[Mlir-commits] [mlir] [mlir][tblgen] Don't echo absolute paths into rewrite pattern source (PR #168984)

[Krzysztof Drewniak]

https://github.com/krzysz00 created https://github.com/llvm/llvm-project/pull/168984

Currently, the declarative pattern rewrite generator will always print the [source]:[line](s) from which a pattern came. This is a useful debugging hint, but it causes problem when absolute paths are used as arguments to mlir-tblgen (which LLVM's build rules automatically do). Specifially, it causes the source to be tied to the build location, harning reproducability and our collective ability to get ccache hits from, say, separate worktrees.

This commit resolves the issue by replacing absolute paths in thes "Generated from:" comments with their filenames. (The alternative would have been to implement an entire file-prefix-map the way the C compilers do, but since this is an isolated incident, I chose to resolve it locally.)

>From dabd1173df98d8b83c78d71354617389b2874ee7 Mon Sep 17 00:00:00 2001
From: Krzysztof Drewniak <Krzysztof.Drewniak at amd.com>
Date: Fri, 21 Nov 2025 01:49:39 +0000
Subject: [PATCH] [mlir][tblgen] Don't echo absolute paths into rewrite pattern
 source

Currently, the declarative pattern rewrite generator will always print
the [source]:[line](s) from which a pattern came. This is a useful
debugging hint, but it causes problem when absolute paths are used as
arguments to mlir-tblgen (which LLVM's build rules automatically do).
Specifially, it causes the source to be tied to the build location,
harning reproducability and our collective ability to get ccache hits
from, say, separate worktrees.

This commit resolves the issue by replacing absolute paths in thes
"Generated from:" comments with their filenames. (The alternative
would have been to implement an entire file-prefix-map the way the C
compilers do, but since this is an isolated incident, I chose to
resolve it locally.)
---
 mlir/include/mlir/TableGen/Pattern.h   |  6 ++++--
 mlir/lib/TableGen/Pattern.cpp          | 21 +++++++++++++++++----
 mlir/tools/mlir-tblgen/RewriterGen.cpp |  3 ++-
 3 files changed, 23 insertions(+), 7 deletions(-)

diff --git a/mlir/include/mlir/TableGen/Pattern.h b/mlir/include/mlir/TableGen/Pattern.h
index 49b2dae62dc22..d2610f09b6241 100644
--- a/mlir/include/mlir/TableGen/Pattern.h
+++ b/mlir/include/mlir/TableGen/Pattern.h
@@ -643,8 +643,10 @@ class Pattern {
   using IdentifierLine = std::pair<StringRef, unsigned>;
 
   // Returns the file location of the pattern (buffer identifier + line number
-  // pair).
-  std::vector<IdentifierLine> getLocation() const;
+  // pair). If `forSourceOutput` is true, replace absolute paths in the buffer
+  // identifier with just their filename so that we don't leak build paths into
+  // the generated code.
+  std::vector<IdentifierLine> getLocation(bool forSourceOutput = false) const;
 
   // Recursively collects all bound symbols inside the DAG tree rooted
   // at `tree` and updates the given `infoMap`.
diff --git a/mlir/lib/TableGen/Pattern.cpp b/mlir/lib/TableGen/Pattern.cpp
index 1a1a58ad271bb..ce09f5c3f5183 100644
--- a/mlir/lib/TableGen/Pattern.cpp
+++ b/mlir/lib/TableGen/Pattern.cpp
@@ -18,6 +18,7 @@
 #include "llvm/ADT/Twine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/FormatVariadic.h"
+#include "llvm/Support/Path.h"
 #include "llvm/TableGen/Error.h"
 #include "llvm/TableGen/Record.h"
 
@@ -771,15 +772,27 @@ int Pattern::getBenefit() const {
   return initBenefit + dyn_cast<IntInit>(delta->getArg(0))->getValue();
 }
 
-std::vector<Pattern::IdentifierLine> Pattern::getLocation() const {
+std::vector<Pattern::IdentifierLine>
+Pattern::getLocation(bool forSourceOutput) const {
   std::vector<std::pair<StringRef, unsigned>> result;
   result.reserve(def.getLoc().size());
   for (auto loc : def.getLoc()) {
     unsigned buf = llvm::SrcMgr.FindBufferContainingLoc(loc);
     assert(buf && "invalid source location");
-    result.emplace_back(
-        llvm::SrcMgr.getBufferInfo(buf).Buffer->getBufferIdentifier(),
-        llvm::SrcMgr.getLineAndColumn(loc, buf).first);
+
+    StringRef bufferName =
+        llvm::SrcMgr.getBufferInfo(buf).Buffer->getBufferIdentifier();
+    // If we're emitting a generated file, we'd like to have some indication of
+    // where our patterns came from. However, LLVM's build rules use absolute
+    // paths as arguments to TableGen, and naively echoing such paths makes the
+    // contents of the generated source file depend on the build location,
+    // making MLIR builds substantially less reproducable. As a compromise, we
+    // trim absolute paths back to only the filename component.
+    if (forSourceOutput && llvm::sys::path::is_absolute(bufferName))
+      bufferName = llvm::sys::path::filename(bufferName);
+
+    result.emplace_back(bufferName,
+                        llvm::SrcMgr.getLineAndColumn(loc, buf).first);
   }
   return result;
 }
diff --git a/mlir/tools/mlir-tblgen/RewriterGen.cpp b/mlir/tools/mlir-tblgen/RewriterGen.cpp
index c3034bb843c4a..7698355c8fa6e 100644
--- a/mlir/tools/mlir-tblgen/RewriterGen.cpp
+++ b/mlir/tools/mlir-tblgen/RewriterGen.cpp
@@ -28,6 +28,7 @@
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/FormatAdapters.h"
+#include "llvm/Support/Path.h"
 #include "llvm/Support/PrettyStackTrace.h"
 #include "llvm/Support/Signals.h"
 #include "llvm/TableGen/Error.h"
@@ -1129,7 +1130,7 @@ void PatternEmitter::emit(StringRef rewriteName) {
   LLVM_DEBUG(llvm::dbgs() << "done collecting ops used in result patterns\n");
 
   // Emit RewritePattern for Pattern.
-  auto locs = pattern.getLocation();
+  auto locs = pattern.getLocation(/*forSourceOutput=*/true);
   os << formatv("/* Generated from:\n    {0:$[ instantiating\n    ]}\n*/\n",
                 llvm::reverse(locs));
   os << formatv(R"(struct {0} : public ::mlir::RewritePattern {