If we can get a backtrace, we can automatically eliminate identical
failures.  Surprisingly backtrace() is quite fast, but the savings for
the (naively-written) rbtree tests are impressive.  ccanlint -v time
drops from 43 seconds to 6 seconds.

When a failtest_hook returns fail_probe(), we start a counter to
continue for a little way.  Replace this hack with a simple flag,
which disables further failures.

Interestingly, this doesn't reduce coverage on the current modules
which use failtest.

This is useful for debugging failtest itself, as well as for things
like tracing.

We use file descriptors inside failtest; use dup2 to try to avoid
changing the file descriptor numbers used in the program (particularly
when programs mess with invalid file descriptors, they're likely to
still be invalid).

This avoids a silly realloc, but more importantly it gets us closer to
being runtime extensible, as each history element can be a different
size.

Less cut & paste means less patching as failtest changes.

failtest_malloc should use p->u.malloc not p->u.calloc.  The layouts
are identical, so it doesn't matter, but it's confusing and leaves us
open to weird bugs in future should one change.

This depends on reduce_features; currently it tends to run before
that, so it doesn't think there are any features to reduce and doesn't
build or run the tests with reduced features.

Saves me doing it manually.

This change initializes a counter of children, otherwise is may contain
arbitrary value.
Signed-off-by: Vladimir Zapolskiy <vz@mleia.com>

As noted by Jan Engelhardt; libHX fixed this already.

The two tests are identical with one line different, so make it clear.

Broken by 758ab1f4 "ccanlint: drop -d, interpret any arguments as
directories, and allow multiple."

Compulsory means "malformed", we might get rid of it altogether, since
any test can mark "fail" and make ccanlint exit with non-zero status.

Now we only have four compulsory tests:
   info_exists               Module has _info file
   depends_exist             Module's CCAN dependencies can be found
   objects_build             Module object files can be built
   module_builds             Module can be built from object files

We currently insist that a tdb file be a version1 file if tdb_open() is
passed the TDB_VERSION1 flag; we fail if it's actually a tdb2.

But that makes generic wrappers harder, and is unlikely to be what the
user wants: if they do, they can check tdb_get_flags() & TDB_VERSION1
after opening.

It seems redundant: strset_test() and strset_clear() can only return NULL
when the string is not a member.  However, it became clear in writing
ccan/tsort that it's much more convenient for callers if we set errno
in this case too, so they can pass it up.

It seems redundant: strmap_get() and strmap_del() can only return NULL
when the string is not a member.  However, it became clear in writing
ccan/tsort that it's much more convenient for callers if we set errno
in this case too, so they can pass it up.

We weren't initializing the leak_info field when the test was marked
FAIL in _info (as in tdb2).

Fallout from commit 758ab1f4
(ccanlint: drop -d, interpret any arguments as directories...)

Actually, I don't even think it means that.  But rename it to something
which is sane.

Thanks to David Gibson for reporting.

Hasn't worked for a while now.  Oops.

This gets us closer to 'ccanlint *' rather than relying on Makefiles
to test all modules.  Unfortunately, because of limited parallelism, it's
currently slower than make -j.

Because we fork children to do compilations, and we use stdio, we need
to flush stdout before the fork otherwise the child will flush
afterwards.  The compile tests interpret this output as a compiler
warning.

This shows up if you redirect ccanlint output to a file.

Commit c4ca9f54 renamed run-83-openhook.c to
api-83-openhook.c, but didn't update the suppression in _info.  My
system runs a patched valgrind, which doesn't have this problem.

Use correct printf string.

Help the compiler eliminate untestable code.

We fake up a dual IPv4/IPv6 response to localhost for testing, but it
assumed localhost6 or ip6-localhost entries.  Fall back to "::1" which
should always work.

gcc (Ubuntu 4.4.3-4ubuntu5) 4.4.3 on x86_64 uses a different stack
offset for the second call to test_ptr_recursion(), meaning it sees
new iterators.  Put it inside a loop to reduce the chance of that
happening.

Very minor fix, since this won't happen unless the test fails.