This allows GDB to print values, but also allows us to use them in
'case' statements. This wasn't allowed before because they're not
constant terms.
This also made it clear there's a clash between two error codes,
so move one.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changelog-Changed: JSON-RPC: Error code from bcli plugin changed from 400 to 500.
This is usually fine, but without this, commando (another branch!) has
a race:
1. A command has multiple parts.
2. We start sending them out.
3. We get a response, which completes the cmd.
4. We go to send the next one out, but it's been freed.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
And fold start_json_request() and start_json_rpc() into the core
function since it's the only caller.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Build them from the command which caused them, and take plugin name
as basename with extension stripped.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We were cmd was getting free'd but holding on to reference of the
thing was causing problems.
==523280== Invalid read of size 8
==523280== at 0x1B3E14: del_notifier_property (tal.c:326)
==523280== by 0x1B3E14: tal_del_notifier_ (tal.c:569)
==523280== by 0x1123E7: handle_rpc_reply (libplugin.c:671)
==523280== by 0x1123E7: rpc_read_response_one (libplugin.c:866)
==523280== by 0x1123E7: rpc_conn_read_response (libplugin.c:886)
==523280== by 0x1A7B53: next_plan (io.c:59)
==523280== by 0x1A7B53: do_plan (io.c:407)
==523280== by 0x1A7B53: io_ready (io.c:417)
==523280== by 0x1A9BDB: io_loop (poll.c:453)
==523280== by 0x1141D0: plugin_main (libplugin.c:1708)
==523280== by 0x10D7E4: main (commando.c:937)
==523280== Address 0x52de928 is 8 bytes inside a block of size 40 free'd
==523280== at 0x483F0C3: free (vg_replace_malloc.c:872)
==523280== by 0x1B2CDD: del_tree (tal.c:419)
==523280== by 0x1B37BB: tal_free (tal.c:486)
==523280== by 0x1B37BB: tal_free (tal.c:474)
==523280== by 0x110CB2: command_complete (libplugin.c:255)
==523280== by 0x110CB2: command_done_err (libplugin.c:390)
==523280== by 0x10F511: handle_reply (commando.c:560)
==523280== by 0x10F511: handle_custommsg (commando.c:609)
==523280== by 0x113877: ld_command_handle (libplugin.c:1441)
==523280== by 0x113877: ld_read_json_one (libplugin.c:1491)
==523280== by 0x113877: ld_read_json (libplugin.c:1511)
==523280== by 0x1A7B53: next_plan (io.c:59)
==523280== by 0x1A7B53: do_plan (io.c:407)
==523280== by 0x1A7B53: io_ready (io.c:417)
==523280== by 0x1A9BDB: io_loop (poll.c:453)
==523280== by 0x1141D0: plugin_main (libplugin.c:1708)
==523280== by 0x10D7E4: main (commando.c:937)
==523280== Block was alloc'd at
==523280== at 0x483C855: malloc (vg_replace_malloc.c:381)
==523280== by 0x1B3BBD: allocate (tal.c:250)
==523280== by 0x1B3BBD: add_notifier_property (tal.c:303)
==523280== by 0x1B3BBD: tal_add_destructor2_ (tal.c:529)
==523280== by 0x110725: jsonrpc_request_start_ (libplugin.c:181)
==523280== by 0x10E0EA: send_more_cmd (commando.c:643)
==523280== by 0x11243C: handle_rpc_reply (libplugin.c:696)
==523280== by 0x11243C: rpc_read_response_one (libplugin.c:866)
==523280== by 0x11243C: rpc_conn_read_response (libplugin.c:886)
==523280== by 0x1A7B53: next_plan (io.c:59)
==523280== by 0x1A7B53: do_plan (io.c:407)
==523280== by 0x1A7B53: io_ready (io.c:417)
==523280== by 0x1A9BDB: io_loop (poll.c:453)
==523280== by 0x1141D0: plugin_main (libplugin.c:1708)
==523280== by 0x10D7E4: main (commando.c:937)
==523280==
{
<insert_a_suppression_name_here>
Plugins are supposed to store their data in the datastore, and commando does so:
let's make it easier for them by providing convenience APIs.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Rather than a generic "add member", provide two routines: one which
doesn't quote, and one which does.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
The direction only depends on the ordering between node_ids, not the
short_channel_id, so we can include it and it won't change. This was
causing some trouble loading the `channel_hints` in the `pay` plugin.
