78c312c983 Replace current benchmarking framework with nanobench (Martin Ankerl)
Pull request description:
Replace current benchmarking framework with nanobench
This replaces the current benchmarking framework with nanobench [1], an
MIT licensed single-header benchmarking library, of which I am the
autor. This has in my opinion several advantages, especially on Linux:
* fast: Running all benchmarks takes ~6 seconds instead of 4m13s on
an Intel i7-8700 CPU @ 3.20GHz.
* accurate: I ran e.g. the benchmark for SipHash_32b 10 times and
calculate standard deviation / mean = coefficient of variation:
* 0.57% CV for old benchmarking framework
* 0.20% CV for nanobench
So the benchmark results with nanobench seem to vary less than with
the old framework.
* It automatically determines runtime based on clock precision, no need
to specify number of evaluations.
* measure instructions, cycles, branches, instructions per cycle,
branch misses (only Linux, when performance counters are available)
* output in markdown table format.
* Warn about unstable environment (frequency scaling, turbo, ...)
* For better profiling, it is possible to set the environment variable
NANOBENCH_ENDLESS to force endless running of a particular benchmark
without the need to recompile. This makes it to e.g. run "perf top"
and look at hotspots.
Here is an example copy & pasted from the terminal output:
| ns/byte | byte/s | err% | ins/byte | cyc/byte | IPC | bra/byte | miss% | total | benchmark
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 2.52 | 396,529,415.94 | 0.6% | 25.42 | 8.02 | 3.169 | 0.06 | 0.0% | 0.03 | `bench/crypto_hash.cpp RIPEMD160`
| 1.87 | 535,161,444.83 | 0.3% | 21.36 | 5.95 | 3.589 | 0.06 | 0.0% | 0.02 | `bench/crypto_hash.cpp SHA1`
| 3.22 | 310,344,174.79 | 1.1% | 36.80 | 10.22 | 3.601 | 0.09 | 0.0% | 0.04 | `bench/crypto_hash.cpp SHA256`
| 2.01 | 496,375,796.23 | 0.0% | 18.72 | 6.43 | 2.911 | 0.01 | 1.0% | 0.00 | `bench/crypto_hash.cpp SHA256D64_1024`
| 7.23 | 138,263,519.35 | 0.1% | 82.66 | 23.11 | 3.577 | 1.63 | 0.1% | 0.00 | `bench/crypto_hash.cpp SHA256_32b`
| 3.04 | 328,780,166.40 | 0.3% | 35.82 | 9.69 | 3.696 | 0.03 | 0.0% | 0.03 | `bench/crypto_hash.cpp SHA512`
[1] https://github.com/martinus/nanobench
ACKs for top commit:
laanwj:
ACK 78c312c983
Tree-SHA512: 9e18770b18b6f95a7d0105a4a5497d31cf4eb5efe6574f4482f6f1b4c88d7e0946b9a4a1e9e8e6ecbf41a3f2d7571240677dcb45af29a6f0584e89b25f32e49e
Check that sections are appropriately separated in virtual memory,
based on their (expected) permissions. This checks for missing
-Wl,-z,separate-code and potentially other problems.
Co-authored-by: fanquake <fanquake@gmail.com>
The RandomOrphan function and the function ecdsa_signature_parse_der_lax
in pubkey.cpp were causing non-deterministic test coverage.
Force seed in the beginning of the test to make it deterministic.
The seed is selected carefully so that all branches of the function
ecdsa_signature_parse_der_lax are executed. Prior to this fix, the test
was exhibiting non-deterministic coverage since none of the ECDSA
signatures that were generated during the test had leading zeroes in
either R, S, or both, resulting in some branches of said function not
being executed. The seed ensures that both conditions are hit.
Removed denialofservice_tests test entry from the list of non-deterministic
tests in the coverage script.
Previously, we did not include the macOS SDK libc++ headers in our SDK
creation process and instead used whichever libc++ headers shipped with
the clang package we downloaded in depends.
This change adds a script (which works on both GNU/Linux and macOS) to
correctly generate the macOS SDK including the libc++ headers. This can
be thought of as a simplified rewrite of tpoechtrager's script:
d3392f4eae/tools/gen_sdk_package.sh
The location within the SDK where we place the libc++ headers is chosen
such that clang's search path detection logic for sysroots would pick up
the headers properly.
