backend: support non-ASCII characters in path to llmodel libs on Windows (#2388)

* backend: refactor dlhandle.h into oscompat.{cpp,h}

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* llmodel: alias std::filesystem

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* llmodel: use wide strings for paths on Windows

Using the native path representation allows us to manipulate paths and
call LoadLibraryEx without mangling non-ASCII characters.

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* llmodel: prefer built-in std::filesystem functionality

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* oscompat: fix string type error

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* backend: rename oscompat back to dlhandle

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* dlhandle: fix #includes

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* dlhandle: remove another #include

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* dlhandle: move dlhandle #include

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* dlhandle: remove #includes that are covered by dlhandle.h

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

* llmodel: fix #include order

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

---------

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

gpt4all-backend/CMakeLists.txt

@@ -149,7 +149,7 @@ endforeach()
 add_library(llmodel
     llmodel.h llmodel.cpp llmodel_shared.cpp
     llmodel_c.h llmodel_c.cpp
-    dlhandle.h
+    dlhandle.cpp
 )
 target_compile_definitions(llmodel PRIVATE LIB_FILE_EXT="${CMAKE_SHARED_LIBRARY_SUFFIX}")
 

gpt4all-backend/dlhandle.cpp

@@ -0,0 +1,56 @@
+#include "dlhandle.h"
+
+#ifndef _WIN32
+#   include <dlfcn.h>
+#else
+#   include <sstream>
+#   define WIN32_LEAN_AND_MEAN
+#   ifndef NOMINMAX
+#       define NOMINMAX
+#   endif
+#   include <windows.h>
+#endif
+
+namespace fs = std::filesystem;
+
+
+#ifndef _WIN32
+
+Dlhandle::Dlhandle(const fs::path &fpath) {
+    chandle = dlopen(fpath.c_str(), RTLD_LAZY | RTLD_LOCAL);
+    if (!chandle) {
+        throw Exception("dlopen(\"" + fpath.filename().string() + "\"): " + dlerror());
+    }
+}
+
+Dlhandle::~Dlhandle() {
+    if (chandle) dlclose(chandle);
+}
+
+void *Dlhandle::get_internal(const char *symbol) const {
+    return dlsym(chandle, symbol);
+}
+
+#else // defined(_WIN32)
+
+Dlhandle::Dlhandle(const fs::path &fpath) {
+    auto afpath = fs::absolute(fpath);
+    chandle = LoadLibraryExW(afpath.c_str(), NULL, LOAD_LIBRARY_SEARCH_DEFAULT_DIRS | LOAD_LIBRARY_SEARCH_DLL_LOAD_DIR);
+
+    if (!chandle) {
+        auto err = GetLastError();
+        std::ostringstream ss;
+        ss << "LoadLibraryExW(\"" << fpath.filename().string() << "\") failed with error 0x" << std::hex << err;
+        throw Exception(ss.str());
+    }
+}
+
+Dlhandle::~Dlhandle() {
+    if (chandle) FreeLibrary(HMODULE(chandle));
+}
+
+void *Dlhandle::get_internal(const char *symbol) const {
+    return GetProcAddress(HMODULE(chandle), symbol);
+}
+
+#endif // defined(_WIN32)

gpt4all-backend/dlhandle.h

@@ -1,74 +1,15 @@
-#ifndef DLHANDLE_H
+#pragma once
-#define DLHANDLE_H
-#ifndef _WIN32
-#include <string>
-#include <stdexcept>
-#include <utility>
-#include <dlfcn.h>
 
-
-
-class Dlhandle {
-    void *chandle;
-
-public:
-    class Exception : public std::runtime_error {
-    public:
-        using std::runtime_error::runtime_error;
-    };
-
-    Dlhandle() : chandle(nullptr) {}
-    Dlhandle(const std::string& fpath, int flags = RTLD_LAZY | RTLD_LOCAL) {
-        chandle = dlopen(fpath.c_str(), flags);
-        if (!chandle) {
-            throw Exception("dlopen(\""+fpath+"\"): "+dlerror());
-        }
-    }
-    Dlhandle(const Dlhandle& o) = delete;
-    Dlhandle(Dlhandle&& o) : chandle(o.chandle) {
-        o.chandle = nullptr;
-    }
-    ~Dlhandle() {
-        if (chandle) dlclose(chandle);
-    }
-
-    auto operator =(Dlhandle&& o) {
-        chandle = std::exchange(o.chandle, nullptr);
-    }
-
-    bool is_valid() const {
-        return chandle != nullptr;
-    }
-    operator bool() const {
-        return is_valid();
-    }
-
-    template<typename T>
-    T* get(const std::string& fname) const {
-        auto fres = reinterpret_cast<T*>(dlsym(chandle, fname.c_str()));
-        return (dlerror()==NULL)?fres:nullptr;
-    }
-    auto get_fnc(const std::string& fname) const {
-        return get<void*(...)>(fname);
-    }
-};
-#else
-#include <algorithm>
 #include <filesystem>
-#include <string>
-#include <exception>
 #include <stdexcept>
+#include <string>
+#include <utility>
 
