3f686d1a28 ci: Run fuzz testing test cases (bitcoin-core/qa-assets) under valgrind to catch memory errors (practicalswift)
Pull request description:
Re-introduce the Travis valgrind fuzzing job which was removed by PR #18899. The removal seems to have been made by accident since the removed job does not appear to be the source of the problem the PR set out to fix.
---
Run fuzz testing [test cases (bitcoin-core/qa-assets)](https://github.com/bitcoin-core/qa-assets) under `valgrind`.
This would have caught `util: Avoid potential uninitialized read in FormatISO8601DateTime(int64_t) by checking gmtime_s/gmtime_r return value` (#18162) and similar cases.
This fuzzing job was introduced in #18166.
Top commit has no ACKs.
Tree-SHA512: 6e2681eb0ade6af465c5ea91ac163a337465d2130ec9880ba57a36d9af7c25682734586a32977dc25972d4f78483f339d680ea48c0ae13cf1dfa52b617aae401
37ae687f95 Add tests for CPubKey serialization/unserialization (Elichai Turkel)
9b8907fade Check size after Unserializing CPubKey (Elichai Turkel)
Pull request description:
Found by practicalswift, closes#19235
Currently all the public API(except the pointer-like API) in CPubKey that sets/constructs a pubkey goes through `CPubKey::Set` which checks if that the length and size match and if not invalidates the key.
This adds the same check to `CPubKey::Unserialize`, sadly I don't see an easy way to just push this to the existing checks in `CPubKey::Set` but it's only a simple condition.
The problem with not invalidating is that if you write a pubkey like: `{0x02,0x00}` it will think the actual length is 33(because of `size()`) and will access uninitialized memory if you call any of the functions on CPubKey.
ACKs for top commit:
practicalswift:
re-ACK 37ae687f95
jonatack:
Code review re-ACK 37ae687 per `git diff eab8ee3 37ae687` only change since last review at eab8ee3 is passing the `pubkey` param by reference to const instead of by value in `src/test/key_tests.cpp::CmpSerializationPubkey`
MarcoFalke:
ACK 37ae687f95
Tree-SHA512: 30173755555dfc76d6263fb6a59f41be36049ffae7b4e1b92b922d668f5e5e2331f7374d5fa10d5d59fc53020d2966156905ffcfa8b8129c1f6d0ca062174ff1
If the socket is tearing down macOS will return EPROTOTYPE instead of EPIPE.
Because python doesn't handle this internally we have to do a workaround and retry the request.
See https://bugs.python.org/issue33450
libevent uses getaddrinfo when available, and falls back to gethostbyname
Windows has both, but gethostbyname only supports IPv4
libevent fails to detect Windows's getaddrinfo due to not including the right headers
This patches libevent's configure script to check it correctly
56010f9256 test: hoist p2p values to test framework constants (Jon Atack)
75447f0893 test: improve msg sends and p2p disconnections in p2p_invalid_messages (Jon Atack)
57960192a5 test: refactor test_large_inv() into 3 tests with common method (Jon Atack)
e2b21d8a59 test: add p2p_invalid_messages logging (Jon Atack)
9fa494dc09 net: update misbehavior logging for oversized messages (Jon Atack)
Pull request description:
...seen while reviewing #19264, #19252, #19304 and #19107:
in `net_processing.cpp`
- make the debug logging for oversized message size misbehavior the same for `addr`, `getdata`, `headers` and `inv` messages
in `p2p_invalid_messages`
- add missing logging
- improve assertions/message sends, move cleanup disconnections outside the assertion scopes
- split a slowish 3-part test into 3 order-independent tests
- add a few p2p constants to the test framework
ACKs for top commit:
troygiorshev:
reACK 56010f9256
MarcoFalke:
ACK 56010f9256 🎛
Tree-SHA512: db67b70278f8d4c318907e105af54b54eb3afd15500f9aa0c98034f6fd4bd1cf9ad1663037bd9b237ff4890f3059b37291a6498d8d6ae2cc38efb9f045f73310
bd93e32292 refactor: Replace HexStr(o.begin(), o.end()) with HexStr(o) (Wladimir J. van der Laan)
Pull request description:
HexStr can be called with anything that bas `begin()` and `end()` functions, so clean up the redundant calls.
