Skip to content

Instantly share code, notes, and snippets.

Embed
What would you like to do?
Debugging Bitcoin Core

Moved to https://github.com/fjahr/debugging_bitcoin to allow for better collaboration.

This document is currently optimized for MacOS. If you would like to help me add Linux equivalent commands, please let me know.

Debugging Bitcoin Core

This guide is designed to give beginners of C++ development and/or people new to the bitcoin core code base an overview of the tools available for debugging issues as well as giving hints where issues may trip you up.

Table of contents

General tips

First of all, debugging involves a lot of compiling, so you definitely want to speed it up as much as possible. I recommend looking at the general productivity notes in the bitcoin core docs and install ccache and optimize your configuration.

Also do not forget the disable optimizations using the -O0 flag, otherwise debugging will be impossible as symbol names will not be recognizable.

An example of configure flags: ./configure CXXFLAGS="-O0 -g" CFLAGS="-O0 -g"

Also note this guide is using lldb instead of gdb because I am running MacOS. For Linux users gdb seems to be the standard and even if you are using gdb the guide should still work for you, as the tools are very similar.

Are you in the right spot?

  • Mainnet/Testnet/Regtest all have their own debug.log files
  • Feature tests log to temp files which get cleaned up unless your test fails or you specify --no-cleanup

Debugging

Running your own bitcoind

These are examples where you are interacting with the code yourself and don't rely on tests to reproduce the error.

Logging from own bitcoind

In general you can use to your std::out but this will not appear in any logs. It is rather recommended to use LogPrintf. Insert into your code a line similar to the following example.

LogPrintf("@@@");

You can then grep for the result in your debug.log file:

$ cat ~/Library/Application\ Support/Bitcoin/regtest/debug.log | grep @@@

This example shows the path of the regtest environment debug.log file. Remember to change this if you are logging from testnet or another environment.

If you would like to log from inside files that validate consensus rules (see src/Makefile.am) then you will errors from missing header files and when you have added those the linker will complain. You can make this work, of course, but I would recommend you use a debugger in that context instead.

Debugging from your own bitcoind

You can start your bitcoind using a debugging tool like lldb in order to debug the code:

$ lldb src/bitcoind

Within the lldb console you can set breakpoints before actually running starting the bitcoind process using the run command. You can then interact with the bitcoind process like you normally would using your bitcoin-cli.

Running unit tests

You are running unit tests which are located at src/test/ and use the BOOST library test framework. These are executed by a seperate executable that is also compiled with make. It is located at src/test/test_bitcoin.

So helpful tips about execution (this uses the Boost library).

Run just one test file: src/test/test_bitcoin --log_level=all --run_test=getarg_tests

Run just one test: src/test/test_bitcoin --log_level=all --run_test=*/the_one_test

Logging from unit tests

Logging from the tests you need to use the BOOST framework provided functions to achieve seeing the logging output. Several methods are available, but simplest is probably adding BOOST_TEST_MESSAGE in your code:

BOOST_TEST_MESSAGE("@@@");

Logging from unit tests code

To have log statements from your code appear in the unit test output you will have to print to stderr directly:

fprintf(stderr, "from the code");

Debugging Unit Tests

You can start the test_bitcoin binary using lldb just like bitcoind. This allows you to set breakpoints anywhere in your unit test or the code and then execute the tests any way you want using the run keyword followed by the arguments you would normally pass to test_bitcoin when calling it directly.

$ lldb src/test/test_bitcoin
(lldb) target create "src/test/test_bitcoin"
Traceback (most recent call last):
  File "<input>", line 1, in <module>
  File "/usr/local/Cellar/python@2/2.7.16/Frameworks/Python.framework/Versions/2.7/lib/python2.7/copy.py", line 52, in <module>
    import weakref
  File "/usr/local/Cellar/python@2/2.7.16/Frameworks/Python.framework/Versions/2.7/lib/python2.7/weakref.py", line 14, in <module>
    from _weakref import (
ImportError: cannot import name _remove_dead_weakref
Current executable set to 'src/test/test_bitcoin' (x86_64).
(lldb) run --log_level=all --run_test=*/lthash_tests

Running functional tests

You are running tests located in test/functional/ which are written in python.

Logging from functional tests

for log level debug run functional tests with --loglevel=debug.

self.log.info("foo")
self.log.debug("bar")

Use --tracerpc to see the log outputs from the RPCs of the different nodes running in the functional test in std::out.

If it doesn't fail, make it fail and use this:

TestFramework (ERROR): Hint: Call /Users/FJ/projects/clones/bitcoin/test/functional/combine_logs.py '/var/folders/9z/n7rz_6cj3bq__11k5kcrsvvm0000gn/T/bitcoin_func_test_epkcr926' to consolidate all logs

You can even assert on logging messages using with self.nodes[0].assert_debug_log(["lkagllksa"]):. However, don't change indentation of test lines. Just insert somewhere indented between def run_test and the body of it.

Logging from functional tests code

Using LogPrintf, as seen before, you will be able to see the output in the combined logs.

