[driver] Make --version show if assertions, etc. are enabled #87585

porglezomp · 2024-04-04T00:03:06Z

It's useful to have some significant build options visible in the version when investigating problems with a specific compiler artifact. This makes it easy to see if assertions, expensive checks, sanitizers, etc. are enabled when checking a compiler version.

Example output:

clang version 19.0.0git (git@github.com:apple/llvm-project.git b4db1d4e5ceadcd8c0e0c010b36cdc7ae68ba236)
Target: x86_64-apple-darwin24.0.0
Thread model: posix
InstalledDir: /Users/cassie/Development/apple-llvm-project/build/bin
Build configuration: +assertions, +asan, +ubsan

Discussion thread: https://discourse.llvm.org/t/rfc-printing-more-build-config-in-clang-version/78112

It's useful to have some significant build options visible in the version when investigating problems with a specific compiler artifact. This makes it easy to see if assertions, expensive checks, sanitizers, etc. are enabled when checking a compiler version.

llvmbot · 2024-04-04T00:03:38Z

@llvm/pr-subscribers-llvm-support
@llvm/pr-subscribers-clang-driver

@llvm/pr-subscribers-clang

Author: Cassie Jones (porglezomp)

Changes

It's useful to have some significant build options visible in the version when investigating problems with a specific compiler artifact. This makes it easy to see if assertions, expensive checks, sanitizers, etc. are enabled when checking a compiler version.

Example output:

clang version 19.0.0git (git@<!-- -->github.com:apple/llvm-project.git b4db1d4e5ceadcd8c0e0c010b36cdc7ae68ba236)
Target: x86_64-apple-darwin24.0.0
Thread model: posix
InstalledDir: /Users/cassie/Development/apple-llvm-project/build/bin
Build configuration: +assertions, +asan, +ubsan

Discussion thread: https://discourse.llvm.org/t/rfc-printing-more-build-config-in-clang-version/78112

Full diff: https://github.com/llvm/llvm-project/pull/87585.diff

2 Files Affected:

(modified) clang/lib/Driver/Driver.cpp (+38)
(added) clang/test/Driver/version-build-config.test (+6)

diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index 7a53764364ce4d..37180efb7ea67b 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -2002,6 +2002,44 @@ void Driver::PrintVersion(const Compilation &C, raw_ostream &OS) const {
   // Print out the install directory.
   OS << "InstalledDir: " << Dir << '\n';
 
+  // Print out build configuration options that impact the compiler's runtime
+  // behavior. Intended for identifying the source of issues when reproducing
+  // changes.
+  std::vector<std::string> BuildOptions = {
+#if !__OPTIMIZE__
+      "+unoptimized",
+#endif
+#ifndef NDEBUG
+      "+assertions",
+#endif
+#ifdef EXPENSIVE_CHECKS
+      "+expensive-checks",
+#endif
+#if __has_feature(address_sanitizer)
+      "+asan",
+#endif
+#if __has_feature(undefined_behavior_sanitizer)
+      "+ubsan",
+#endif
+#if __has_feature(memory_sanitizer)
+      "+msan",
+#endif
+#if __has_feature(dataflow_sanitizer)
+      "+dfsan",
+#endif
+  };
+  if (!BuildOptions.empty()) {
+    OS << "Build configuration: ";
+    bool FirstOption = true;
+    for (const auto &Option : BuildOptions) {
+      if (!FirstOption)
+        OS << ", ";
+      OS << Option;
+      FirstOption = false;
+    }
+    OS << '\n';
+  }
+
   // If configuration files were used, print their paths.
   for (auto ConfigFile : ConfigFiles)
     OS << "Configuration file: " << ConfigFile << '\n';
diff --git a/clang/test/Driver/version-build-config.test b/clang/test/Driver/version-build-config.test
new file mode 100644
index 00000000000000..3d183f372908ea
--- /dev/null
+++ b/clang/test/Driver/version-build-config.test
@@ -0,0 +1,6 @@
+# REQUIRES: asserts
+# RUN: %clang --version 2>&1 | FileCheck %s
+
+# CHECK: clang version
+# When assertions are enabled, we should have a build configuration line that reflects that
+# CHECK: Build configuration: {{.*}}+assertions

porglezomp

Added questions I'd like to see reviewers address here.

porglezomp · 2024-04-04T00:04:12Z

clang/lib/Driver/Driver.cpp

+#if !__OPTIMIZE__
+      "+unoptimized",


Question for reviewers: Is this a good enough mechanism for the "unoptimized" build case, or do we actually want to plumb the CMAKE_BUILD_TYPE through? I like this because it doesn't expose extra information to the build, and I don't think distinguishing the different types of optimized builds is helpful.

Downside: doesn't work if you build clang with msvc. I don't know if that's an expected build configuration.

Building clang/LLVM with MSVC is a supported configuration and is tested with post-commit bots, so that is definitely an expected configuration. I'm not exactly sure what the best course of action is though. I don't think there's a universally supported preprocessor macro for whether or not optimizations are enabled.

