]> git.saurik.com Git - apt.git/blobdiff - methods/rred.cc
Merge remote-tracking branch 'mvo/feature/hash-stats' into debian/experimental
[apt.git] / methods / rred.cc
index 6c55880ca1726c782bd1f939c50952a7f083c0ba..cabb3c456ae923b66ba3bb87e3afcf4e378a3865 100644 (file)
-// Includes                                                                    /*{{{*/
+// Copyright (c) 2014 Anthony Towns
+//
+// This program 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 2 of the License, or
+// (at your option) any later version.
+
+#include <config.h>
+
 #include <apt-pkg/fileutl.h>
-#include <apt-pkg/mmap.h>
 #include <apt-pkg/error.h>
 #include <apt-pkg/acquire-method.h>
 #include <apt-pkg/strutl.h>
 #include <apt-pkg/hashes.h>
+#include <apt-pkg/configuration.h>
 
-#include <sys/stat.h>
-#include <sys/uio.h>
-#include <unistd.h>
-#include <utime.h>
+#include <stddef.h>
+#include <iostream>
+#include <string>
+#include <list>
+#include <vector>
+
+#include <assert.h>
 #include <stdio.h>
-#include <errno.h>
-#include <zlib.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <sys/time.h>
+
 #include <apti18n.h>
-                                                                               /*}}}*/
-/** \brief RredMethod - ed-style incremential patch method                     {{{
- *
- *  This method implements a patch functionality similar to "patch --ed" that is
- *  used by the "tiffany" incremental packages download stuff. It differs from
- *  "ed" insofar that it is way more restricted (and therefore secure).
- *  The currently supported ed commands are "<em>c</em>hange", "<em>a</em>dd" and
- *  "<em>d</em>elete" (diff doesn't output any other).
- *  Additionally the records must be reverse sorted by line number and
- *  may not overlap (diff *seems* to produce this kind of output).
- * */
-class RredMethod : public pkgAcqMethod {
-       bool Debug;
-       // the size of this doesn't really matter (except for performance)
-       const static int BUF_SIZE = 1024;
-       // the supported ed commands
-       enum Mode {MODE_CHANGED='c', MODE_DELETED='d', MODE_ADDED='a'};
-       // return values
-       enum State {ED_OK, ED_ORDERING, ED_PARSER, ED_FAILURE, MMAP_FAILED};
-
-       State applyFile(gzFile &ed_cmds, FILE *in_file, FILE *out_file,
-                    unsigned long &line, char *buffer, Hashes *hash) const;
-       void ignoreLineInFile(FILE *fin, char *buffer) const;
-       void ignoreLineInFile(gzFile &fin, char *buffer) const;
-       void copyLinesFromFileToFile(FILE *fin, FILE *fout, unsigned int lines,
-                                   Hashes *hash, char *buffer) const;
-       void copyLinesFromFileToFile(gzFile &fin, FILE *fout, unsigned int lines,
-                                   Hashes *hash, char *buffer) const;
-
-       State patchFile(FileFd &Patch, FileFd &From, FileFd &out_file, Hashes *hash) const;
-       State patchMMap(FileFd &Patch, FileFd &From, FileFd &out_file, Hashes *hash) const;
-
-protected:
-       // the methods main method
-       virtual bool Fetch(FetchItem *Itm);
-
-public:
-       RredMethod() : pkgAcqMethod("1.1",SingleInstance | SendConfig), Debug(false) {};
+
+#define BLOCK_SIZE (512*1024)
+
+class MemBlock {
+   char *start;
+   size_t size;
+   char *free;
+   struct MemBlock *next;
+
+   MemBlock(size_t size) : size(size), next(NULL)
+   {
+      free = start = new char[size];
+   }
+
+   size_t avail(void) { return size - (free - start); }
+
+   public:
+
+   MemBlock(void) {
+      free = start = new char[BLOCK_SIZE];
+      size = BLOCK_SIZE;