We also document this change.
47b49a05ea contrib: Fix SyntaxWarning in Python base58 implementation (Alex Willmer)
Pull request description:
In Python integers should be compared for equality (`i == j`), not identity (`i is j`). Recent versions of CPython 3.x emit a SyntaxWarning when they encounter this incorrect usage, e.g.
```
$ python3 base58.py
base58.py:110: SyntaxWarning: "is" with a literal. Did you mean "=="?
assert get_bcaddress_version('15VjRaDX9zpbA8LVnbrCAFzrVzN7ixHNsC') is 0
Tests passed
```
ACKs for top commit:
MarcoFalke:
ACK 47b49a05ea
Tree-SHA512: 9f8962025dcdfa062c0515c68a1864f5bbeb86bd0510c0ec0e413a5edb6afbfd5f41b4c0255784e53db8eaf39c68b7cfa7cc8a33a2e5214aae463fda374f8719
In Python integers should be compared for equality (`i == j`), not identity (`i is j`). Recent versions of CPython 3.x emit a SyntaxWarning when they encounter this incorrect usage, e.g.
```
$ python3 base58.py
base58.py:110: SyntaxWarning: "is" with a literal. Did you mean "=="?
assert get_bcaddress_version('15VjRaDX9zpbA8LVnbrCAFzrVzN7ixHNsC') is 0
Tests passed
```
This replaces the current benchmarking framework with nanobench [1], an
MIT licensed single-header benchmarking library, of which I am the
autor. This has in my opinion several advantages, especially on Linux:
* fast: Running all benchmarks takes ~6 seconds instead of 4m13s on
an Intel i7-8700 CPU @ 3.20GHz.
* accurate: I ran e.g. the benchmark for SipHash_32b 10 times and
calculate standard deviation / mean = coefficient of variation:
* 0.57% CV for old benchmarking framework
* 0.20% CV for nanobench
So the benchmark results with nanobench seem to vary less than with
the old framework.
* It automatically determines runtime based on clock precision, no need
to specify number of evaluations.
* measure instructions, cycles, branches, instructions per cycle,
branch misses (only Linux, when performance counters are available)
* output in markdown table format.
* Warn about unstable environment (frequency scaling, turbo, ...)
* For better profiling, it is possible to set the environment variable
NANOBENCH_ENDLESS to force endless running of a particular benchmark
without the need to recompile. This makes it to e.g. run "perf top"
and look at hotspots.
Here is an example copy & pasted from the terminal output:
| ns/byte | byte/s | err% | ins/byte | cyc/byte | IPC | bra/byte | miss% | total | benchmark
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 2.52 | 396,529,415.94 | 0.6% | 25.42 | 8.02 | 3.169 | 0.06 | 0.0% | 0.03 | `bench/crypto_hash.cpp RIPEMD160`
| 1.87 | 535,161,444.83 | 0.3% | 21.36 | 5.95 | 3.589 | 0.06 | 0.0% | 0.02 | `bench/crypto_hash.cpp SHA1`
| 3.22 | 310,344,174.79 | 1.1% | 36.80 | 10.22 | 3.601 | 0.09 | 0.0% | 0.04 | `bench/crypto_hash.cpp SHA256`
| 2.01 | 496,375,796.23 | 0.0% | 18.72 | 6.43 | 2.911 | 0.01 | 1.0% | 0.00 | `bench/crypto_hash.cpp SHA256D64_1024`
| 7.23 | 138,263,519.35 | 0.1% | 82.66 | 23.11 | 3.577 | 1.63 | 0.1% | 0.00 | `bench/crypto_hash.cpp SHA256_32b`
| 3.04 | 328,780,166.40 | 0.3% | 35.82 | 9.69 | 3.696 | 0.03 | 0.0% | 0.03 | `bench/crypto_hash.cpp SHA512`
[1] https://github.com/martinus/nanobench
* Adds support for asymptotes
This adds support to calculate asymptotic complexity of a benchmark.
This is similar to #17375, but currently only one asymptote is
supported, and I have added support in the benchmark `ComplexMemPool`
as an example.