CMAKE_BUILD_TYPE I think would be the best bet. #72021 made it so that there are a fixed number of values (unless additional flags are set). That would also let the user differentiate between something like RELEASE which is -O3 and MINSIZEREL, which I believe is -Oz. If there is a different value specified, just displaying something like custom I think should be fine.

I don't really have any opinions on this though and will leave to others to decide what the best course of action should be. Just trying to provide additional context.

I think +unoptimized is good enough for me. CMAKE_BUILD_TYPE doesn't give as much signal as it suggests, since toolchain files can stick in whatever CXX flags they want per build type.

I was also thinking about the fact that the build type doesn't directly correspond to optimized or not.
I should at least make it so it at least doesn't always display +unoptimized on MSVC though, I can fix that.

It turns out llvm/Support/CommandLine has a definition that's approximately correct for MSVC and uses __OPTIMIZE__ for gnu compatible compilers, I switched to using that.

clang/lib/Driver/Driver.cpp

porglezomp · 2024-04-04T00:05:22Z

clang/lib/Driver/Driver.cpp

+      "+dfsan",
+#endif
+  };
+  if (!BuildOptions.empty()) {


Do we want it to be hidden for the default config?
My opinion: yes—because this is helpful to llvm developers but not to compiler end-users.

Do we want a build option to force it hidden that vendors could use?

Making it opt-out, and having the release scripts disable it seems like a good idea to me. It's not actionable for end-users, but if one has to remember to turn it on (or even know it's an option), one might not see the benefit of it in day-to-day development.

Added this as LLVM_VERSION_PRINTER_SHOW_BUILD_CONFIG which defaults on, is there anything I should do to update the release scripts here or is that for someone else to do?

clang/lib/Driver/Driver.cpp

jroelofs · 2024-04-04T01:02:55Z

clang/lib/Driver/Driver.cpp

+  // Print out build configuration options that impact the compiler's runtime
+  // behavior. Intended for identifying the source of issues when reproducing
+  // changes.
+  std::vector<std::string> BuildOptions = {


llvm::SmallVector<llvm::StringRef>

This can be const StringRef BuildOptions[] = {

It seems interleaveComma doesn't work with plain arrays currently (because of using .begin() instead of std::begin()).

Since these are all const char*, you can make this an ArrayRef<StringRef>.
Hopefully that aligns all the comments in this thread :)

I moved this to a global variable to support making this a reusable API between tools, so I think ArrayRef<StringRef> ends up being the appropriate type in the end.

JDevlieghere · 2024-04-04T01:06:38Z

Would it be worth doing this for the llvm tools like dsymutil and dwarfdump as well? They already print "Optimized build (with assertions)" but this PR covers a few more things, is more legible and it would be nice have more uniformity between the two.

clang/lib/Driver/Driver.cpp

Allow us to reuse this logic in the feature printing.

github-actions · 2024-04-04T05:47:28Z

✅ With the latest revision this PR passed the C/C++ code formatter.

This reverts commit 1213ba2.

porglezomp · 2024-04-04T06:06:04Z

Moved the LLVM_IS_DEBUG_BUILD into the llvm/Support headers to reuse it, but thinking more about it I think it would be better to move the printing logic into llvm/Support/CommandLine(?) instead, to make it easier to reuse in other tools as @JDevlieghere is suggesting.

Some compilers released to end-users will want to force-disable printing the build config, since that's intended for compiler developers and not really end-users. Add that flag, defaulting to on, so compiler developers get the most benefit but release engineers can hide it.

porglezomp · 2024-04-04T21:06:24Z

The result of those refactors I now have a new build option to support hiding the version, should I be touching the utils/bazel with this patch?

jroelofs · 2024-04-04T21:16:36Z

The result of those refactors I now have a new build option to support hiding the version, should I be touching the utils/bazel with this patch?

cc @nico

clang/test/Driver/version-build-config.test

llvm/lib/Support/CommandLine.cpp

Co-authored-by: Jon Roelofs <jroelofs@gmail.com>

JDevlieghere

Thanks, this LGTM!

MaskRay · 2024-04-04T23:15:03Z

llvm/lib/Support/CommandLine.cpp

+#if __has_feature(address_sanitizer)
+    "+asan",
+#endif
+#if __has_feature(undefined_behavior_sanitizer)


sort sanitizers alphabetically?

