h) are not actually used anywhere and the absence of any of these
headers doesn't cause configure to fail, so just delete these useless
checks.
this reworks memory allocation throughout the support library.
the old DO_ALLOC macro had several problems:
1) only by reading the macro implementation is it possible to tell
what actually occurs. consider:
DO_ALLOC(x, ...);
vs.:
x = do_alloc(...);
only in the latter is it clear that x is being assigned to.
2) it looks like a typical macro/function call, except it alters the
control flow, usually by return'ing or executing a goto in the
enclosing function. this type of anti-pattern is explicitly called out
in the linux kernel coding style.
3) in one instance, setting the destination pointer to NULL is
actually a *success* return. in parse_required_member(), when parsing
a PROTOBUF_C_TYPE_BYTES wire field, it is possible that the field is
present but of zero length, in which case memory shouldn't be
allocated and nothing should actually be copied. this is not apparent
from reading:
DO_ALLOC(bd->data, allocator, len - pref_len, return 0);
memcpy(bd->data, data + pref_len, len - pref_len);
instead, make this behavior explicit:
if (len - pref_len > 0) {
bd->data = do_alloc(allocator, len - pref_len);
if (bd->data == NULL)
return 0;
memcpy(bd->data, data + pref_len, len - pref_len);
}
this is much more readable and makes it possible to write a
replacement for DO_ALLOC which returns NULL on failures.
this changes the protobuf_c_default_allocator to contain only NULL
values; if a replacement function pointer is not present (non-NULL) in
this struct, the default malloc/free implementations are used. this
makes it impossible to call the default allocator functions directly and
represents an API/ABI break, which required a fix to the
PROTOBUF_C_BUFFER_SIMPLE_CLEAR macro.
despite turning one-line allocations in the simple case:
DO_ALLOC(rv, allocator, desc->sizeof_message, return NULL);
into three-line statements like:
rv = do_alloc(allocator, desc->sizeof_message);
if (!rv)
return (NULL);
this changeset actually *reduces* the total number of lines in the
support library.
in general, libraries shouldn't be responsible for terminating the
program if memory allocation fails. if we need to allocate memory and
can't, we should be returning a failure indicator, not providing a
strange interface to the user for receiving a callback in the event of
such an error.
also in general, libraries should never write to stdout or stderr.
this breaks the API/ABI and will require a note in the ChangeLog.
i'm confused as to why these fields exist, since the typical
implementation of a "temporary alloc" would be something like alloca(),
and alloca() is usually just inlined code that adjusts the stack
pointer, which is not a function whose address could be taken.
this breaks the API/ABI and will require a note in the ChangeLog.
possibly we could revisit the idea of "temporary allocations" by using
C99 variable length arrays. this would have the advantage of being
standardized, unlike alloca().
this should silence Coverity #1153648, which complains because
tmp.length_prefix_len is uninitialized for certain wire types when
copied on line 2486:
scanned_member_slabs[which_slab][in_slab_index++] = tmp;
dave's original style drives me crazy. reformat the C code in
protobuf-c/ with "indent -kr -i8" and manually reflow for readability.
try to fit most lines in 80 columns, but due to the lengthy type and
function names in protobuf-c, enforcing an 80 column rule would result
in a lot of cramped statements, so try to fit lines in up to 100 columns
if it would improve readability. (e.g., one <=100 column line is
probably better than 3-4 <=80 column lines.)
ultimately i'd like to adopt most of the recommendations in the linux
coding style: https://www.kernel.org/doc/Documentation/CodingStyle.
this commit gets us most of the kernel indentation and comment coding
style recommendations. later commits will tackle style recommendations
that require more intrusive changes: breaking up large functions,
replacing macros that affect control flow (e.g., DO_ALLOC). this will
hopefully facilitate review and make the code base easier to maintain.
i ran the old and new versions of protobuf-c.c through something like:
gcc -S -D__PRETTY_FUNCTION__=0 -D__FILE__=0 -D__LINE__=0 -Wall -O0 \
-o protobuf-c.S -c protobuf-c.c
and reviewed the diffs of the assembly output to spot any functions that
changed, and went back to make sure that any differences were
functionally equivalent.
