Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

It's got some useful stuff in it, but doesn't pass ccanlint here.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Using ccanlint's --deps-fail-ignore, we can just test every module
with "make check".  Still exclude them from the 'libccan.a' build
though.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Otherwise ccanlint complains about missing license comments.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Useful for bulk testing modules, where some might not have required
external (non-CCAN) dependencies.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

These version helpers help to compare major.minor style version numbers,
without the need for open-coded and error-prone bitshifting, multiplication,
and similar.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Creating a server for IPv4 and IPv6 has similar issues to clients,
with some novel twists.  Slightly different arguments need to be given
to getaddrinfo(), but worse, some platforms (Linux without
/proc/sys/net/ipv6/bindv6only set) automatically bind IPv6 sockets to IPv4
ports as well.

Thus we need a function which can bind (and listen) to one or two fds.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

list_top + list_del, as suggested by Ben Herrenschmidt.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Operations for moving the entire contents of a list.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

If you have sections, it might still work for some cases though: on x86
Ubuntu it worked once I changed the section name from autodata to xautodata.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

hi all,

here is a trival path to fix warning [-Wunused-but-set-variable].

Thanks.
Tinggong
Signed-off-by: Tinggong <wangtinggong@gmail.com>

hi rusty,

here is a trival path to fix warning [-Wformat].

Thanks.
Tinggong
Signed-off-by: Tinggong <wangtinggong@gmail.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Andreas Schlick reports failure on systems without IPv6 support (in particular
ones with no IPv6 addresses for localhost).  We should be more robust in
these cases, and also where IPv4 support is missing.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Andreas Schlick reported a failure on 64-bit systems; we should simply
test that the number of iterators does not grow on second iteration,
rather than assuming an explicit limit as compilers may vary.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

7 in ~7 needs to be size_t for it to be correct when len > INT_MAX.

A simple implementation compiles very well with gcc and clang on ppc and
x86, and will handle any-aligned input.

"engla" on IRC reported that ccanlint on linux/powerpc would loop
infinitely calling dep_failed after all tests are run and the score is
output.

Reproduced, and discovered that cannot_run() takes a container_of(),
except our top node is not a struct ccanlint.  The result was harmless on
x86, but set dep_failed to the return address on powerpc, causing that
to be called repeatedly.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

This is good form for datastructures, so caller can track weird bugs.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

We use TAL_USE_TALLOC to use libtalloc as the backend: you can test that with
ccanlint --compiler="cc -DTAL_USE_TALLOC".
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

We can't really tell if non-ccan dependencies are used, so don't try.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

David Gibson points out that a project may already be using talloc, so
this gives a simple adapter if one module uses tal.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

It may interfere with other at_exit() calls, so let them call it manually.
Also, use memset to zero, which really does make valgrind notice any leaks.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

In this case, CCAN_CFLAGS should be "".
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Since we never post-add a length property, we can save some cycles by
allocating it together with the object itself:

Before:
	$ ./samba-allocs talloc.dump --tal
	Tal time:                6058997-6215211(6102645)ns
	Tal_free time:           4791112-4968964(4824814)ns
	Single tal_free time:    3173647-3331584(3200183)ns
	$ ./samba-allocs talloc.dump --tal-size
	Virtual size = 9809920, RSS = 3952640

After:
	$ ./samba-allocs talloc.dump --tal
	Tal time:                5911994-6043808(5956914)ns
	Tal_free time:           4676026-4834598(4719908)ns
	Single tal_free time:    2888195-3067365(2922298)ns
	$ ./samba-allocs talloc.dump --tal-size
	Virtual size = 9809920, RSS = 3948544
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Since we added the length property, this penalizes us with an extra
alloc.  Unfortunately, we can't tell from the talloc.dump which are
array tallocs and which are single tallocs, but grepping the sources
gives a rate of 1 in 27.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

The first ok() line was actually failing, because sigaction returns 0
on success, but because the program exit(0) out the abort handler,
that wasn't being seen.

So remove tap and just abort and assert our way through the test,
which I think is clearer.

We use macros to mug the talloc entry points, since talloc_set_allocator()
is not in mainline talloc.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>