You can add thread_sanitizer and hwaddress_sanitizer as well

Good catch on TSan.
I followed the documented list of __has_feature that I could see, I don't see hwaddress_sanitizer in the list here: https://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-dynamic-analysis

I see it in the tests though, adding.

llvm/lib/Support/CommandLine.cpp

- Sort sanitizers by name - Add some missing sanitizers - Oops! Fix a lifetime issue with the compiler build config

llvm/lib/Support/CommandLine.cpp

llvmbot added clang Clang issues not falling into any other category clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' labels Apr 4, 2024

porglezomp commented Apr 4, 2024

View reviewed changes

porglezomp requested review from jroelofs, AaronBallman and pogo59 April 4, 2024 00:08

jroelofs reviewed Apr 4, 2024

View reviewed changes

MaskRay reviewed Apr 4, 2024

View reviewed changes

clang/lib/Driver/Driver.cpp Outdated Show resolved Hide resolved

porglezomp added 2 commits April 3, 2024 22:01

Address review feedback

efc90ba

Move LLVM_IS_DEBUG_BUILD into Support/CommandLine.h

1213ba2

Allow us to reuse this logic in the feature printing.

llvmbot added the llvm:support label Apr 4, 2024

Revert "Move LLVM_IS_DEBUG_BUILD into Support/CommandLine.h"

82fa43a

This reverts commit 1213ba2.

porglezomp added 2 commits April 4, 2024 13:38

[Support] Expose config printing as llvm::cl::printBuildConfig

732388b

porglezomp requested review from boomanaiden154, felipepiovezan, jroelofs and MaskRay April 4, 2024 21:10

jroelofs mentioned this pull request Apr 4, 2024

[llvm][ADT] Use ADL to find begin()/end() in interleave* #87669

Merged

jroelofs reviewed Apr 4, 2024

View reviewed changes

clang/test/Driver/version-build-config.test Outdated Show resolved Hide resolved

jroelofs reviewed Apr 4, 2024

View reviewed changes

llvm/lib/Support/CommandLine.cpp Outdated Show resolved Hide resolved

Apply suggestions from code review

14bc6ff

Co-authored-by: Jon Roelofs <jroelofs@gmail.com>

JDevlieghere approved these changes Apr 4, 2024

View reviewed changes

MaskRay approved these changes Apr 4, 2024

View reviewed changes

porglezomp commented Apr 4, 2024

View reviewed changes

llvm/lib/Support/CommandLine.cpp Outdated Show resolved Hide resolved

More review fixups

16e6332

- Sort sanitizers by name - Add some missing sanitizers - Oops! Fix a lifetime issue with the compiler build config

porglezomp commented Apr 5, 2024

View reviewed changes

llvm/lib/Support/CommandLine.cpp Show resolved Hide resolved

porglezomp added 2 commits April 4, 2024 23:15

Fix empty array issue in getCompilerBuildConfig

1b32113

clang-format

59ba94c

porglezomp merged commit 68b939f into llvm:main Apr 5, 2024
4 checks passed

porglezomp deleted the build-configuration branch April 5, 2024 17:01

chapuni added a commit that referenced this pull request Apr 7, 2024

[Bazel] Enable LLVM_VERSION_PRINTER_SHOW_BUILD_CONFIG (for #87585)

ba5dad3

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[driver] Make --version show if assertions, etc. are enabled #87585

[driver] Make --version show if assertions, etc. are enabled #87585

porglezomp commented Apr 4, 2024

llvmbot commented Apr 4, 2024 •

edited

porglezomp left a comment

porglezomp Apr 4, 2024 •

edited

boomanaiden154 Apr 4, 2024

jroelofs Apr 4, 2024

porglezomp Apr 4, 2024

porglezomp Apr 4, 2024

porglezomp Apr 4, 2024 •

edited

jroelofs Apr 4, 2024

porglezomp Apr 4, 2024

jroelofs Apr 4, 2024

MaskRay Apr 4, 2024

porglezomp Apr 4, 2024

felipepiovezan Apr 4, 2024 •

edited

jroelofs Apr 4, 2024

porglezomp Apr 4, 2024 •

edited

JDevlieghere commented Apr 4, 2024

github-actions bot commented Apr 4, 2024 •

edited

porglezomp commented Apr 4, 2024

porglezomp commented Apr 4, 2024

jroelofs commented Apr 4, 2024

JDevlieghere left a comment

MaskRay Apr 4, 2024

porglezomp Apr 4, 2024 •

edited

[driver] Make --version show if assertions, etc. are enabled #87585

[driver] Make --version show if assertions, etc. are enabled #87585

Conversation

porglezomp commented Apr 4, 2024

llvmbot commented Apr 4, 2024 • edited

porglezomp left a comment

Choose a reason for hiding this comment

porglezomp Apr 4, 2024 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

porglezomp Apr 4, 2024 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

felipepiovezan Apr 4, 2024 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

porglezomp Apr 4, 2024 • edited

Choose a reason for hiding this comment

JDevlieghere commented Apr 4, 2024

github-actions bot commented Apr 4, 2024 • edited

porglezomp commented Apr 4, 2024

porglezomp commented Apr 4, 2024

jroelofs commented Apr 4, 2024

JDevlieghere left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

porglezomp Apr 4, 2024 • edited

Choose a reason for hiding this comment

llvmbot commented Apr 4, 2024 •

edited

porglezomp Apr 4, 2024 •

edited

porglezomp Apr 4, 2024 •

edited

felipepiovezan Apr 4, 2024 •

edited

porglezomp Apr 4, 2024 •

edited

github-actions bot commented Apr 4, 2024 •

edited

porglezomp Apr 4, 2024 •

edited