+      next = NULL;
+   }
+
+   ~MemBlock() {
+      delete [] start;
+      delete next;
+   }
+
+   void clear(void) {
+      free = start;
+      if (next)
+        next->clear();
+   }
+
+   char *add_easy(char *src, size_t len, char *last)
+   {
+      if (last) {
+        for (MemBlock *k = this; k; k = k->next) {
+           if (k->free == last) {
+              if (len <= k->avail()) {
+                 char *n = k->add(src, len);
+                 assert(last == n);
+                 if (last == n)
+                    return NULL;
+                 return n;
+              } else {
+                 break;
+              }
+           } else if (last >= start && last < free) {
+              break;
+           }
+        }
+      }
+      return add(src, len);
+   }
+
+   char *add(char *src, size_t len) {
+      if (len > avail()) {
+        if (!next) {
+           if (len > BLOCK_SIZE)  {
+              next = new MemBlock(len);
+           } else {
+              next = new MemBlock;
+           }
+        }
+        return next->add(src, len);
+      }
+      char *dst = free;
+      free += len;
+      memcpy(dst, src, len);
+      return dst;
+   }
 };
-                                                                               /*}}}*/
-/** \brief applyFile - in reverse order with a tail recursion                  {{{
- *
- *  As it is expected that the commands are in reversed order in the patch file
- *  we check in the first half if the command is valid, but doesn't execute it
- *  and move a step deeper. After reaching the end of the file we apply the
- *  patches in the correct order: last found command first.
- *
- *  \param ed_cmds patch file to apply
- *  \param in_file base file we want to patch
- *  \param out_file file to write the patched result to
- *  \param line of command operation
- *  \param buffer internal used read/write buffer
- *  \param hash the created file for correctness
- *  \return the success State of the ed command executor
- */
-RredMethod::State RredMethod::applyFile(gzFile &ed_cmds, FILE *in_file, FILE *out_file,
-                       unsigned long &line, char *buffer, Hashes *hash) const {
-       // get the current command and parse it
-       if (gzgets(ed_cmds, buffer, BUF_SIZE) == NULL) {
-               if (Debug == true)
-                       std::clog << "rred: encounter end of file - we can start patching now." << std::endl;
-               line = 0;
-               return ED_OK;
-       }
-
-       // parse in the effected linenumbers
-       char* idx;
-       errno=0;
-       unsigned long const startline = strtol(buffer, &idx, 10);
-       if (errno == ERANGE || errno == EINVAL) {
-               _error->Errno("rred", "startline is an invalid number");
-               return ED_PARSER;
-       }
-       if (startline > line) {
-               _error->Error("rred: The start line (%lu) of the next command is higher than the last line (%lu). This is not allowed.", startline, line);
-               return ED_ORDERING;
-       }
-       unsigned long stopline;
-       if (*idx == ',') {
-               idx++;
-               errno=0;
-               stopline = strtol(idx, &idx, 10);
-               if (errno == ERANGE || errno == EINVAL) {
-                       _error->Errno("rred", "stopline is an invalid number");
-                       return ED_PARSER;
-               }
-       }
-       else {
-               stopline = startline;
-       }
-       line = startline;
-
-       // which command to execute on this line(s)?