Usage is e.g. like this:
```
./bench_bitcoin -filter=ComplexMemPool -asymptote=25,50,100,200,400,600,800
```
This runs the benchmark `ComplexMemPool` several times but with
different complexityN settings. The benchmark can extract that number
and use it accordingly. Here, it's used for `childTxs`. The output is
this:
| complexityN | ns/op | op/s | err% | ins/op | cyc/op | IPC | total | benchmark
|------------:|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|----------:|:----------
| 25 | 1,064,241.00 | 939.64 | 1.4% | 3,960,279.00 | 2,829,708.00 | 1.400 | 0.01 | `ComplexMemPool`
| 50 | 1,579,530.00 | 633.10 | 1.0% | 6,231,810.00 | 4,412,674.00 | 1.412 | 0.02 | `ComplexMemPool`
| 100 | 4,022,774.00 | 248.58 | 0.6% | 16,544,406.00 | 11,889,535.00 | 1.392 | 0.04 | `ComplexMemPool`
| 200 | 15,390,986.00 | 64.97 | 0.2% | 63,904,254.00 | 47,731,705.00 | 1.339 | 0.17 | `ComplexMemPool`
| 400 | 69,394,711.00 | 14.41 | 0.1% | 272,602,461.00 | 219,014,691.00 | 1.245 | 0.76 | `ComplexMemPool`
| 600 | 168,977,165.00 | 5.92 | 0.1% | 639,108,082.00 | 535,316,887.00 | 1.194 | 1.86 | `ComplexMemPool`
| 800 | 310,109,077.00 | 3.22 | 0.1% |1,149,134,246.00 | 984,620,812.00 | 1.167 | 3.41 | `ComplexMemPool`
| coefficient | err% | complexity
|--------------:|-------:|------------
| 4.78486e-07 | 4.5% | O(n^2)
| 6.38557e-10 | 21.7% | O(n^3)
| 3.42338e-05 | 38.0% | O(n log n)
| 0.000313914 | 46.9% | O(n)
| 0.0129823 | 114.4% | O(log n)
| 0.0815055 | 133.8% | O(1)
The best fitting curve is O(n^2), so the algorithm seems to scale
quadratic with `childTxs` in the range 25 to 800.
f852761aec guix: Add clarifying documentation for V env var (Carl Dong)
85f4a4b082 guix: Make V=1 more powerful for debugging (Carl Dong)
Pull request description:
```
- Print commands in both unexpanded and expanded forms
- Set VERBOSE=1 for CMake
```
Ping MarcoFalke hopefully you use `V=1` already for the Guix builds on DrahtBot?
ACKs for top commit:
fanquake:
ACK f852761aec. Ran a Windows Guix build and compared the output from master and this PR when using `V=1`. i.e `HOSTS=x86_64-w64-mingw32 PATH="/root/.config/guix/current/bin${PATH:+:}$PATH" V=1 ./contrib/guix/guix-build.sh`.
Tree-SHA512: 8bc466fa7b869618bbd5a0a91c6b23d4785009289f8dfb93b0349317463a9ab9ece128c72436e02a0819722a63e703100aed15807867a716fda891292fcb9d9d
bfe1ba2f5b rel-builds: Specify core.abbrev for git-rev-parse (Carl Dong)
27e63e01cc build: Accomodate makensis v2.x (Carl Dong)
1f2c39a30e guix: Remove logical cores requirement (Carl Dong)
a4f6ffa71e lint: Also enable source statements for non-gitian (Carl Dong)
d256f91cb1 rel-builds: Directly deploy win installer to OUTDIR (Carl Dong)
fa791da02f nsis: Specify OutFile path only once (Carl Dong)
14701604d0 guix: Expose GIT_COMMON_DIR in container as readonly (Carl Dong)
f5a6ac4f48 guix: Make source tarball using git-archive (Carl Dong)
395c1137f6 gitian: Limit sourced script to just assignments (Carl Dong)
Pull request description:
Based on: #18556
Related: https://github.com/bitcoin/bitcoin/pull/17595#discussion_r399728721
ACKs for top commit:
fanquake:
ACK bfe1ba2f5b - I agree with Carl, and am going to merge this. I'd like for Linux Guix builds to be working again, and we can rebase #18818.
Tree-SHA512: c87ada7e3de17ca0b692a91029b86573442ded5780fc081c214773f6b374a0cdbeaf6f6898c36669c2e247ee32aa7f82defb1180f8decac52c65f0c140f18674
fa13090d20 contrib: Remove optimize-pngs.py script, which lives in the maintainer repo (MarcoFalke)
Pull request description:
Moved to https://github.com/bitcoin-core/bitcoin-maintainer-tools/blob/master/optimize-pngs.py
Bitcoin Core should focus on the full node implementation, not on scripts to compress png images.