(context: I tried to convert `HexStr` to use span, but this turns out to be somewhat more involved than I thought, because of the limitation to pre-c++17 Span lacking iterator-based constructor) . This commit is a first step which stands on its own though)
ACKs for top commit:
jonatack:
ACK bd93e32292
troygiorshev:
ACK bd93e32292
MarcoFalke:
review ACK bd93e32292🔌
Tree-SHA512: 7e4c9d0259b8d23271d233095f1c51db1ee021e865361d74c05c10dd5129aa6d34a243323e2b4596d648e2d7b25c7ebdee37a3e4f99a27883cb4c3cd26432b08
4f9d9efb4e qt: Remove needless headers (Hennadii Stepanov)
Pull request description:
No symbols from the removed headers are used in the `qt/walletview.cpp`.
This is a small followup of https://github.com/bitcoin/bitcoin/pull/18027.
Top commit has no ACKs.
Tree-SHA512: 986ed5c8f3bac4c0053736ce84d738f8593d3dbf713109af3cb9b7051cd838f23152a39bb3c1e9694a993c4e7accf14e94e5beff5e7881155638cd44fbf7f46f
On OS X, when searching Homebrew keg-only packages for BDB 4.8, if we find it,
use BDB_CPPFLAGS and BDB_LIBS instead of CFLAGS and LIBS for the result. This
is (1) more correct, and (2) necessary in order to give this location
priority over other directories in the include search path, which may include
system include directories with other versions of BDB.
The TRY_LOCK(cs_inventory) in DisconnectNodes() is taken after the CNode
object has been removed from vNodes and when the CNode's nRefCount is
zero.
The only other places that cs_inventory can be taken are:
- In ProcessMessages() or SendMessages(), when the CNode's nRefCount
must be >0 (see ThreadMessageHandler(), where the refcount is
incremented before calling ProcessMessages() and SendMessages()).
- In a ForEachNode() lambda in PeerLogicValidation::UpdatedBlockTip().
ForEachNode() locks cs_vNodes and calls the function on the CNode
objects in vNodes.
Therefore, cs_inventory is never locked by another thread when the
TRY_LOCK(cs_inventory) is reached in DisconnectNodes(). Since the
only purpose of this TRY_LOCK is to ensure that the lock is not
taken by another thread, this always succeeds. Remove the check.
PushBlockInventory() and PushBlockHash() are functions that can
be replaced with single-line statements. This also eliminates
the single place that cs_inventory is taken recursively.
adf543d714 darwin: pass mlinker-version so that clang enables new features (Cory Fields)
2418f739f7 macos: Bump to xcode 11.3.1 and 10.15 SDK (Cory Fields)
5c2c835433 depends: bump MacOS toolchain (Cory Fields)
85b5e42088 contrib: macdeploy: Remove historical extraction notes (Carl Dong)
351beb5c9a contrib: macdeploy: Use apple-sdk-tools instead of xar+pbzx (Carl Dong)
fbcfcf6954 native_cctools: Don't use libc++ from pinned clang (Carl Dong)
3381e4a189 Adapt rest of tooling to new SDK naming scheme (Carl Dong)
b3394ab235 contrib: macdeploy: Correctly generate macOS SDK (Carl Dong)
Pull request description:
This PR achieves 3 main things:
1. It simplifies the macOS SDK generation by putting the logic inside a (semi-)portable python3 script `gen-sdk`
2. It transitions us to using `libc++` headers extracted from the `Xcode.app`, which is more correct as those headers better match the `.tbd` library stubs we use from the `MacOSX.sdk` (located under the same `Xcode.app`). Previously, we used `libc++` headers copied from our downloaded, pinned clang (see `native_cctools.mk`).