Debugging from functional tests

1. Compile Bitcoin for debugging

In case you have not done it yet, compile bitcoin for debugging (change other config flags as you need them).

$ make clean
$ ./configure CXXFLAGS="-O0 -ggdb3"
$ make -j "$(($(sysctl -n hw.physicalcpu)+1))"
2. Ensure you have lldb installed

Your should see something like this:

$ lldb -v
lldb-1001.0.13.3
3. Halt functional test

You could halt the test with a sleep as well but much cleaner (although a little confusing maybe) is to use another debugger within python: pdb.

Add the following line before the functional test causing the error that you want to debug:

import pdb; pdb.set_trace()

Then run your test. You will need to call the test directly and not run it through the test_runner.py because then you could not get access to the pdb console.

$ ./test/functional/example_test.py
4. Attach lldb to running process

Start lldb with the running bitcoind process (might not work if you have other bitcoind processes running). Just running lldb instead of PATH=/usr/bin /usr/bin/lldb might work for you, too, but lots of people seem to run into problems when lldb tries to use their system python in that case.

$ PATH=/usr/bin /usr/bin/lldb -p $(pgrep bitcoind)
5. Set your breakpoints in lldb

Set you breakpoint with b, then you have to enter continue since lldb is setting a stop to the process as well.

(lldb) b createwallet
Breakpoint 1: [...]
(lldb) continue
Process XXXXX resuming
6. Let test continue

You can now let you test continue so that the process is actually running into your breakpoint.

(Pdb) continue

You should now see something like this in you lldb and can start debugging:

Process XXXXX stopped
* thread #10, name = 'bitcoin-httpworker.3', stop reason = breakpoint 1.1
    frame #0: 0x00000001038c8e43 bitcoind`createwallet(request=0x0000700004d55a10) at rpcwallet.cpp:2642:9
   2639	static UniValue createwallet(const JSONRPCRequest& request)
   2640	{
   2641	    RPCHelpMan{
-> 2642	        "createwallet",
   2643	        "\nCreates and loads a new wallet.\n",
   2644	        {
   2645	            {"wallet_name", RPCArg::Type::STR, RPCArg::Optional::NO, "The name for the new wallet. If this is a path, the wallet will be created at the path location."},
Target 0: (bitcoind) stopped.
(lldb) 

More Tools for Segfaults

Core dumps

A core dump is a full dump of the working memory of your computer. When activated it will be created in /cores on your computer in case a segfault happens.

To activate core dumps on macOS you have to run

$ ulimit -c unlimited

then you have to run the process where you are observing the segfault in the same terminal. You can then inspect them in your /cores directory.

You should always make sure to not generate more core dumps than you need to and clean up your /cores directory when you have solved your issue. Core dumps are huge files and can clutter up your disc space very quickly.

valgrind

Install valgrind

On newer MacOS versions this is harder than you might think because you can only install it through homebrew up to version 10.13. If you currently have an up-to-date MacOS Mojave installed you will have to use the following script to make it work.

$ brew install --HEAD https://raw.githubusercontent.com/sowson/valgrind/master/valgrind.rb

Run bitcoind with valgrind

Using valgrind works similar to lldb. You start the executable where the segfault might happen using valgrind, when an error is observed you will be able to inspect it.

For example:

$ sudo valgrind src/bitcoind -regtest

Now you can interact with your bitcoind normally using bitcoin-cli, for example, and trigger the error by hand using the right parameters.

Further resources

Getting help

READMEs

Other resources

https://medium.com/provoost-on-crypto/debugging-bitcoin-core-functional-tests-cc0aa6e7fd3e https://bitcoin.stackexchange.com/questions/76521/debugging-bitcoin-unit-tests

lldb

Using lldb as a standalone debugger

LLVM Tutorial on lldb

BOOST

@jonatack

This comment has been minimized.

Copy link

jonatack commented Sep 10, 2019

To have log statements from your code appear in the unit test output you will have to print to stderr directly:

fprintf(stderr, "from the code");

Could also mention stdout?

@jonatack

This comment has been minimized.

Copy link

jonatack commented Sep 10, 2019

Could possibly mention Berkeley DB commands (located in db4/bin) for inspecting wallet file changes, notably db_dump and db_load:

      db4/bin/db_dump ~/.bitcoin/wallets/wallet.dat
      db4/bin/db_dump -p ~/.bitcoin/wallets/wallet.dat | grep -C 2 0X
      db4/bin/db_load
@michaelfolkson

This comment has been minimized.

Copy link

michaelfolkson commented Oct 25, 2019

As I've said in person, great work. Personally I would find a maintained version of this doc very useful. Some minor suggestions and typos. I would open a PR but it doesn't appear you can on Gists.

This guide is designed to give beginners of C++ development and/or people new to the Bitcoin Core codebase an overview of the tools available for debugging issues as well as providing hints for regularly encountered challenges

I recommend looking at the general productivity notes in the Bitcoin Core docs

Also do not forget to disable optimizations

Mainnet/Testnet/Regtest all have their own debug.log files

Perhaps say where they are located here? You do say where the regtest debug.log file is later in the document but it should probably made clear earlier.