This script is only used when new PNG files are added to the repo. This happens about once every two years. So fetching the script from the other repo should not be a burden, but removing it from this repo is going to cut down on the meta files we need to maintain in the main repo.
ACKs for top commit:
practicalswift:
ACK fa13090d20 -- `+0 lines, -82 lines` :)
promag:
ACK fa13090d20.
hebasto:
ACK fa13090d20, verified that script is already [moved](https://github.com/bitcoin-core/bitcoin-maintainer-tools/pull/56).
Tree-SHA512: 37d111adae769bcddc6ae88041032d5a2b8b228fec67f555c8333c38de3992f5138b30bea868d7d6d6b7f966a47133e5853134373b149ab23cba3b8b560ecb31
When using worktrees or submodules, you'll see a `.git' plain text file
at the root of your working tree instead of the usual `.git' directory.
This plain text file will point to the real GIT_DIR, under the
GIT_COMMON_DIR. From experimentation, the full GIT_COMMON_DIR is
required to exist for operations such as git-archive(1), so we expose it
as readonly inside the container.
Previously, the sourced script would create the source tarball. Now, it
only assigns variables and the source-ing script has more flexibility in
determining what to do with these variables.
See later commit showing how this flexibility is useful in our Guix
builds.
2aa48edec0 refactor: Drop unused ${WRAP_DIR}/${HOST} directory (Hennadii Stepanov)
1362be0447 build: Drop make dist in gitian builds (Hennadii Stepanov)
Pull request description:
After the merge of #18331, the packaged source tarball is created by `git archive`, but the binaries are built from another one which is made by `make dist`.
With this PR the only source tarball, created by `git archive`, is used both for binaries building and for packaging to users.
Close#16588.
Close#18547.
As a good side-effect, #18349 becomes redundant.
**Change in behavior**
The following variables 1b151e3ffc/configure.ac (L2-L6)
are no longer used for naming of directories and tarballs.
Instead of them the gitian descriptors use a git tag (if available) or a commit hash.
---
Also a small refactor commit picked from #18404.
ACKs for top commit:
dongcarl:
ACK 2aa48edec0
MarcoFalke:
ACK 2aa48edec0
fanquake:
ACK 2aa48edec0 - I've had a quick look over this, and don't want to block merging if this actually gets as closer to finally having this all sorted out. Obviously we've still got #18741, and after speaking to Carl this morning, there will likely be even more changes after that (not Guix specific).
Tree-SHA512: d3b16f87e48d1790a3264940c28acd5d881bfd10f3ce94fb0c8a6af76d8039289d01e0cd4972adac49ae24362857251f6c1e5e09e3e9fbf636c10708b4015a7c
3e38023af7 scripts: add PE .reloc section check to security-check.py (fanquake)
Pull request description:
The `ld` in binutils has historically had a few issues with PE binaries, there's a good summary in this [thread](https://sourceware.org/bugzilla/show_bug.cgi?id=19011).
One issue in particular was `ld` stripping the `.reloc` section out of PE binaries, even though it's required for functioning ASLR. This was [reported by a Tor developer in 2014](https://sourceware.org/bugzilla/show_bug.cgi?id=17321) and they have been patching their [own binutils](https://gitweb.torproject.org/builders/tor-browser-build.git/tree/projects/binutils) ever since. However their patch only made it into binutils at the [start of this year](https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=commit;h=dc9bd8c92af67947db44b3cb428c050259b15cd0). It adds an `--enable-reloc-section` flag, which is turned on by default if you are using `--dynamic-base`. In the mean time this issue has also been worked around by other projects, such as FFmpeg, see [this commit](91b668acd6).
I have checked our recent supported Windows release binaries, and they do contain a `.reloc` section. From what I understand, we are using all the right compile/linker flags, including `-pie` & `-fPIE`, and have never run into the crashing/entrypoint issues that other projects might have seen.
One other thing worth noting here, it how Debian/Ubuntu patch the binutils that they distribute, because that's what we end up using in our gitian builds.