-#define WIN32_LEAN_AND_MEAN
+namespace fs = std::filesystem;
-#ifndef NOMINMAX
-#   define NOMINMAX
-#endif
-#include <windows.h>
-#include <libloaderapi.h>
 
 
 class Dlhandle {
-    HMODULE chandle;
+    void *chandle = nullptr;
 
 public:
     class Exception : public std::runtime_error {
@@ -76,34 +17,31 @@ public:
         using std::runtime_error::runtime_error;
     };
 
-    Dlhandle() : chandle(nullptr) {}
+    Dlhandle() = default;
-    Dlhandle(const std::string& fpath) {
+    Dlhandle(const fs::path &fpath);
-        std::string afpath = std::filesystem::absolute(fpath).string();
-        std::replace(afpath.begin(), afpath.end(), '/', '\\');
-        chandle = LoadLibraryExA(afpath.c_str(), NULL, LOAD_LIBRARY_SEARCH_DEFAULT_DIRS | LOAD_LIBRARY_SEARCH_DLL_LOAD_DIR);
-        if (!chandle) {
-            throw Exception("dlopen(\""+fpath+"\"): Error");
-        }
-    }
     Dlhandle(const Dlhandle &o) = delete;
-    Dlhandle(Dlhandle&& o) : chandle(o.chandle) {
+    Dlhandle(Dlhandle &&o)
+        : chandle(o.chandle)
+    {
         o.chandle = nullptr;
     }
-    ~Dlhandle() {
-        if (chandle) FreeLibrary(chandle);
-    }
 
-    bool is_valid() const {
+    ~Dlhandle();
-        return chandle != nullptr;
+
+    Dlhandle &operator=(Dlhandle &&o) {
+        chandle = std::exchange(o.chandle, nullptr);
+        return *this;
     }
 
     template <typename T>
-    T* get(const std::string& fname) const {
+    T *get(const std::string &symbol) const {
-        return reinterpret_cast<T*>(GetProcAddress(chandle, fname.c_str()));
+        return reinterpret_cast<T *>(get_internal(symbol.c_str()));
     }
-    auto get_fnc(const std::string& fname) const {
+
-        return get<void*(...)>(fname);
+    auto get_fnc(const std::string &symbol) const {
+        return get<void*(...)>(symbol);
     }
+
+private:
+    void *get_internal(const char *symbol) const;
 };
-#endif
-#endif // DLHANDLE_H

gpt4all-backend/llmodel.cpp

@@ -1,6 +1,4 @@
 #include "llmodel.h"
-#include "dlhandle.h"
-#include "sysinfo.h"
 
 #include <cassert>
 #include <cstdlib>
@@ -15,6 +13,9 @@
 #include <unordered_map>
 #include <vector>
 
+#include "dlhandle.h"
+#include "sysinfo.h"
+
 #ifdef _WIN32
 #   define WIN32_LEAN_AND_MEAN
 #   ifndef NOMINMAX
@@ -27,6 +28,8 @@
 #   include <intrin.h>
 #endif
 
+namespace fs = std::filesystem;
+
 #ifndef __APPLE__
 static const std::string DEFAULT_BACKENDS[] = {"kompute", "cpu"};
 #elif defined(__aarch64__)
@@ -129,17 +132,17 @@ const std::vector<LLModel::Implementation> &LLModel::Implementation::implementat
             std::string path;
             // Split the paths string by the delimiter and process each path.
             while (std::getline(ss, path, ';')) {
-                std::filesystem::path fs_path(path);
+                std::u8string u8_path(path.begin(), path.end());
                 // Iterate over all libraries
-                for (const auto& f : std::filesystem::directory_iterator(fs_path)) {
+                for (const auto &f : fs::directory_iterator(u8_path)) {
-                    const std::filesystem::path& p = f.path();
+                    const fs::path &p = f.path();
 
                     if (p.extension() != LIB_FILE_EXT) continue;
                     if (!std::regex_search(p.stem().string(), re)) continue;
 
                     // Add to list if model implementation
                     try {
-                        Dlhandle dl(p.string());
+                        Dlhandle dl(p);
                         if (!isImplementation(dl))
                             continue;
                         fres.emplace_back(Implementation(std::move(dl)));