Category-dependent block compression

CMakeLists.txt

@@ -354,6 +354,7 @@ list(
   LIBDWARFS_SRC
   src/dwarfs/block_cache.cpp
   src/dwarfs/block_compressor.cpp
+  src/dwarfs/block_compressor_parser.cpp
   src/dwarfs/block_manager.cpp
   src/dwarfs/block_range.cpp
   src/dwarfs/builtin_script.cpp

include/dwarfs/block_compressor.h

@@ -65,6 +65,8 @@ class block_compressor {
 
   compression_type type() const { return impl_->type(); }
 
+  std::string describe() const { return impl_->describe(); }
+
   class impl {
    public:
     virtual ~impl() = default;
@@ -77,6 +79,7 @@ class block_compressor {
     compress(std::vector<uint8_t>&& data) const = 0;
 
     virtual compression_type type() const = 0;
+    virtual std::string describe() const = 0;
   };
 
  private:

include/dwarfs/block_compressor_parser.h

@@ -0,0 +1,37 @@
+/* vim:set ts=2 sw=2 sts=2 et: */
+/**
+ * \author     Marcus Holland-Moritz (github@mhxnet.de)
+ * \copyright  Copyright (c) Marcus Holland-Moritz
+ *
+ * This file is part of dwarfs.
+ *
+ * dwarfs is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * dwarfs is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dwarfs.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <string>
+#include <string_view>
+
+#include "dwarfs/block_compressor.h"
+
+namespace dwarfs {
+
+struct block_compressor_parser {
+ public:
+  block_compressor parse(std::string_view arg) const;
+  std::string to_string(block_compressor const& opts) const;
+};
+
+} // namespace dwarfs

include/dwarfs/contextual_option.h

@@ -41,22 +41,22 @@ class contextual_option {
   using policy_type = Policy;
   using context_argument_type = typename policy_type::ContextArgumentType;
   using context_type = typename policy_type::ContextType;
-  using option_type = typename policy_type::OptionType;
+  using value_type = typename policy_type::ValueType;
 
   template <typename OptionType, typename ContextParser, typename OptionParser>
   friend class contextual_option_parser;
 
   contextual_option() = default;
-  explicit contextual_option(option_type const& def)
+  explicit contextual_option(value_type const& def)
       : default_{def} {}
 
-  void set_default(option_type const& val) { default_ = val; }
+  void set_default(value_type const& val) { default_ = val; }
 
-  void add_contextual(context_type const& ctx, option_type const& val) {
-    contextual_[ctx] = val;
+  bool add_contextual(context_type const& ctx, value_type const& val) {
+    return contextual_.emplace(ctx, val).second;
   }
 
-  std::optional<option_type>
+  std::optional<value_type>
   get_optional(context_argument_type const& arg) const {
     if constexpr (std::is_same_v<context_type, context_argument_type>) {
       return get_optional_impl(arg);
@@ -65,7 +65,7 @@ class contextual_option {
     }
   }
 
-  option_type get(context_argument_type const& arg) const {
+  value_type get(context_argument_type const& arg) const {
     if constexpr (std::is_same_v<context_type, context_argument_type>) {
       return get_impl(arg);
     } else {
@@ -73,9 +73,9 @@ class contextual_option {
     }
   }
 
-  std::optional<option_type> get_optional() const { return default_; }
+  std::optional<value_type> get_optional() const { return default_; }
 
-  option_type get() const { return default_.value(); }
+  value_type get() const { return default_.value(); }
 
   template <typename T>
   bool any_is(T&& pred) const {
@@ -88,22 +88,22 @@ class contextual_option {
   }
 
  private:
-  std::optional<option_type> get_optional_impl(context_type const& ctx) const {
+  std::optional<value_type> get_optional_impl(context_type const& ctx) const {
     if (auto it = contextual_.find(ctx); it != contextual_.end()) {
       return it->second;
     }
     return default_;
   }
 