-       switch (*idx) {
-               case MODE_CHANGED:
-                       if (Debug == true)
-                               std::clog << "Change from line " << startline << " to " << stopline << std::endl;
-                       break;
-               case MODE_ADDED:
-                       if (Debug == true)
-                               std::clog << "Insert after line " << startline << std::endl;
-                       break;
-               case MODE_DELETED:
-                       if (Debug == true)
-                               std::clog << "Delete from line " << startline << " to " << stopline << std::endl;
-                       break;
-               default:
-                       _error->Error("rred: Unknown ed command '%c'. Abort.", *idx);
-                       return ED_PARSER;
-       }
-       unsigned char mode = *idx;
-
-       // save the current position
-       unsigned const long pos = gztell(ed_cmds);
-
-       // if this is add or change then go to the next full stop
-       unsigned int data_length = 0;
-       if (mode == MODE_CHANGED || mode == MODE_ADDED) {
-               do {
-                       ignoreLineInFile(ed_cmds, buffer);
-                       data_length++;
-               }
-               while (strncmp(buffer, ".", 1) != 0);
-               data_length--; // the dot should not be copied
-       }
-
-       // do the recursive call - the last command is the one we need to execute at first
-       const State child = applyFile(ed_cmds, in_file, out_file, line, buffer, hash);
-       if (child != ED_OK) {
-               return child;
-       }
-
-       // change and delete are working on "line" - add is done after "line"
-       if (mode != MODE_ADDED)
-               line++;
-
-       // first wind to the current position and copy over all unchanged lines
-       if (line < startline) {
-               copyLinesFromFileToFile(in_file, out_file, (startline - line), hash, buffer);
-               line = startline;
-       }
-
-       if (mode != MODE_ADDED)
-               line--;
-
-       // include data from ed script
-       if (mode == MODE_CHANGED || mode == MODE_ADDED) {
-               gzseek(ed_cmds, pos, SEEK_SET);
-               copyLinesFromFileToFile(ed_cmds, out_file, data_length, hash, buffer);
-       }
-
-       // ignore the corresponding number of lines from input
-       if (mode == MODE_CHANGED || mode == MODE_DELETED) {
-               while (line < stopline) {
-                       ignoreLineInFile(in_file, buffer);
-                       line++;
-               }
-       }
-       return ED_OK;
-}
-                                                                               /*}}}*/
-void RredMethod::copyLinesFromFileToFile(FILE *fin, FILE *fout, unsigned int lines,/*{{{*/
-                                       Hashes *hash, char *buffer) const {
-       while (0 < lines--) {
-               do {
-                       fgets(buffer, BUF_SIZE, fin);
-                       size_t const written = fwrite(buffer, 1, strlen(buffer), fout);
-                       hash->Add((unsigned char*)buffer, written);
-               } while (strlen(buffer) == (BUF_SIZE - 1) &&
-                      buffer[BUF_SIZE - 2] != '\n');
-       }
-}
-                                                                               /*}}}*/
-void RredMethod::copyLinesFromFileToFile(gzFile &fin, FILE *fout, unsigned int lines,/*{{{*/
-                                       Hashes *hash, char *buffer) const {
-       while (0 < lines--) {
-               do {
-                       gzgets(fin, buffer, BUF_SIZE);
-                       size_t const written = fwrite(buffer, 1, strlen(buffer), fout);
-                       hash->Add((unsigned char*)buffer, written);
-               } while (strlen(buffer) == (BUF_SIZE - 1) &&
-                      buffer[BUF_SIZE - 2] != '\n');
-       }
-}
-                                                                               /*}}}*/
-void RredMethod::ignoreLineInFile(FILE *fin, char *buffer) const {             /*{{{*/
-       fgets(buffer, BUF_SIZE, fin);
-       while (strlen(buffer) == (BUF_SIZE - 1) &&
-              buffer[BUF_SIZE - 2] != '\n') {
-               fgets(buffer, BUF_SIZE, fin);
-               buffer[0] = ' ';
-       }
-}
-                                                                               /*}}}*/
-void RredMethod::ignoreLineInFile(gzFile &fin, char *buffer) const {           /*{{{*/
-       gzgets(fin, buffer, BUF_SIZE);
-       while (strlen(buffer) == (BUF_SIZE - 1) &&
-              buffer[BUF_SIZE - 2] != '\n') {
-               gzgets(fin, buffer, BUF_SIZE);
-               buffer[0] = ' ';
-       }
-}
-                                                                               /*}}}*/
-RredMethod::State RredMethod::patchFile(FileFd &Patch, FileFd &From,           /*{{{*/