In general you can use to your std::out but this will not appear in any logs. It is instead recommended to use LogPrintf.

LogPrintf("@@@");

Perhaps add that the @ symbol could be replaced by anything and is only included for visual and search purposes.

If you would like to log from inside files that validate consensus rules (see src/Makefile.am) then you will see errors

Within the lldb console you can set breakpoints before actually starting the bitcoind process using the run command.

These are executed by a separate executable"

Some helpful tips about execution (this uses the Boost library).

Logging from the tests requires you to use the BOOST framework provided functions to be able to see* the logging output. Several methods are available, but the simplest is probably adding BOOST_TEST_MESSAGE in your code

To have log statements from your code appear in the unit test output you have to print to stderr directly

For log level debug run functional tests with --loglevel=debug.

If it doesn't fail, make it fail and use this:

I don't get what you mean here. Why do you want to make it fail? So that you can practise for when it does fail?

You can even assert on logging messages with self.nodes[0].assert_debug_log(["lkagllksa"]):

However, don't change the indentation of test lines.

You could halt the test with a sleep but it is cleaner to use another debugger within Python

Then run your test. You will need to call the test directly and not run it through the test_runner.py because if you do that you won't get access to the pdb console.

You can now let your test continue so that the process is actually running into your breakpoint.

You should now see something like this in your lldb and can start debugging:

More Tools for Segfaults

Perhaps explain what a segfault is and when will you most likely experience them working on Bitcoin Core

valgrind

Similarly with Valgrind. What is it and when would you use it?

@svanstaa

This comment has been minimized.

Copy link

svanstaa commented Nov 6, 2019

Great work, Fabian! I have found something analog to this using gdb, which i found helpful for adapting your approach to Linux.
https://gist.github.com/gubatron/36784ee38e45cb4bf4c7a82ecc87b6a8

@LarryRuane

This comment has been minimized.

Copy link

LarryRuane commented Apr 14, 2020

Fabian, these are awesome tips, thanks! The idea of using pdb to hold a functional test so that you can then attach the debugger to bitcoind is great. In the past, I've solved this problem without using pdb (and I mention this here in case this may be useful in some situations, such as if bitcoind fails during startup) by adding the following line of code to bitcoind where I want to set a breakpoint:

    {static int spin=1;while(spin);}

(You can insert that line in more than one place.) Then I run the functional test as usual, and, using top (I'm on Linux), when I see bitcoind's CPU is a steady 100%, I attach to it with the debugger (top shows its PID, and I run gdb /proc/12345/exe 12345 if that's the PID). If I run bt (backtrace) it's usually on the wrong thread, so I run info threads and it's clear which is the spinning thread. I switch to it using thr 17 (whichever thread id) then bt to confirm I'm at the spinning code. Then I type set var spin=0 so that it doesn't stop when it reaches here in the future (that's why the static qualifier is needed), and I also set a breakpoint there (or anywhere) and continue.

@LarryRuane

This comment has been minimized.

Copy link

LarryRuane commented Apr 14, 2020

It's sometimes useful to see the output of the C++ preprocessor -- so much magic happens there it's sometimes almost impossible to figure out what code is actually being compiled. In those cases, I do this (using validation.cpp as an example):

$ cd src
$ touch validation.cpp
$ make V=1 libbitcoin_server_a-validation.o

This prints the compile command line with all the arguments, for example (a bunch of stuff deleted at [...]):

g++ -std=c++11 -DHAVE_CONFIG_H -I. -I../src [...] -c -o libbitcoin_server_a-validation.o `test -f 'validation.cpp' || echo './'`validation.cpp

Then I use the mouse to copy and paste most of this, except replacing the -c with -E and libbitcoin_server_a-validation.o with cpp.out (or any filename you like). This file will contain the output of the preprocessor, which will be huge, but you can search for the code of interest.

A similar idea is to add -H (headers) anywhere in the command line, then add to the end: >headers.out 2>&1. This file will show the hierarchy of header file includes, sometimes very useful. It looks like this, for example (the number of leading dots indicates the include nesting level):

. ./validation.h
.. ./config/bitcoin-config.h
.. ./amount.h
... /usr/lib/gcc/x86_64-linux-gnu/7/include/stdint.h
.... /usr/include/stdint.h
[... lots more ...]
@nickycutesc

This comment has been minimized.

Copy link

nickycutesc commented Apr 27, 2020

Hey Fabian, great work! I came upon this guide in order to become familiar with debugging a unit test issue I filed yesterday: bitcoin/bitcoin#18776

I noticed this issue while building bitcoin, and I'm trying to use your guide to help me understand why this unit test is failing. I'm having trouble setting a breakpoint:

  1. $ lldb src/test/test_bitcoin
  2. $ b /src/wallet/test/psbt_wallet_tests.cpp:19

image

The file in question is here: ./src/wallet/test/psbt_wallet_tests.cpp. It's not very apparent why I'm not able to set a breakpoint in the file. Thanks for your time!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.