-  option_type get_impl(context_type const& ctx) const {
+  value_type get_impl(context_type const& ctx) const {
     if (auto it = contextual_.find(ctx); it != contextual_.end()) {
       return it->second;
     }
     return default_.value();
   }
 
-  std::optional<option_type> default_;
-  std::unordered_map<context_type, option_type> contextual_;
+  std::optional<value_type> default_;
+  std::unordered_map<context_type, value_type> contextual_;
 };
 
 template <typename OptionType, typename ContextParser, typename OptionParser>
@@ -132,10 +132,10 @@ class contextual_option_parser {
                           std::invoke_result_t<decltype(&ContextParser::parse),
                                                ContextParser, decltype(ctx)>,
                           typename option_type::context_type>) {
-          opt_.add_contextual(cp_.parse(ctx), val);
+          add_contextual(cp_.parse(ctx), val);
         } else {
           for (auto c : cp_.parse(ctx)) {
-            opt_.add_contextual(c, val);
+            add_contextual(c, val);
           }
         }
       }
@@ -172,6 +172,14 @@ class contextual_option_parser {
   }
 
  private:
+  void add_contextual(typename option_type::context_type const& ctx,
+                      typename option_type::value_type const& val) const {
+    if (!opt_.add_contextual(ctx, val)) {
+      throw std::runtime_error(fmt::format(
+          "duplicate context '{}' for option '{}'", cp_.to_string(ctx), name_));
+    }
+  }
+
   OptionType& opt_;
   ContextParser const& cp_;
   OptionParser const& op_;

include/dwarfs/fragment_order_parser.h

@@ -21,6 +21,7 @@
 
 #pragma once
 
+#include <string>
 #include <string_view>
 
 #include "dwarfs/options.h"

include/dwarfs/options.h

@@ -44,7 +44,7 @@ template <typename T>
 struct categorized_option_policy {
   using ContextArgumentType = fragment_category;
   using ContextType = fragment_category::value_type;
-  using OptionType = T;
+  using ValueType = T;
 
   static ContextType context_from_arg(ContextArgumentType const& arg) {
     return arg.value();
@@ -53,9 +53,9 @@ struct categorized_option_policy {
 
 } // namespace detail
 
-template <typename OptionType>
+template <typename ValueType>
 using categorized_option =
-    contextual_option<detail::categorized_option_policy<OptionType>>;
+    contextual_option<detail::categorized_option_policy<ValueType>>;
 
 enum class mlock_mode { NONE, TRY, MUST };
 

src/dwarfs/block_compressor_parser.cpp

@@ -0,0 +1,35 @@
+/* vim:set ts=2 sw=2 sts=2 et: */
+/**
+ * \author     Marcus Holland-Moritz (github@mhxnet.de)
+ * \copyright  Copyright (c) Marcus Holland-Moritz
+ *
+ * This file is part of dwarfs.
+ *
+ * dwarfs is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * dwarfs is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dwarfs.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include "dwarfs/block_compressor_parser.h"
+
+namespace dwarfs {
+
+block_compressor block_compressor_parser::parse(std::string_view arg) const {
+  return block_compressor(std::string(arg));
+}
+
+std::string
+block_compressor_parser::to_string(block_compressor const& bc) const {
+  return bc.describe();
+}
+
+} // namespace dwarfs

src/dwarfs/category_parser.cpp

@@ -33,7 +33,7 @@ std::vector<fragment_category::value_type>
 category_parser::parse(std::string_view arg) const {
   if (!catmgr_) {
     throw std::runtime_error(
-        "cannot configure category-specific options without any categories");
+        "cannot configure category-specific options without any categorizers");
   }
 
   std::vector<fragment_category::value_type> rv;

src/dwarfs/compression/brotli.cpp

@@ -75,6 +75,10 @@ class brotli_block_compressor final : public block_compressor::impl {
 