-                                       FileFd &out_file, Hashes *hash) const {
-   char buffer[BUF_SIZE];
-   FILE* fFrom = fdopen(From.Fd(), "r");
-   gzFile fPatch = Patch.gzFd();
-   FILE* fTo = fdopen(out_file.Fd(), "w");
-
-   /* we do a tail recursion to read the commands in the right order */
-   unsigned long line = -1; // assign highest possible value
-   State const result = applyFile(fPatch, fFrom, fTo, line, buffer, hash);
-   
-   /* read the rest from infile */
-   if (result == ED_OK) {
-      while (fgets(buffer, BUF_SIZE, fFrom) != NULL) {
-         size_t const written = fwrite(buffer, 1, strlen(buffer), fTo);
-         hash->Add((unsigned char*)buffer, written);
+
+struct Change {
+   /* Ordering:
+    *
+    *   1. write out <offset> lines unchanged
+    *   2. skip <del_cnt> lines from source
+    *   3. write out <add_cnt> lines (<add>/<add_len>)
+    */
+   size_t offset;
+   size_t del_cnt;
+   size_t add_cnt; /* lines */
+   size_t add_len; /* bytes */
+   char *add;
+
+   Change(int off)
+   {
+      offset = off;
+      del_cnt = add_cnt = add_len = 0;
+      add = NULL;
+   }
+
+   /* actually, don't write <lines> lines from <add> */
+   void skip_lines(size_t lines)
+   {
+      while (lines > 0) {
+        char *s = (char*) memchr(add, '\n', add_len);
+        assert(s != NULL);
+        s++;
+        add_len -= (s - add);
+        add_cnt--;
+        lines--;
+        if (add_len == 0) {
+           add = NULL;
+           assert(add_cnt == 0);
+           assert(lines == 0);
+        } else {
+           add = s;
+           assert(add_cnt > 0);
+        }
       }
-      fflush(fTo);
    }
-   return result;
-}
-                                                                               /*}}}*/
-/* struct EdCommand                                                            {{{*/
-#ifdef _POSIX_MAPPED_FILES
-struct EdCommand {
-  size_t data_start;
-  size_t data_end;
-  size_t data_lines;
-  size_t first_line;
-  size_t last_line;
-  char type;
 };
-#define IOV_COUNT 1024 /* Don't really want IOV_MAX since it can be arbitrarily large */
-#endif
-                                                                               /*}}}*/
-RredMethod::State RredMethod::patchMMap(FileFd &Patch, FileFd &From,           /*{{{*/
-                                       FileFd &out_file, Hashes *hash) const {
-#ifdef _POSIX_MAPPED_FILES
-       MMap ed_cmds(MMap::ReadOnly);
-       if (Patch.gzFd() != NULL) {
-               unsigned long mapSize = Patch.Size();
-               DynamicMMap* dyn = new DynamicMMap(0, mapSize, 0);
-               if (dyn->validData() == false) {
-                       delete dyn;
-                       return MMAP_FAILED;
-               }
-               dyn->AddSize(mapSize);
-               gzread(Patch.gzFd(), dyn->Data(), mapSize);
-               ed_cmds = *dyn;
-       } else
-               ed_cmds = MMap(Patch, MMap::ReadOnly);
-
-       MMap in_file(From, MMap::ReadOnly);
-
-       if (ed_cmds.Size() == 0 || in_file.Size() == 0)
-               return MMAP_FAILED;
-
-       EdCommand* commands = 0;
-       size_t command_count = 0;
-       size_t command_alloc = 0;
-
-       const char* begin = (char*) ed_cmds.Data();
-       const char* end = begin;
-       const char* ed_end = (char*) ed_cmds.Data() + ed_cmds.Size();
-
-       const char* input = (char*) in_file.Data();
-       const char* input_end = (char*) in_file.Data() + in_file.Size();
-
-       size_t i;
-
-       /* 1. Parse entire script.  It is executed in reverse order, so we cather it
-        *    in the `commands' buffer first
-        */
-
-       for(;;) {
-               EdCommand cmd;
-               cmd.data_start = 0;
-               cmd.data_end = 0;
-
-               while(begin != ed_end && *begin == '\n')
-                       ++begin;
-               while(end != ed_end && *end != '\n')
-                       ++end;
-               if(end == ed_end && begin == end)
-                       break;
-
-               /* Determine command range */
-               const char* tmp = begin;
-
-               for(;;) {
-                       /* atoll is safe despite lacking NUL-termination; we know there's an
-                        * alphabetic character at end[-1]
-                        */
-                       if(tmp == end) {
-                               cmd.first_line = atol(begin);
-                               cmd.last_line = cmd.first_line;
-                               break;
-                       }
-                       if(*tmp == ',') {
-                               cmd.first_line = atol(begin);
-                               cmd.last_line = atol(tmp + 1);
-                               break;
-                       }
-                       ++tmp;
-               }
-
-               // which command to execute on this line(s)?
