[llvm] [llvm-symbolizer] Have --obj participate in markup resolution (PR #183444)
Joshua Seaton via llvm-commits
llvm-commits at lists.llvm.org
Sun Mar 8 18:25:45 PDT 2026
https://github.com/joshuaseaton updated https://github.com/llvm/llvm-project/pull/183444
>From 4dbb209133aa7c422a6cfdfa7fbff97bf42d3cbd Mon Sep 17 00:00:00 2001
From: Joshua Seaton <josh.a.seaton at gmail.com>
Date: Wed, 25 Feb 2026 21:36:05 -0800
Subject: [PATCH 1/2] [llvm-symbolizer] Have --obj participate in markup
resolution
Sometimes in development one is looking to symbolize the backtrace from
an executable that's *right there*. It is ergonomic to be able to simply
pass to to llvm-symbolizer via --obj, while in that case installing it
in a build ID directory and passing it through indirection is less than.
In this change we update --obj and --debug-file-directory to be able to
both independently contribute to the set of binaries to participating in
markup resolution.
---
llvm/lib/DebugInfo/Symbolize/Symbolize.cpp | 6 ++++++
.../DebugInfo/symbolize-filter-markup-obj.test | 16 ++++++++++++++++
llvm/tools/llvm-symbolizer/llvm-symbolizer.cpp | 18 +++++++++++-------
3 files changed, 33 insertions(+), 7 deletions(-)
create mode 100644 llvm/test/DebugInfo/symbolize-filter-markup-obj.test
diff --git a/llvm/lib/DebugInfo/Symbolize/Symbolize.cpp b/llvm/lib/DebugInfo/Symbolize/Symbolize.cpp
index 3821f53d26b98..8276d50fe0158 100644
--- a/llvm/lib/DebugInfo/Symbolize/Symbolize.cpp
+++ b/llvm/lib/DebugInfo/Symbolize/Symbolize.cpp
@@ -769,6 +769,12 @@ LLVMSymbolizer::getOrCreateModuleInfo(StringRef ModuleName) {
}
ObjectPair Objects = ObjectsOrErr.get();
+ // Register the build ID so that build-ID-based lookups (e.g., from
+ // --filter-markup) can resolve to this path.
+ BuildIDRef BID = getBuildID(Objects.first);
+ if (!BID.empty())
+ BuildIDPaths[getBuildIDStr(BID)] = std::string(BinaryName);
+
std::unique_ptr<DIContext> Context;
// If this is a COFF object containing PDB info and not containing DWARF
// section, use a PDBContext to symbolize. Otherwise, use DWARF.
diff --git a/llvm/test/DebugInfo/symbolize-filter-markup-obj.test b/llvm/test/DebugInfo/symbolize-filter-markup-obj.test
new file mode 100644
index 0000000000000..017295d400bd5
--- /dev/null
+++ b/llvm/test/DebugInfo/symbolize-filter-markup-obj.test
@@ -0,0 +1,16 @@
+## Test that --obj populates the build ID cache for --filter-markup.
+## The debug binary is reused from symbolize-build-id.test.
+
+# RUN: split-file %s %t
+# RUN: llvm-symbolizer --obj=%p/Inputs/.build-id/ab/b50d82b6bdc861.debug \
+# RUN: --filter-markup < %t/input > %t.output 2>&1
+# RUN: FileCheck %s --input-file=%t.output --match-full-lines \
+# RUN: --implicit-check-not {{.}}
+
+# CHECK: [[BEGIN:\[{3}]]ELF module #0x0 "a.o"; BuildID=abb50d82b6bdc861 [0x0-0x2fffff](r)[[END:\]{3}]]
+# CHECK: main[/tmp/dbginfo[[SEP:[/\\]]]dwarfdump-test.cc:16]
+
+;--- input
+{{{module:0:a.o:elf:abb50d82b6bdc861}}}
+{{{mmap:0:0x300000:load:0:r:0}}}
+{{{pc:0x20112f}}}
diff --git a/llvm/tools/llvm-symbolizer/llvm-symbolizer.cpp b/llvm/tools/llvm-symbolizer/llvm-symbolizer.cpp
index 4784dafeb2948..dfad7f991ac64 100644
--- a/llvm/tools/llvm-symbolizer/llvm-symbolizer.cpp
+++ b/llvm/tools/llvm-symbolizer/llvm-symbolizer.cpp
@@ -463,7 +463,8 @@ static object::BuildID parseBuildIDArg(const opt::InputArgList &Args, int ID) {
}
// Symbolize markup from stdin and write the result to stdout.
