[Windows] Fix Registry static data members not exported by extract_symbols.py in static builds with plugin support

[zond]

Summary

This PR fixes an issue where external plugins cannot register through llvm::Registry<clang::PluginASTAction> when building LLVM statically on Windows with LLVM_EXPORT_SYMBOLS_FOR_PLUGINS=ON.

Problem

When building LLVM with a static configuration (the default) and enabling plugin support via LLVM_EXPORT_SYMBOLS_FOR_PLUGINS=ON, the extract_symbols.py script filters out Registry static data members (Head and Tail) during symbol export because they don't match the function signature patterns that the script looks for.

These static data members are essential for the plugin registry mechanism - they form the linked list that stores registered plugins.

Solution

This patch modifies extract_symbols.py to add pattern matching that recognizes and exports Registry static data members. The pattern matches mangled symbols like:

• ?Head@?$Registry@<template_args>@llvm@@
• ?Tail@?$Registry@<template_args>@llvm@@

Technical Details

When both LLVM and plugins are built with /Zc:dllexportInlines-, the static methods (add_node(), begin()) are inlined by the plugin at call sites. This means only the data symbols (Head, Tail) need to be exported from the executable - no changes to driver.cpp are necessary.

Testing

Verified on Firefox's clang-plugin build for Windows with Clang 20, which uses:

• Static LLVM build
• LLVM_EXPORT_SYMBOLS_FOR_PLUGINS=ON
• /Zc:dllexportInlines- for both LLVM and plugin compilation (which makes inline functions inlined instead of exported)

After this patch, all required Registry symbols are correctly exported from clang.exe and the plugin successfully registers.

Fixes #163367

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-clang

Author: None (zond)

Changes

When building LLVM with CLANG_LINK_CLANG_DYLIB=ON on Windows, external plugins cannot register through llvm::Registry<clang::PluginASTAction> due to two issues:

1. Static data members (Head, Tail) are filtered out during symbol export
2. Static methods (add_node(), begin()) are not linked into the executable

This change fixes both issues:

• Modified extract_symbols.py to recognize and export Registry<T>::Head and Registry<T>::Tail static members using pattern matching instead of filtering them out as non-function symbols.

• Added code in driver.cpp that references Registry methods, forcing the linker to include these symbols in clang.exe even though they're not directly called by the driver.

Fixes #163367

---

Full diff: https://github.com/llvm/llvm-project/pull/163391.diff

2 Files Affected:

• (modified) clang/tools/driver/driver.cpp (+19)
• (modified) llvm/utils/extract_symbols.py (+5)

diff --git a/clang/tools/driver/driver.cpp b/clang/tools/driver/driver.cpp
index 7390d7d610ec0..855c8285e841c 100644
--- a/clang/tools/driver/driver.cpp
+++ b/clang/tools/driver/driver.cpp
@@ -55,6 +55,25 @@
 #include <set>
 #include <system_error>
 
+#if defined(CLANG_PLUGIN_SUPPORT) && defined(_WIN32)
+#include "clang/Frontend/FrontendPluginRegistry.h"
+
+// Force plugin registry symbols into clang.exe on Windows so plugins can
+// register. These methods exist in libraries but aren't linked by default
+// because they're unreferenced. Taking their addresses forces the linker to
+// include them.
+namespace {
+void ForcePluginRegistrySymbols() {
+  using PluginRegistry = llvm::Registry<clang::PluginASTAction>;
+  // Use volatile to prevent the compiler from optimizing away these references
+  volatile auto add_node_ptr = &PluginRegistry::add_node;
+  volatile auto begin_ptr = &PluginRegistry::begin;
+  (void)add_node_ptr;
+  (void)begin_ptr;
+}
+} // anonymous namespace
+#endif
+
 using namespace clang;
 using namespace clang::driver;
 using namespace llvm::opt;
diff --git a/llvm/utils/extract_symbols.py b/llvm/utils/extract_symbols.py
index 388723421d660..72f992f560c7f 100755
--- a/llvm/utils/extract_symbols.py
+++ b/llvm/utils/extract_symbols.py
@@ -105,6 +105,11 @@ def should_keep_microsoft_symbol(symbol, calling_convention_decoration):
     # Skip X86GenMnemonicTables functions, they are not exposed from llvm/include/.
     elif re.match(r"\?is[A-Z0-9]*@X86@llvm", symbol):
         return None
+    # Keep Registry<T>::Head and Registry<T>::Tail static members for plugin support.
+    # Pattern matches: ?Head@?$Registry@<template_args>@llvm@@ or ?Tail@?$Registry@...
+    elif ("?$Registry@" in symbol and "@llvm@@" in symbol and
+          (symbol.startswith("?Head@") or symbol.startswith("?Tail@"))):
+        return symbol
     # Keep mangled llvm:: and clang:: function symbols. How we detect these is a
     # bit of a mess and imprecise, but that avoids having to completely demangle
     # the symbol name. The outermost namespace is at the end of the identifier

[zmodem]

@john-brawn-arm and @glandium are probably more familiar with extract_symbols.py.

[zmodem]

This doesn't look right to me. I think the plugin registry symbols are used in clang.exe already, e.g. at

llvm-project/clang/lib/Frontend/FrontendAction.cpp

Line 435 in 558d935

FrontendPluginRegistry::entries()) {

[zond]

Alright - with /Zc:dllexportInlines- this wasn't necessary - the symbols are now inlined into the plugin as well!

[zond]

@zmodem Alright - now we removed some crud from this PR, funnily with the help of the solution to #163349

[github-actions]

✅ With the latest revision this PR passed the Python code formatter.

[john-brawn-arm]

Adjusting extract_symbols.py seems like a reasonable way of fixing this, as hopefully once #109483 is finished we can get rid of extract_symbols.py and not have to worry about this kind of thing,

[john-brawn-arm]

LGTM

[zond]

LGTM

Should I ping someone to actually merge it in, or will it happen on its own? :D

[john-brawn-arm]

I've merged this.

[github-actions]

@zond Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!