-               switch (end[-1]) {
-                       case MODE_CHANGED:
-                               if (Debug == true)
-                                       std::clog << "Change from line " << cmd.first_line << " to " << cmd.last_line << std::endl;
-                               break;
-                       case MODE_ADDED:
-                               if (Debug == true)
-                                       std::clog << "Insert after line " << cmd.first_line << std::endl;
-                               break;
-                       case MODE_DELETED:
-                               if (Debug == true)
-                                       std::clog << "Delete from line " << cmd.first_line << " to " << cmd.last_line << std::endl;
-                               break;
-                       default:
-                               _error->Error("rred: Unknown ed command '%c'. Abort.", end[-1]);
-                               free(commands);
-                               return ED_PARSER;
-               }
-               cmd.type = end[-1];
-
-               /* Determine the size of the inserted text, so we don't have to scan this
-                * text again later.
-                */
-               begin = end + 1;
-               end = begin;
-               cmd.data_lines = 0;
-
-               if(cmd.type == MODE_ADDED || cmd.type == MODE_CHANGED) {
-                       cmd.data_start = begin - (char*) ed_cmds.Data();
-                       while(end != ed_end) {
-                               if(*end == '\n') {
-                                       if(end[-1] == '.' && end[-2] == '\n')
-                                               break;
-                                       ++cmd.data_lines;
-                               }
-                               ++end;
-                       }
-                       cmd.data_end = end - (char*) ed_cmds.Data() - 1;
-                       begin = end + 1;
-                       end = begin;
-               }
-               if(command_count == command_alloc) {
-                       command_alloc = (command_alloc + 64) * 3 / 2;
-                       commands = (EdCommand*) realloc(commands, command_alloc * sizeof(EdCommand));
-               }
-               commands[command_count++] = cmd;
-       }
-
-       struct iovec* iov = new struct iovec[IOV_COUNT];
-       size_t iov_size = 0;
-
-       size_t amount, remaining;
-       size_t line = 1;
-       EdCommand* cmd;
-
-       /* 2. Execute script.  We gather writes in a `struct iov' array, and flush
-        *    using writev to minimize the number of system calls.  Data is read
-        *    directly from the memory mappings of the input file and the script.
-        */
-
-       for(i = command_count; i-- > 0; ) {
-               cmd = &commands[i];
-               if(cmd->type == MODE_ADDED)
-                       amount = cmd->first_line + 1;
-               else
-                       amount = cmd->first_line;
-
-               if(line < amount) {
-                       begin = input;
-                       while(line != amount) {
-                               input = (const char*) memchr(input, '\n', input_end - input);
-                               if(!input)
-                                       break;
-                               ++line;
-                               ++input;
-                       }
-
-                       iov[iov_size].iov_base = (void*) begin;
-                       iov[iov_size].iov_len = input - begin;
-                       hash->Add((const unsigned char*) begin, input - begin);
-
-                       if(++iov_size == IOV_COUNT) {
-                               writev(out_file.Fd(), iov, IOV_COUNT);
-                               iov_size = 0;
-                       }
-               }
-
-               if(cmd->type == MODE_DELETED || cmd->type == MODE_CHANGED) {
-                       remaining = (cmd->last_line - cmd->first_line) + 1;
-                       line += remaining;
-                       while(remaining) {