As per proposal in https://github.com/lightning/bolts/pull/962
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changelog-Removed: protocol: support for legacy onion format removed, since everyone supports the new one.
1. We don't keep a pointer to payments (unlike pay, which keeps a
linked list), so mark it notleak.
2. plugin->our_features is overloaded for "features we want to set" (used by keysend)
and then "features we have". Create a new field, which is cleaner.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
And turn "" includes into full-path (which makes it easier to put
config.h first, and finds some cases check-includes.sh missed
previously).
config.h sets _GNU_SOURCE which really needs to be done before any
'#includes': we mainly got away with it with glibc, but other platforms
like Alpine may have stricter requirements.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We always allocate a new `struct command` when we get a full JSON
object from stdin:
b2df01dc73/plugins/libplugin.c (L1229-L1233)
If it happens to be a notification, we pass the `struct command` to
the handler, and not free it ourselves:
b2df01dc73/plugins/libplugin.c (L1270-L1275)
There are only nine points in `plugins/libplugin.c` where we `tal_free`
anything, and only one of them frees a `struct command`:
b2df01dc73/plugins/libplugin.c (L224-L234)
The above function `command_complete` is not appropriate for
notification handlers; the above function sends out a response
to our stdout, which a notification handler should not do.
However, as-is, it does mean that notification handling leaks
`struct command` objects, which can be problematic if we ever
have future built-in plugins which are significantly more
dependent on notifications.
This commit changes notification handlers to return
`struct command_result *`, because possibly in the future
notification handlers may want to perform `send_outreq`, so we
might as well use our standard convention for callbacks, and
to encourage future developers to check how to properly
terminate notification handlers (and free up the
`struct command`).
We also now provide a `notification_handled` function which a
notification handler must eventually call, as well as a
`notification_handler_pending` which is just a snowclone of
`command_still_pending`.
Before:
Ten builds, laptop -j5, no ccache:
```
real 0m36.686000-38.956000(38.608+/-0.65)s
user 2m32.864000-42.253000(40.7545+/-2.7)s
sys 0m16.618000-18.316000(17.8531+/-0.48)s
```
Ten builds, laptop -j5, ccache (warm):
```
real 0m8.212000-8.577000(8.39989+/-0.13)s
user 0m12.731000-13.212000(12.9751+/-0.17)s
sys 0m3.697000-3.902000(3.83722+/-0.064)s
```
After:
Ten builds, laptop -j5, no ccache: 8% faster
```
real 0m33.802000-35.773000(35.468+/-0.54)s
user 2m19.073000-27.754000(26.2542+/-2.3)s
sys 0m15.784000-17.173000(16.7165+/-0.37)s
```
Ten builds, laptop -j5, ccache (warm): 1% faster
```
real 0m8.200000-8.485000(8.30138+/-0.097)s
user 0m12.485000-13.100000(12.7344+/-0.19)s
sys 0m3.702000-3.889000(3.78787+/-0.056)s
```
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changelog-Fixed: Plugins: C plugins would could leak memory on every command (esp. seen when hammering topology's listchannels).
We weren't actually getting the last log out; this does that.
We have to fix test_bitcoin_failure which now notices the BROKEN
log message.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changelog-Fixed: libplugin: Fatal error messages from plugin_exit() now logged in lightningd.
As mentioned in previous commits: "result" must be an object,
and anything else is an antipattern.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This makes for more useful errors. It prints where it was up to in
the guide, but doesn't print the entire JSON it's scanning.
Suggested-by: Christian Decker
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