3. It bumps the macOS toolchain in a way that fulfills all of the following constraints:
1. The new SDK should support compiling with C++17 (our current one doesn't)
2. The new toolchain should not change our minimum supported macOS version (`-mmacosx-version-min`)
3. The new toolchain should expect to use a version of `cctools` that is supported by https://github.com/tpoechtrager/cctools-port
For the constraints in (3), you can reference [this chart](https://en.wikipedia.org/wiki/Xcode#Xcode_7.0_-_11.x_(since_Free_On-Device_Development)) to see that the newest toolchain we can use with our `cctools-port` is `11.3.1`, and the rest of the constraints were tested with local builds.
#### But [the other Wikipedia chart](https://en.wikipedia.org/wiki/Xcode#Xcode_11.x_(since_SwiftUI_framework)) says that the "min macOS to run" for Xcode 11.3.1 is 10.14.4, doesn't that violate constraint (ii)?
This confused me at first too, but the "min macOS to run" is for the Xcode.app App itself. The SDK still supports 10.12, as evident in a few plist files and as proven through local builds.
#### Why bundle all of this together in a single PR?
We need (1) and (2) together, because if we don't, manually adding the `libc++` headers and writing that out in a `README.md` is going to result in a lot of user error, so it's great to have these together to be more correct and also make it easier on the user at the same time.
We need (3) together with everything else because bumping (or in the case of (1), renaming) the SDK requires some human coordination and may break some builds. And since it's not that complicated a change, it makes sense to do it together with the rest.
ACKs for top commit:
theuni:
ACK adf543d714.
fanquake:
ACK adf543d714 - I'll take a look at the linker issue.
Tree-SHA512: 3813b69ebfe9610bee14100f26296fb5438d9bf0dd184ea55e6c38f5ebd94f7c171d98b150fc9e52fde626533f347f7ec51a2b72b79859d946284f578c1084a3
Instead of returning a Dbc (BDB cursor object) and having the caller
deal with the cursor, make BerkeleyBatch handle the cursor internally.
This prepares BerkeleyBatch to work with other database systems as Dbc
objects are BDB specific.
Updates Python linters, spellchecking, and ShellCheck versions. The PR links are updated for
the dependency versions in test/README.md. ShellCheck SC2230 removed to align with with new
behaviour in v0.7.1.
Fixes#19346.
f1d21ef1c3 doc: add C++17 release note for 0.21.0 (fanquake)
Pull request description:
TLDR: Mention that the codebase is now compatible with C++17, and that the
intention is to require C++17 starting with 0.22.0.
Following some discussion with Cory/Carl, and in #16684, I think this is the next step in the C++17 migration.
While #16684 mentions a gitian/Guix release with C++17, it's not yet clear how that would be done. Are we just going to pass `--enable-c++17` in gitian/Guix?. Are we changing our default in configure.ac?
According to the [last comment](https://github.com/bitcoin/bitcoin/issues/16684#issuecomment-643778757) in #16684, we wouldn't be changing anything in depends:
> No, everything (including depends) will stay at C++11.
However I don't think we want to be mixing C++11 built dependencies, with a C++17 built bitcoind, if there is any potential for compatibility issues.
Instead, I'd suggest we build the 0.21.0 release as C++11, and do a complete switch to C++17 for 0.22.0. Also, if we actually wanted to use C++17 in depends for 0.21.0, we couldn't without breaking C++11 compat (Qt). See below.
Here is a potential timeline/TODOs for the migration:
Potential Timeline
* 17 / 6 / 2020 - Today
* Some time prior to split-off:
* Confirm that compiling with C++17 works.
* Confirm that C++11 compatibility has not been broken.
* 1 / 11 / 2020
* [0.21.0 split off happens](https://github.com/bitcoin/bitcoin/issues/18947).
