Recently usage of undeclared functions became an error rather than a
warning, in C2x. https://reviews.llvm.org/D122983?id=420290
This change has migrated into the build tools of Ubuntu 23.10 which now
causes the USDT tests to fail to compile, see
https://github.com/bitcoin/bitcoin/issues/28600
Fix this by setting `-Wno-error=implicit-function-declaration` for the
tracing programs.
9f55773a37 test: refactor: usdt_mempool: store all events (stickies-v)
bc43270450 test: refactor: remove unnecessary nonlocal (stickies-v)
326db63a68 test: log sanity check assertion failures (stickies-v)
f5525ad680 test: store utxocache events (stickies-v)
f1b99ac94f test: refactor: deduplicate handle_utxocache_* logic (stickies-v)
ad90ba36bd test: refactor: rename inbound to is_inbound (stickies-v)
afc0224cdb test: refactor: remove unnecessary blocks_checked counter (stickies-v)
Pull request description:
Various cleanups to the USDT functional tests, largely (but not exclusively) follow-ups to https://github.com/bitcoin/bitcoin/pull/27831#pullrequestreview-1491438045. Except for slightly different logging behaviour in "test: store utxocache events" and "test: log sanity check assertion failures", this is a refactor PR, removing unnecessary code and (imo) making it more readable and maintainable.
The rationale for each change is in the corresponding commit message.
Note: except for "test: store utxocache events" (which relies on its parent, and I separated into two commits because we may want the parent but not the child), all commits are stand-alone and I'm okay with dropping one/multiple commits if they turn out to be controversial or undesired.
ACKs for top commit:
0xB10C:
ACK 9f55773a37. Reviewed the code and ran the USDT interface tests. I stepped through the commits and think all changes are reasonable.
Tree-SHA512: 6c37a0265b6c26d4f9552a056a690b8f86f7304bd33b4419febd8b17369cf6af799cb87c16df35d0c2a1b839ad31de24661d4384eafa88816c2051c522fd3bf5
Even though we expect these functions to only produce one event,
we still keep a counter to check if that's true. By simply storing
all the events, we can remove the counters and make debugging
easier, by allowing pdb to access the events.
61f4b9b7ad Manage exceptions in bcc callback functions (virtu)
Pull request description:
Address #27380 (and similar future issues) by handling failed `assert_equal()` assertions in bcc callback functions
### Problem
Exceptions are not propagated in ctype callback functions used by bcc. This means an AssertionError exception raised by `assert_equal()` to signal a failed assertion is not getting caught and properly logged. Instead, the error is logged to stdout and execution of the callback stops.
The current workaround to check whether all `assert_equal()` assertions in a callback succeeded is to increment a success counter after the assertions (which only gets incremented if none exception is raised and stops execution). Then, outside the callback, the success counter can be used to check whether a callback executed successfully.
One issue with the described workaround is that when an exception occurs, there is no way of telling which of the `assert_equal()` statements caused the exception; moreover, there is no way of inspecting how the pieces of data that got compared in `assert_equal()` differed (often a crucial clue when debugging what went wrong).
This problem is happening in #27380: Sporadically, in the `mempool:rejected` test, execution does not reach the end of the callback function and the success counter is not incremented. Thus, the test fails when comparing the counter to its expected value of one. Without knowing which of the asserts failed any why it failed, this issue is hard to debug.
### Solution
Two fixes come to mind. The first involves having the callback function make event data accessible outside the callback and inspecting the event using `assert_equal()` outside the callback. This solution still requires a counter in the callback in order to tell whether a callback was actually executed or if instead the call to perf_buffer_poll() timed out.
The second fix entails wrapping all relevant `assert_equal()` statements inside callback functions into try-catch blocks and manually logging AssertionErrors. While not as elegant in terms of design, this approach can be more pragmatic for more complex tests (e.g., ones involving multiple events, events of different types, or the order of events).
The solution proposed here is to select the most pragmatic fix on a case-by-case basis: Tests in `interface_usdt_net.py`, `interface_usdt_mempool.py` and `interface_usdt_validation.py` have been refactored to use the first approach, while the second approach was chosen for `interface_usdt_utxocache.py` (partly to provide a reference for the second approach, but mainly because the utxocache tests are the most intricate tests, and refactoring them to use the first approach would negatively impact their readability). Lastly, `interface_usdt_coinselection.py` was kept unchanged because it does not use `assert_equal()` statements inside callback functions.
ACKs for top commit:
0xB10C:
Reviewed the changes since my last review. ACK 61f4b9b7ad. I've tested that the combined log contains both exceptions by modifying `interface_usdt_utxocache.py`.
willcl-ark:
utACK 61f4b9b
stickies-v:
utACK 61f4b9b7a
Tree-SHA512: 85cdaabf370d4f09a9eab6af9ce7c796cd9d08cb91f38f021f71adda34c5f643331022dd09cadb95be2185dad6016c95cbb8942e41e4fbd566a49bf431c5141a
Exceptions are not propagated in ctype callback functions used by bcc.
This means an AssertionError exception raised by check_equal() to signal
a failed assertion is not getting caught and properly logged. Instead,
the error is logged to stdout and execution of the handler stops.
The current workaround to check whether all check_equal() assertions in
a callback succeeded is to increment a success counter after the
assertions (which only gets incremented if none exception is raised and
stops execution). Then, outside the callback, the success counter can be
used to check whether a callback executed successfully.
One issue with the described workaround is that when an exception
occurs, there is no way of telling which of the check_equal() statements
caused the exception; moreover, there is no way of inspecting how the
pieces of data that got compared in check_equal() differed (often
a crucial clue when debugging what went wrong).
Two fixes to this problem come to mind. The first involves having the
callback function make event data accessible outside the callback and
inspecting the event using check_equal() outside the callback. This
solution still requires a counter in the callback to tell whether
a callback was actually executed or if instead the call to
perf_buffer_poll() timed out.
The second fix entails wrapping all relevant check_equal() statements
inside callback functions into try-catch blocks and manually logging
AssertionErrors. While not as elegant in terms of design, this approach
can be more pragmatic for more complex tests (e.g., ones involving
multiple events, events of different types, or the order of events).
The solution proposed here is to select the most pragmatic fix on
a case-by-case basis: Tests in interface_usdt_net.py,
interface_usdt_mempool.py and interface_usdt_validation.py have been
refactored to use the first approach, while the second approach was
chosen for interface_usdt_utxocache.py (partly to provide a reference
for the second approach, but mainly because the utxocache tests are the
most intricate tests, and refactoring them to use the first approach
would negatively impact their readability). Lastly,
interface_usdt_coinselection.py was kept unchanged because it does not
use check_equal() statements inside callback functions.
Tracepoints for added, removed, replaced, and rejected transactions.
The removal reason is passed as string instead of a numeric value, since
the benefits of not having to maintain a redundant enum-string mapping
seem to outweigh the small cost of string generation. The reject reason
is passed as string as well, although here the string does not have to
be generated but is readily available.
So far, tracepoint PRs typically included two demo scripts: a naive
bpftrace script to show raw tracepoint data and a bcc script for a more
refined view. However, as some of the ongoing changes to bpftrace
introduce a certain degree of unreliability (running some of the
existing bpftrace scripts was not possible with standard kernels and
bpftrace packages on latest stable Ubuntu, Debian, and NixOS), this PR
includes only a single bcc script that fuses the functionality of former
bpftrace and bcc scripts.