-                               input = (const char*) memchr(input, '\n', input_end - input);
-                               if(!input)
-                                       break;
-                               --remaining;
-                               ++input;
-                       }
-               }
-
-               if(cmd->type == MODE_CHANGED || cmd->type == MODE_ADDED) {
-                       if(cmd->data_end != cmd->data_start) {
-                               iov[iov_size].iov_base = (void*) ((char*)ed_cmds.Data() + cmd->data_start);
-                               iov[iov_size].iov_len = cmd->data_end - cmd->data_start;
-                               hash->Add((const unsigned char*) ((char*)ed_cmds.Data() + cmd->data_start),
-                               iov[iov_size].iov_len);
-
-                               if(++iov_size == IOV_COUNT) {
-                                       writev(out_file.Fd(), iov, IOV_COUNT);
-                                       iov_size = 0;
-                               }
-                       }
-               }
-       }
-
-       if(input != input_end) {
-               iov[iov_size].iov_base = (void*) input;
-               iov[iov_size].iov_len = input_end - input;
-               hash->Add((const unsigned char*) input, input_end - input);
-               ++iov_size;
-       }
-
-       if(iov_size) {
-               writev(out_file.Fd(), iov, iov_size);
-               iov_size = 0;
-       }
-
-       for(i = 0; i < iov_size; i += IOV_COUNT) {
-               if(iov_size - i < IOV_COUNT)
-                       writev(out_file.Fd(), iov + i, iov_size - i);
-               else
-                       writev(out_file.Fd(), iov + i, IOV_COUNT);
-       }
-
-       delete [] iov;
-       free(commands);
-
-       return ED_OK;
+
+class FileChanges {
+   std::list<struct Change> changes;
+   std::list<struct Change>::iterator where;
+   size_t pos; // line number is as far left of iterator as possible
+
+   bool pos_is_okay(void)
+   {
+#ifdef POSDEBUG
+      size_t cpos = 0;
+      std::list<struct Change>::iterator x;
+      for (x = changes.begin(); x != where; ++x) {
+        assert(x != changes.end());
+        cpos += x->offset + x->add_cnt;
+      }
+      return cpos == pos;
 #else
-       return MMAP_FAILED;
+      return true;
 #endif
-}
-                                                                               /*}}}*/
-bool RredMethod::Fetch(FetchItem *Itm)                                         /*{{{*/
-{
-   Debug = _config->FindB("Debug::pkgAcquire::RRed", false);
-   URI Get = Itm->Uri;
-   string Path = Get.Host + Get.Path; // To account for relative paths
-
-   FetchResult Res;
-   Res.Filename = Itm->DestFile;
-   if (Itm->Uri.empty() == true) {
-      Path = Itm->DestFile;
-      Itm->DestFile.append(".result");
-   } else
-      URIStart(Res);
-
-   if (Debug == true) 
-      std::clog << "Patching " << Path << " with " << Path 
-         << ".ed and putting result into " << Itm->DestFile << std::endl;
-   // Open the source and destination files (the d'tor of FileFd will do 
-   // the cleanup/closing of the fds)
-   FileFd From(Path,FileFd::ReadOnly);
-   FileFd Patch(Path+".ed",FileFd::ReadOnlyGzip);
-   FileFd To(Itm->DestFile,FileFd::WriteAtomic);   
-   To.EraseOnFailure();
-   if (_error->PendingError() == true)
-      return false;
-   
-   Hashes Hash;
-   // now do the actual patching
-   State const result = patchMMap(Patch, From, To, &Hash);
-   if (result == MMAP_FAILED) {
-      // retry with patchFile
-      Patch.Seek(0);
-      From.Seek(0);
-      To.Open(Itm->DestFile,FileFd::WriteAtomic);
-      if (_error->PendingError() == true)
-         return false;
-      if (patchFile(Patch, From, To, &Hash) != ED_OK) {
-        return _error->WarningE("rred", _("Could not patch %s with mmap and with file operation usage - the patch seems to be corrupt."), Path.c_str());
-      } else if (Debug == true) {
-        std::clog << "rred: finished file patching of " << Path  << " after mmap failed." << std::endl;