In the binutils-mingw-w64 in Bionic (18.04), which we currently use in gitian, PE hardening options/security flags are enabled by default. See the [changelog](https://changelogs.ubuntu.com/changelogs/pool/universe/b/binutils-mingw-w64/binutils-mingw-w64_8ubuntu1/changelog) and the [relevant commit](452b3013b8).
However in Focal (20.04), this has now been reversed. PE hardening options are no-longer the default. See the [changelog](https://changelogs.ubuntu.com/changelogs/pool/universe/b/binutils-mingw-w64/binutils-mingw-w64_8.8/changelog) and [relevant commit](7bd8b2fbc2), which cites same .reloc issue mentioned here.
Given that we explicitly specify/opt-in to everything that we want to use, the defaults aren't necessarily an issue for us. However I think it highlights the importance of continuing to be explicit about what we want, and not falling-back or relying on upstream.
This was also prompted by the possibility of us doing link time garbage collection, see #18579 & #18605. It seemed some sanity checks would be worthwhile in-case the linker goes haywire while garbage collecting.
I think Guix is going to bring great benefits when dealing with these kinds of issues. Carl you might have something to say in that regard.
ACKs for top commit:
dongcarl:
ACK 3e38023af7
Tree-SHA512: af14d63bdb334bde548dd7de3e0946556b7e2598d817b56eb4e75b3f56c705c26aa85dd9783134c4b6a7aeb7cb4de567eed996e94d533d31511f57ed332287da
eb37275a6f Fix naming of macOS SDK and clarify version (Andrew Chow)
Pull request description:
Fixes the `MacOSX10.14.sdk.tar.gz` creation command to have `MacOSX.sdk` be correctly named as `MacOSX10.14.sdk` and for the resulting file to be placed in the current directory. Gitian requires that `tar.gz` contains a folder named `MacOSX10.14.sdk` and the command did not do this originally. Having the file be placed in the current directory is a convenience so builders don't have to go find it.
Also clarifies which version of Xcode to download and where it can be downloaded.
ACKs for top commit:
fanquake:
ACK eb37275a6f - tested the macOS and Linux SDK extraction. Also noticed something seemingly broken with Apple `tar`, but will open an issue to follow up.
Sjors:
ACK eb37275 for the macOS instruction
Tree-SHA512: d691e14711cf195999291dd6fb7ffe552c86f8b30d2b1a77e88b4db6050dd817ba128b047cf36d29b0bb0d4183e709b7c03aa27f31b64e562ea8cd948434ca55
fa4632c417 test: Move boost/stdlib includes last (MarcoFalke)
fa488f131f scripted-diff: Bump copyright headers (MarcoFalke)
fac5c37300 scripted-diff: Sort test includes (MarcoFalke)
Pull request description:
When writing tests, often includes need to be added or removed. Currently the list of includes is not sorted, so developers that write tests and have `clang-format` installed will either have an unrelated change (sorting) included in their commit or they will have to manually undo the sort.
This pull preempts both issues by just sorting all includes in one commit.
Please be aware that this is **NOT** a change to policy to enforce clang-format or any other developer guideline or process. Developers are free to use whatever tool they want, see also #18651.
Edit: Also includes a commit to bump the copyright headers, so that the touched files don't need to be touched again for that.
ACKs for top commit:
practicalswift:
ACK fa4632c417
jonatack:
ACK fa4632c417, light review and sanity checks with gcc build and clang fuzz build
Tree-SHA512: 130a8d073a379ba556b1e64104d37c46b671425c0aef0ed725fd60156a95e8dc83fb6f0b5330b2f8152cf5daaf3983b4aca5e75812598f2626c39fd12b88b180
e44aeefaae gitian: Add missing automake package to gitian-win-signer.yml (Andrew Chow)
Pull request description:
automake is needed to build osslsigncode otherwise autogen.sh fails with the docker virtualization method.
ACKs for top commit:
hebasto:
ACK e44aeefaae, for `osslsigncode-1.7.1` we did not run `autogen.sh` in the past.
fanquake:
ACK e44aeefaae
jonatack:
ACK e44aeef
Tree-SHA512: a0e615c1b099ee1c469ce41f886f2ece6746234a5a800743a4e8be671e4114fd30e1c35bc0ddcb75778409564129d0fde7ac4e3d70b0f7691f97f729f34c8e0c
Wladimir J. van der Laan