   compression_type type() const override { return compression_type::BROTLI; }
 
+  std::string describe() const override {
+    return fmt::format("brotli [quality={}, lgwin={}]", quality_, window_bits_);
+  }
+
  private:
   uint32_t const quality_;
   uint32_t const window_bits_;

src/dwarfs/compression/lz4.cpp

@@ -38,6 +38,8 @@ struct lz4_compression_policy {
         reinterpret_cast<const char*>(src), reinterpret_cast<char*>(dest),
         folly::to<int>(size), folly::to<int>(destsize)));
   }
+
+  static std::string describe(int /*level*/) { return "lz4"; }
 };
 
 struct lz4hc_compression_policy {
@@ -47,6 +49,10 @@ struct lz4hc_compression_policy {
         reinterpret_cast<const char*>(src), reinterpret_cast<char*>(dest),
         folly::to<int>(size), folly::to<int>(destsize), level));
   }
+
+  static std::string describe(int level) {
+    return fmt::format("lz4hc [level={}]", level);
+  }
 };
 
 template <typename Policy>
@@ -84,6 +90,8 @@ class lz4_block_compressor final : public block_compressor::impl {
 
   compression_type type() const override { return compression_type::LZ4; }
 
+  std::string describe() const override { return Policy::describe(level_); }
+
  private:
   const int level_;
 };

src/dwarfs/compression/lzma.cpp

@@ -71,6 +71,8 @@ class lzma_block_compressor final : public block_compressor::impl {
 
   compression_type type() const override { return compression_type::LZMA; }
 
+  std::string describe() const override { return description_; }
+
  private:
   std::vector<uint8_t>
   compress(const std::vector<uint8_t>& data, const lzma_filter* filters) const;
@@ -107,11 +109,16 @@ class lzma_block_compressor final : public block_compressor::impl {
 
   lzma_options_lzma opt_lzma_;
   std::array<lzma_filter, 3> filters_;
+  std::string description_;
 };
 
 lzma_block_compressor::lzma_block_compressor(unsigned level, bool extreme,
                                              const std::string& binary_mode,
-                                             unsigned dict_size) {
+                                             unsigned dict_size)
+    : description_{
+          fmt::format("lzma [level={}, dict_size={}{}{}]", level, dict_size,
+                      extreme ? ", extreme" : "",
+                      binary_mode.empty() ? "" : ", binary=" + binary_mode)} {
   if (lzma_lzma_preset(&opt_lzma_, get_preset(level, extreme))) {
     DWARFS_THROW(runtime_error, "unsupported preset, possibly a bug");
   }

src/dwarfs/compression/null.cpp

@@ -47,6 +47,8 @@ class null_block_compressor final : public block_compressor::impl {
   }
 
   compression_type type() const override { return compression_type::NONE; }
+
+  std::string describe() const override { return "null"; }
 };
 
 class null_block_decompressor final : public block_decompressor::impl {

src/dwarfs/compression/zstd.cpp

@@ -63,6 +63,10 @@ class zstd_block_compressor final : public block_compressor::impl {
 
   compression_type type() const override { return compression_type::ZSTD; }
 
+  std::string describe() const override {
+    return fmt::format("zstd [level={}]", level_);
+  }
+
  private:
   class scoped_context;
 

src/mkdwarfs_main.cpp

@@ -51,6 +51,7 @@
 #include <fmt/format.h>
 
 #include "dwarfs/block_compressor.h"
+#include "dwarfs/block_compressor_parser.h"
 #include "dwarfs/block_manager.h"
 #include "dwarfs/builtin_script.h"
 #include "dwarfs/categorizer.h"
@@ -275,13 +276,14 @@ int mkdwarfs_main(int argc, sys_char** argv) {
 
