[LLD] [COFF] Pick timestamps from the SOURCE_DATE_EPOCH variable

[mstorsjo]

The SOURCE_DATE_EPOCH environment variable can be set in order to get reproducible build.

When linking PE/COFF modules with LLD, the timestamp field is set to the current time, unless either the /timestamp: or /Brepro option is set. If neither of them is set, check the SOURCE_DATE_EPOCH variable, before resorting to using the actual current date and time.

[llvmbot]

@llvm/pr-subscribers-lld
@llvm/pr-subscribers-lld-coff

@llvm/pr-subscribers-platform-windows

Author: Martin Storsjö (mstorsjo)

Changes

The SOURCE_DATE_EPOCH environment variable can be set in order to get reproducible build.

When linking PE/COFF modules with LLD, the timestamp field is set to the current time, unless either the /timestamp: or /Brepro option is set. If neither of them is set, check the SOURCE_DATE_EPOCH variable, before resorting to using the actual current date and time.

---

Full diff: https://github.com/llvm/llvm-project/pull/81326.diff

2 Files Affected:

• (modified) lld/COFF/Driver.cpp (+9-1)
• (modified) lld/test/COFF/timestamp.test (+2)

diff --git a/lld/COFF/Driver.cpp b/lld/COFF/Driver.cpp
index e0afb6b18805b2..22ee2f133be98a 100644
--- a/lld/COFF/Driver.cpp
+++ b/lld/COFF/Driver.cpp
@@ -1825,7 +1825,15 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
     }
   } else {
     config->repro = false;
-    config->timestamp = time(nullptr);
+    if (std::optional<std::string> epoch =
+            Process::GetEnv("SOURCE_DATE_EPOCH")) {
+      StringRef value(*epoch);
+      if (value.getAsInteger(0, config->timestamp))
+        fatal(Twine("invalid SOURCE_DATE_EPOCH timestamp: ") + value +
+              ".  Expected 32-bit integer");
+    } else {
+      config->timestamp = time(nullptr);
+    }
   }
 
   // Handle /alternatename
diff --git a/lld/test/COFF/timestamp.test b/lld/test/COFF/timestamp.test
index fbdc5788a33a55..f94eeee8793f01 100644
--- a/lld/test/COFF/timestamp.test
+++ b/lld/test/COFF/timestamp.test
@@ -3,9 +3,11 @@ RUN: yaml2obj %p/Inputs/generic.yaml -o %t.obj
 RUN: lld-link %t.obj /debug /Brepro /entry:main /nodefaultlib /out:%t.1.exe
 RUN: lld-link %t.obj /debug /Brepro /entry:main /nodefaultlib /out:%t.2.exe
 RUN: lld-link %t.obj /debug /timestamp:0 /entry:main /nodefaultlib /out:%t.3.exe
+RUN: env SOURCE_DATE_EPOCH=0 lld-link %t.obj /debug /entry:main /nodefaultlib /out:%t.4.exe
 RUN: llvm-readobj --file-headers --coff-debug-directory %t.1.exe | FileCheck %s --check-prefix=HASH
 RUN: llvm-readobj --file-headers --coff-debug-directory %t.2.exe | FileCheck %s --check-prefix=HASH
 RUN: llvm-readobj --file-headers --coff-debug-directory %t.3.exe | FileCheck %s --check-prefix=ZERO
+RUN: llvm-readobj --file-headers --coff-debug-directory %t.4.exe | FileCheck %s --check-prefix=ZERO
 
 HASH: ImageFileHeader {
 HASH: TimeDateStamp: [[STAMP:.*]]

[MaskRay]

It'd be useful to test another value like clang/test/Preprocessor/SOURCE_DATE_EPOCH.c

[mstorsjo]

Thanks, I added more tests for more interesting cases.

What do you think of backporting this to 18.x? It’s not a bug fix per se, but a pretty safe improvement.

[MaskRay]

LGTM for a backport to improve reproducibility!

https://reproducible-builds.org/

[aganea]

There’s not much info online about this. Do you know where it was first introduced?

[aganea]

Ah nvm found it: https://reproducible-builds.org/docs/source-date-epoch/

[aganea]

Does this deserve an update of the release notes for 18.0?

[mstorsjo]

Does this deserve an update of the release notes for 18.0?

Good point - yeah I think it does; I can make a separate PR to update the notes on the branch.

[MaskRay]

Does this deserve an update of the release notes for 18.0?

Good point - yeah I think it does; I can make a separate PR to update the notes on the branch.

The release notes for non-ELF ports are a barebone. @mstorsjo Do you want to find all interesting commits using git log origin/release/17.x..origin/release/18.x -- lld/COFF?

You may create a feature request with the base branch set to release/18.x and then add the release 18.x label.

[mstorsjo]

The release notes for non-ELF ports are a barebone. @mstorsjo Do you want to find all interesting commits using git log origin/release/17.x..origin/release/18.x -- lld/COFF?

I did something similar already - filling in the stuff which is already there - but I only looked at the commits that I have been involved in myself, that I'm more familiar with. But I guess it would be useful to have a look through all of them - it's just a little bit more work. I'll see if I can get to it in the next few days.

[DavidSpickett]

FYI: https://lab.llvm.org/buildbot/#/builders/178/builds/6797/steps/7/logs/FAIL__lld__timestamp_test

Only happens on the 32 bit bot.

[mstorsjo]

FYI: https://lab.llvm.org/buildbot/#/builders/178/builds/6797/steps/7/logs/FAIL__lld__timestamp_test

Only happens on the 32 bit bot.

Oops, sorry about that, and sorry for not noticing the buildbot mail earlier.

It looks like the code itself is right, but llvm-readobj might not be able to print the value correctly on hosts with a 32 bit time_t.

Although, https://learn.microsoft.com/en-us/windows/win32/debug/pe-format doesn't seem to say anything explicitly about the signedness of this field either, but treating it as an unsigned seems reasonable. It does say that the value 0xFFFFFFFF that we're using in the test doesn't represent a real timestamp. So perhaps we should test with a different value.

I guess the simplest way forward is to test with 0x7FFFFFFF which should work safely everywhere. (And potentially test with 0xFFFFFFFE if we know the host isn't limited to a 32 bit time_t?)