-static void filterMarkup(const opt::InputArgList &Args, LLVMSymbolizer &Symbolizer) {
+static void filterMarkup(const opt::InputArgList &Args,
+ LLVMSymbolizer &Symbolizer) {
MarkupFilter Filter(outs(), Symbolizer, parseColorArg(Args));
std::string InputString;
while (std::getline(std::cin, InputString)) {
@@ -540,11 +541,6 @@ int llvm_symbolizer_main(int argc, char **argv, const llvm::ToolContext &) {
if (Args.hasFlag(OPT_debuginfod, OPT_no_debuginfod, canUseDebuginfod()))
enableDebuginfod(Symbolizer, Args);
- if (Args.hasArg(OPT_filter_markup)) {
- filterMarkup(Args, Symbolizer);
- return 0;
- }
-
auto Style = IsAddr2Line ? OutputStyle::GNU : OutputStyle::LLVM;
if (const opt::Arg *A = Args.getLastArg(OPT_output_style_EQ)) {
if (strcmp(A->getValue(), "GNU") == 0)
@@ -559,7 +555,6 @@ int llvm_symbolizer_main(int argc, char **argv, const llvm::ToolContext &) {
errs() << "error: cannot specify both --build-id and --obj\n";
return EXIT_FAILURE;
}
- object::BuildID BuildID = parseBuildIDArg(Args, OPT_build_id_EQ);
std::unique_ptr<DIPrinter> Printer;
if (Style == OutputStyle::GNU)
@@ -583,6 +578,15 @@ int llvm_symbolizer_main(int argc, char **argv, const llvm::ToolContext &) {
}
}
+ // This is sequenced after object loading so that the build ID cache can
+ // participate in the resolution of modules referenced in the markup.
+ if (Args.hasArg(OPT_filter_markup)) {
+ filterMarkup(Args, Symbolizer);
+ return 0;
+ }
+
+ object::BuildID BuildID = parseBuildIDArg(Args, OPT_build_id_EQ);
+
std::vector<std::string> InputAddresses = Args.getAllArgValues(OPT_INPUT);
if (InputAddresses.empty()) {
const int kMaxInputStringLength = 1024;
>From 05a73f1d48f087daa8ee25236441027328aa2fc7 Mon Sep 17 00:00:00 2001
From: Joshua Seaton <josh.a.seaton at gmail.com>
Date: Sun, 8 Mar 2026 18:17:46 -0700
Subject: [PATCH 2/2] [llvm-symbolizer] Have --obj participate in markup
resolution
This change concerns the interplay between --filter-markup and --obj.
Today the --filter-markup path does not consult any of the --obj
specifications when it comes to build ID lookup. We update things so
that it does.
Two reasons for doing so:
* Sometimes in development one is looking to symbolize the backtrace
that an executable that's *right there*. It is ergonomic to be able to
simplypass to to llvm-symbolizer via --obj, while in that case
installing it in a build ID directory and passing it through indirection
is less than.
* In the current command-line interface, it feels natural for --obj
--debug-file-directory to both be able to independently contribute to
the set of binaries participating in markup resolution.
---
llvm/lib/DebugInfo/Symbolize/Symbolize.cpp | 11 ++++++++++-
1 file changed, 10 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/DebugInfo/Symbolize/Symbolize.cpp b/llvm/lib/DebugInfo/Symbolize/Symbolize.cpp
index 8276d50fe0158..84da834cef183 100644
--- a/llvm/lib/DebugInfo/Symbolize/Symbolize.cpp
+++ b/llvm/lib/DebugInfo/Symbolize/Symbolize.cpp
@@ -555,8 +555,17 @@ LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
DbgObj = lookUpBuildIDObject(Path, ELFObj, ArchName);
if (!DbgObj)
DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
- if (!DbgObj)
+ if (!DbgObj) {
+ // No previously cached debug binary found. As a best-effort, we use the
+ // provided object instead, and also register its build ID (if any) so that
+ // build-ID-based lookups (e.g., from --filter-markup) can resolve to this
+ // path.
DbgObj = Obj;
+
+ BuildIDRef BID = getBuildID(Obj);
+ if (!BID.empty())
+ BuildIDPaths[getBuildIDStr(BID)] = std::string(Path);
+ }
ObjectPair Res = std::make_pair(Obj, DbgObj);
auto Pair =
ObjectPairForPathArch.emplace(std::make_pair(Path, ArchName), Res);
More information about the llvm-commits
mailing list