   block_manager::config cfg;
   sys_string path_str, output_str;
-  std::string memory_limit, script_arg, compression, header, schema_compression,
+  std::string memory_limit, script_arg, header, schema_compression,
       metadata_compression, log_level_str, timestamp, time_resolution,
       progress_mode, recompress_opts, pack_metadata, file_hash_algo,
       debug_filter, max_similarity_size, input_list_str, chmod_str,
       categorizer_list_str;
   std::vector<sys_string> filter;
-  std::vector<std::string> order, max_lookback_blocks, window_size, window_step;
+  std::vector<std::string> order, max_lookback_blocks, window_size, window_step,
+      compression;
   size_t num_workers, num_scanner_workers;
   bool no_progress = false, remove_header = false, no_section_index = false,
        force_overwrite = false;
@@ -292,11 +294,12 @@ int mkdwarfs_main(int argc, sys_char** argv) {
   integral_value_parser<size_t> max_lookback_parser;
   integral_value_parser<unsigned> window_size_parser(6, 24);
   integral_value_parser<unsigned> window_step_parser(0, 8);
+  fragment_order_parser order_parser;
+  block_compressor_parser compressor_parser;
 
   scanner_options options;
 
-  auto order_desc =
-      "inode fragments order (" + fragment_order_parser::choices() + ")";
+  auto order_desc = "inode fragments order (" + order_parser.choices() + ")";
 
   auto progress_desc = "progress mode (" +
                        (from(progress_modes) | get<0>() | unsplit(", ")) + ")";
@@ -430,7 +433,8 @@ int mkdwarfs_main(int argc, sys_char** argv) {
   po::options_description compressor_opts("Compressor options");
   compressor_opts.add_options()
     ("compression,C",
-        po::value<std::string>(&compression),
+        // po::value<std::string>(&compression), // TODO
+        po::value<std::vector<std::string>>(&compression)->multitoken(),
         "block compression algorithm")
     ("schema-compression",
         po::value<std::string>(&schema_compression),
@@ -590,7 +594,7 @@ int mkdwarfs_main(int argc, sys_char** argv) {
   }
 
   if (!vm.count("compression")) {
-    compression = defaults.data_compression;
+    compression.push_back(std::string(defaults.data_compression));
   }
 
   if (!vm.count("schema-compression")) {
@@ -913,13 +917,14 @@ int mkdwarfs_main(int argc, sys_char** argv) {
 
   progress prog(std::move(updater), interval_ms);
 
-  block_compressor bc(compression);
+  block_compressor bc(compression.front()); // TODO
   block_compressor schema_bc(schema_compression);
   block_compressor metadata_bc(metadata_compression);
 
   auto min_memory_req = num_workers * (UINT64_C(1) << cfg.block_size_bits);
 
-  if (mem_limit < min_memory_req && compression != "null") {
+  // TODO:
+  if (mem_limit < min_memory_req /* && compression != "null" */) {
     LOG_WARN << "low memory limit (" << size_with_unit(mem_limit) << "), need "
              << size_with_unit(min_memory_req) << " to efficiently compress "
              << size_with_unit(UINT64_C(1) << cfg.block_size_bits)
@@ -951,6 +956,77 @@ int mkdwarfs_main(int argc, sys_char** argv) {
     os = std::make_unique<std::ostringstream>();
   }
 
