From dd116e699ac15750389ec84fd3c1086722862e2f Mon Sep 17 00:00:00 2001
From: danielzheng <danielzheng@google.com>
Date: Sun, 6 May 2018 01:00:49 -0700
Subject: [PATCH] Fix "Couldn't lookup symbols" error when linking external
 Swift libraries on Linux.

Addresses https://github.com/google/swift/issues/4.

On Linux, the REPL fails with "Couldn't lookup symbols" errors when:
- Linking an external Swift shared library (produced by `swift build`) and
  importing the corresponding Swift module.
- Importing the `TensorFlow` and `Python` modules, without manually linking
  `libswiftPython.so` and `libswiftTensorFlow.so`.

A manual workaround involves specifying the `-lswiftPython` and
`-lswiftTensorFlow` flags (in that specific order) when invoking the REPL.
Also, `Python` and `TensorFlow` must be imported before the external Swift
module to avoid the error.

Conditionally adding the linker flags here seems to solve the issue. This is
robust assuming that toolchain artifacts are not manipulated (so that somehow
`Python.swiftmodule` exists while `libswiftPython.so` doesn't).

PiperOrigin-RevId: 195572160
---
 lib/Driver/UnixToolChains.cpp | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/lib/Driver/UnixToolChains.cpp b/lib/Driver/UnixToolChains.cpp
index e738362f6b1c6..a1bb4788f1a45 100644
--- a/lib/Driver/UnixToolChains.cpp
+++ b/lib/Driver/UnixToolChains.cpp
@@ -283,6 +283,34 @@ toolchains::GenericUnix::constructInvocation(const LinkJobAction &job,
   } else {
     Arguments.push_back(context.Args.MakeArgString(SharedRuntimeLibPath));
     Arguments.push_back("-lswiftCore");
+
+    // SWIFT_ENABLE_TENSORFLOW
+    // On Linux, the REPL fails with "Couldn't lookup symbols" errors when:
+    // - Linking an external Swift shared library (produced by `swift build`)
+    //   and importing the corresponding Swift module.
+    // - Importing the `TensorFlow` and `Python` modules, without manually
+    //   linking `libswiftPython.so` and `libswiftTensorFlow.so`.
+    //
+    // A manual workaround involves specifying the `-lswiftPython` and
+    // `-lswiftTensorFlow` flags (in that specific order) when invoking the
+    // REPL. Also, `Python` and `TensorFlow` must be imported before the
+    // external Swift module to avoid the error.
+    //
+    // Conditionally adding the linker flags here seems to solve the issue.
+    // This is robust assuming that toolchain artifacts are not manipulated
+    // (so that somehow Python.swiftmodule exists while libswiftPython.so
+    // doesn't).
+    //
+    // https://github.com/google/swift/issues/4
+    SmallString<128> swiftPythonLibPath = SharedRuntimeLibPath;
+    llvm::sys::path::append(swiftPythonLibPath, "libswiftPython.so");
+    if (llvm::sys::fs::exists(swiftPythonLibPath))
+      Arguments.push_back("-lswiftPython");
+
+    SmallString<128> swiftTensorFlowLibPath = SharedRuntimeLibPath;
+    llvm::sys::path::append(swiftTensorFlowLibPath, "libswiftTensorFlow.so");
+    if (llvm::sys::fs::exists(swiftTensorFlowLibPath))
+      Arguments.push_back("-lswiftTensorFlow");
   }
 
   // Explicitly pass the target to the linker