* 2 / 11 / 2020
* Merge an "incompatible with C++11" change into master.
* Switch configure to use C++17 mode by default.
* Update minimum compiler requirements. At least:
* Clang 5: https://clang.llvm.org/cxx_status.html#cxx17
* GCC 7: https://gcc.gnu.org/projects/cxx-status.html#cxx17
* While GCC has some support from 5, it seems a more complete support landed in GCC 7.
* https://gcc.gnu.org/gcc-7/changes.html#cxx
* Switch depends packages to use C++17 where applicable.
* Bump Qt from 5.9.x (no c++17 mode) to, likely, 5.15.x (LTS).
* Drop support for macOS < 10.14.x
* The c++ dylib shipped with macOS [doesn't support c++17, prior to macOS 10.14](https://github.com/bitcoin/bitcoin/issues/16684#issuecomment-643722538).
* Building Qt 5.12 or 5.15 in C++17 mode will also require a minimum macOS deployment target of 10.14. https://codereview.qt-project.org/c/qt/qtbase/+/283832.
* Begin merging PRs like #19183 and #19245.
* I've left some comments in #19183 if the macOS runtime issue interests anyone.
* 3 / 12 / 2020
* 0.21.0 released.
* Built as C++11.
* Contains warning in release notes that compiling 0.22.0 will require C++17.
* 3 / 6 / 2021
* 0.22.0 released.
* Full of C++17 code.
One thing worth noting, is that we cannot bump our Qt to a newer LTS for 0.21.0, without breaking C++11 compatibility. Qt 5.12 is not compilable in C++11 mode, as the project has started using C++14 features throughout at least the macOS portions of it's codebase, and seemingly "forgotten" that the release is meant to be C++11 compatible.
Upstream bug here: https://bugreports.qt.io/browse/QTBUG-77310.
> Building Qt requires C+11, at a minimum, but in practice we use later features, usually under a feature define, or with a fallback of some kind. On platforms that support > C11, we've (apparently) not considered the fallback necessary, under the assumption C+14 is always available.
ACKs for top commit:
MarcoFalke:
ACK f1d21ef1c3 can't hurt to give an advance warning
Sjors:
ACK f1d21ef1c3
laanwj:
ACK f1d21ef1c3
theStack:
ACK f1d21ef1c3
Tree-SHA512: 706baceb07d9584783ba6e437cdf447531c20f586285b9797edc21f3adb1e9d386059d1c543c70eb298d0f8e555dafb6682a55d35c5836979fc12132e8ba06f5
20b6e95944 test: refactor functional tests to use restart_node (Christopher Coverdale)
Pull request description:
fixes#19345
This PR replaces consecutive calls to `stop_node()` and `start_node()` with `restart_node()` where appropriate in the functional tests.
The commit messages are repetitive but focused on each file changed with the intention of squashing if applicable.
ACKs for top commit:
laanwj:
ACK 20b6e95944
Tree-SHA512: 1cfa1fb8c5f01a7b00fe44e80dbef072147f21e3891098817acd4275b0c5d91dc1c787594209e117edd418f2fa3a7b2dfcbafdf87efc07f740040938d641f3a9
ccef5d7bf0 test: add two edge case tests for CSubNet (Vasil Dimov)
Pull request description:
This is chopped off from https://github.com/bitcoin/bitcoin/pull/19031. It is needed because later 19031 modifies the related code and the tests ensure that no surprising changes in behavior sneak in.
ACKs for top commit:
practicalswift:
ACK ccef5d7bf0 -- more test coverage is better than less test coverage :)
laanwj:
ACK ccef5d7bf0
hebasto:
ACK ccef5d7bf0, I have reviewed the code and it looks OK, I agree it can be merged.
Tree-SHA512: 6d386672b6598aeddd33dabe3512e816cf548d5c1af56c4c9e6f897d513b62ba4659cde73405811a0df286ffee3a3f084ab7caf8e3a2086fa9ddecd1bdcb3c67