Use GetShortPathNameW workaround on Windows for help indexer

Lucene does not accept Unicode paths; it uses thread encoding. On
Windows, that makes it fail for any path that contains characters
not representable in ACP. Using short path name tries to workaround
that when first attempt has failed.

Since Windows 10 build 17134, the UCRT supports using a UTF-8 code
page: see
https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/setlocale-wsetlocale.
This has benefit of using Unicode, but I hesitate using it, since I
am not sure if setting the locale globally to UTF8 is a good idea
(what side effects may it have on other external libraries or our
use of CRT functions, when thread encoding differs from C locale?),
and if only setting it at object creation, it would crash later,
when attempting to access the stored Unicode path with restored old
locale.

Change-Id: I6a3a5a84ba54d8046ec011989289c9c7d5fa5c92
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/108484
Tested-by: Jenkins
Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>

1 files changed, 58 insertions, 5 deletions

diff --git a/helpcompiler/source/HelpIndexer.cxx b/helpcompiler/source/HelpIndexer.cxx
index b8889a619ee9..ad8bd9f4e6df 100644
--- a/helpcompiler/source/HelpIndexer.cxx
+++ b/helpcompiler/source/HelpIndexer.cxx
@@ -20,6 +20,12 @@
 #include <CLucene.h>
 #include <CLucene/analysis/LanguageBasedAnalyzer.h>
 
+#if defined _WIN32
+#include <o3tl/char16_t2wchar_t.hxx>
+#include <prewin.h>
+#include <postwin.h>
+#endif
+
 using namespace lucene::document;
 
 HelpIndexer::HelpIndexer(OUString const &lang, OUString const &module,
@@ -31,6 +37,38 @@ HelpIndexer::HelpIndexer(OUString const &lang, OUString const &module,
     d_contentDir = OUString::Concat(srcDir) + "/content";
 }
 
+#if defined _WIN32
+namespace
+{
+template <class Constructor>
+auto TryWithUnicodePathWorkaround(const OUString& ustrPath, const Constructor& constructor)
+{
+    const rtl_TextEncoding eThreadEncoding = osl_getThreadTextEncoding();
+    OString sPath = OUStringToOString(ustrPath, eThreadEncoding);
+    try
+    {
+        // First try path in thread encoding (ACP in case of Windows).
+        return constructor(sPath);
+    }
+    catch (const CLuceneError&)
+    {
+        // Maybe the path contains characters not representable in ACP. There's no API in lucene
+        // that takes Unicode strings (they take 8-bit strings, and pass them to CRT library
+        // functions without conversion).
+
+        // For a workaround, try short name, which should only contain ASCII characters. Would
+        // not help (i.e., would return original long name) if short (8.3) file name creation is
+        // disabled in OS or volume settings.
+        wchar_t buf[32767];
+        if (GetShortPathNameW(o3tl::toW(ustrPath.getStr()), buf, std::size(buf)) == 0)
+            throw;
+        sPath = OUStringToOString(o3tl::toU(buf), eThreadEncoding);
+        return constructor(sPath);
+    }
+}
+}
+#endif
+
 bool HelpIndexer::indexDocuments()
 {
     if (!scanForFiles())
@@ -51,25 +89,34 @@ bool HelpIndexer::indexDocuments()
         OUString ustrSystemPath;
         osl::File::getSystemPathFromFileURL(d_indexDir, ustrSystemPath);
 
+#if defined _WIN32
+        // Make sure the path exists, or GetShortPathNameW (if attempted) will fail.
+        osl::Directory::createPath(d_indexDir);
+        auto writer = TryWithUnicodePathWorkaround(ustrSystemPath, [&analyzer](const OString& s) {
+            return std::make_unique<lucene::index::IndexWriter>(s.getStr(), analyzer.get(), true);
+        });
+#else
         OString indexDirStr = OUStringToOString(ustrSystemPath, osl_getThreadTextEncoding());
-        lucene::index::IndexWriter writer(indexDirStr.getStr(), analyzer.get(), true);
+        auto writer = std::make_unique<lucene::index::IndexWriter>(indexDirStr.getStr(),
+                                                                   analyzer.get(), true);
+#endif
+
         //Double limit of tokens allowed, otherwise we'll get a too-many-tokens
         //exception for ja help. Could alternative ignore the exception and get
         //truncated results as per java-Lucene apparently
-        writer.setMaxFieldLength(lucene::index::IndexWriter::DEFAULT_MAX_FIELD_LENGTH*2);
+        writer->setMaxFieldLength(lucene::index::IndexWriter::DEFAULT_MAX_FIELD_LENGTH*2);
 
         // Index the identified help files
         Document doc;
         for (auto const& elem : d_files)
         {
             helpDocument(elem, &doc);
-            writer.addDocument(&doc);
+            writer->addDocument(&doc);
             doc.clear();
         }
-        writer.optimize();
 
         // Optimize the index
-        writer.optimize();
+        writer->optimize();
     }
     catch (CLuceneError &e)
     {
@@ -137,8 +184,14 @@ lucene::util::Reader *HelpIndexer::helpFileReader(OUString const & path) {
         file.close();
         OUString ustrSystemPath;
         osl::File::getSystemPathFromFileURL(path, ustrSystemPath);
+#if defined _WIN32
+        return TryWithUnicodePathWorkaround(ustrSystemPath, [](const OString& s) {
+            return _CLNEW lucene::util::FileReader(s.getStr(), "UTF-8");
+        });
+#else
         OString pathStr = OUStringToOString(ustrSystemPath, osl_getThreadTextEncoding());
         return _CLNEW lucene::util::FileReader(pathStr.getStr(), "UTF-8");
+#endif
     } else {
         return _CLNEW lucene::util::StringReader(L"");
     }