+  // TODO: the whole re-writing thing will be a bit weird in combination
+  //       with categories; we'd likely require a "category"-section to be
+  //       present (which we'll also require for bit-identical mode)
+
+  if (!categorizer_list_str.empty()) {
+    std::vector<std::string> categorizer_list;
+    boost::split(categorizer_list, categorizer_list_str, boost::is_any_of(","));
+
+    options.inode.categorizer_mgr = std::make_shared<categorizer_manager>(lgr);
+
+    for (auto const& name : categorizer_list) {
+      options.inode.categorizer_mgr->add(catreg.create(lgr, name, vm));
+    }
+  }
+
+  category_parser cp(options.inode.categorizer_mgr);
+
+  try {
+    contextual_option_parser cop("--order", options.inode.fragment_order, cp,
+                                 order_parser);
+    cop.parse(order);
+    cop.dump(std::cerr);
+  } catch (std::exception const& e) {
+    LOG_ERROR << e.what();
+    return 1;
+  }
+
+  try {
+    categorized_option<size_t> max_lookback_opt;
+    contextual_option_parser cop("--max-lookback-blocks", max_lookback_opt, cp,
+                                 max_lookback_parser);
+    cop.parse(max_lookback_blocks);
+    cop.dump(std::cerr);
+  } catch (std::exception const& e) {
+    LOG_ERROR << e.what();
+    return 1;
+  }
+
+  try {
+    categorized_option<unsigned> window_size_opt;
+    contextual_option_parser cop("--window-size", window_size_opt, cp,
+                                 window_size_parser);
+    cop.parse(window_size);
+    cop.dump(std::cerr);
+  } catch (std::exception const& e) {
+    LOG_ERROR << e.what();
+    return 1;
+  }
+
+  try {
+    categorized_option<unsigned> window_step_opt;
+    contextual_option_parser cop("--window-step", window_step_opt, cp,
+                                 window_step_parser);
+    cop.parse(window_step);
+    cop.dump(std::cerr);
+  } catch (std::exception const& e) {
+    LOG_ERROR << e.what();
+    return 1;
+  }
+
+  try {
+    categorized_option<block_compressor> compression_opt;
+    contextual_option_parser cop("--compression", compression_opt, cp,
+                                 compressor_parser);
+    cop.parse(compression);
+    cop.dump(std::cerr);
+  } catch (std::exception const& e) {
+    LOG_ERROR << e.what();
+    return 1;
+  }
+
   filesystem_writer fsw(*os, lgr, wg_compress, prog, bc, schema_bc, metadata_bc,
                         fswopts, header_ifs.get());
 
@@ -962,61 +1038,6 @@ int mkdwarfs_main(int argc, sys_char** argv) {
                              fsw, rw_opts);
       wg_compress.wait();
     } else {
-      if (!categorizer_list_str.empty()) {
-        std::vector<std::string> categorizer_list;
-        boost::split(categorizer_list, categorizer_list_str,
-                     boost::is_any_of(","));
-
-        options.inode.categorizer_mgr =
-            std::make_shared<categorizer_manager>(lgr);
-
-        for (auto const& name : categorizer_list) {
-          options.inode.categorizer_mgr->add(catreg.create(lgr, name, vm));
-        }
-      }
-
-      category_parser cp(options.inode.categorizer_mgr);
-
-      try {
-        fragment_order_parser fop;
-        contextual_option_parser("--order", options.inode.fragment_order, cp,
-                                 fop)
-            .parse(order);
-      } catch (std::exception const& e) {
-        LOG_ERROR << e.what();
-        return 1;
-      }
-
-      try {
-        categorized_option<size_t> max_lookback_opt;
-        contextual_option_parser("--max-lookback-blocks", max_lookback_opt, cp,
-                                 max_lookback_parser)
-            .parse(max_lookback_blocks);
-      } catch (std::exception const& e) {
-        LOG_ERROR << e.what();
-        return 1;
-      }
-
-      try {
-        categorized_option<unsigned> window_size_opt;
-        contextual_option_parser("--window-size", window_size_opt, cp,
-                                 window_size_parser)
-            .parse(window_size);
-      } catch (std::exception const& e) {
-        LOG_ERROR << e.what();
-        return 1;
-      }
-
-      try {
-        categorized_option<unsigned> window_step_opt;
-        contextual_option_parser("--window-step", window_step_opt, cp,
-                                 window_step_parser)
-            .parse(window_step);
-      } catch (std::exception const& e) {
-        LOG_ERROR << e.what();
-        return 1;
-      }
-
       scanner s(lgr, wg_scanner, cfg, entry_factory::create(),
                 std::make_shared<os_access_generic>(), std::move(script),
                 options);