+   }
+
+   public:
+   FileChanges() {
+      where = changes.end();
+      pos = 0;
+   }
+
+   std::list<struct Change>::iterator begin(void) { return changes.begin(); }
+   std::list<struct Change>::iterator end(void) { return changes.end(); }
+
+   std::list<struct Change>::reverse_iterator rbegin(void) { return changes.rbegin(); }
+   std::list<struct Change>::reverse_iterator rend(void) { return changes.rend(); }
+
+   void add_change(Change c) {
+      assert(pos_is_okay());
+      go_to_change_for(c.offset);
+      assert(pos + where->offset == c.offset);
+      if (c.del_cnt > 0)
+        delete_lines(c.del_cnt);
+      assert(pos + where->offset == c.offset);
+      if (c.add_len > 0) {
+        assert(pos_is_okay());
+        if (where->add_len > 0)
+           new_change();
+        assert(where->add_len == 0 && where->add_cnt == 0);
+
+        where->add_len = c.add_len;
+        where->add_cnt = c.add_cnt;
+        where->add = c.add;
       }
-   } else if (result != ED_OK) {
-      return _error->Errno("rred", _("Could not patch %s with mmap (but no mmap specific fail) - the patch seems to be corrupt."), Path.c_str());
-   } else if (Debug == true) {
-      std::clog << "rred: finished mmap patching of " << Path << std::endl;
+      assert(pos_is_okay());
+      merge();
+      assert(pos_is_okay());
    }
 
-   // write out the result
-   From.Close();
-   Patch.Close();
-   To.Close();
-
-   /* Transfer the modification times from the patch file
-      to be able to see in which state the file should be
-      and use the access time from the "old" file */
-   struct stat BufBase, BufPatch;
-   if (stat(Path.c_str(),&BufBase) != 0 ||
-       stat(string(Path+".ed").c_str(),&BufPatch) != 0)
-      return _error->Errno("stat",_("Failed to stat"));
-
-   struct utimbuf TimeBuf;
-   TimeBuf.actime = BufBase.st_atime;
-   TimeBuf.modtime = BufPatch.st_mtime;
-   if (utime(Itm->DestFile.c_str(),&TimeBuf) != 0)
-      return _error->Errno("utime",_("Failed to set modification time"));
-
-   if (stat(Itm->DestFile.c_str(),&BufBase) != 0)
-      return _error->Errno("stat",_("Failed to stat"));
-
-   // return done
-   Res.LastModified = BufBase.st_mtime;
-   Res.Size = BufBase.st_size;
-   Res.TakeHashes(Hash);
-   URIDone(Res);
-
-   return true;
-}
-                                                                               /*}}}*/
-/** \brief Wrapper class for testing rred */                                   /*{{{*/
-class TestRredMethod : public RredMethod {
-public:
-       /** \brief Run rred in debug test mode
-        *
-        *  This method can be used to run the rred method outside
-        *  of the "normal" acquire environment for easier testing.
-        *
-        *  \param base basename of all files involved in this rred test
-        */
-       bool Run(char const *base) {
-               _config->CndSet("Debug::pkgAcquire::RRed", "true");
-               FetchItem *test = new FetchItem;
-               test->DestFile = base;
-               return Fetch(test);
-       }
+   private:
+   void merge(void)
+   {
+      while (where->offset == 0 && where != changes.begin()) {
+        left();
+      }
+      std::list<struct Change>::iterator next = where;
+      ++next;
+
+      while (next != changes.end() && next->offset == 0) {
+        where->del_cnt += next->del_cnt;
+        next->del_cnt = 0;
+        if (next->add == NULL) {
+           next = changes.erase(next);
+        } else if (where->add == NULL) {
+           where->add = next->add;
+           where->add_len = next->add_len;
+           where->add_cnt = next->add_cnt;
+           next = changes.erase(next);
+        } else {
+           ++next;
+        }
+      }
+   }
+
+   void go_to_change_for(size_t line)
+   {
+      while(where != changes.end()) {
+        if (line < pos) {
+           left();
+           continue;
+        }
+        if (pos + where->offset + where->add_cnt <= line) {
+           right();
+           continue;
+        }
+        // line is somewhere in this slot
+        if (line < pos + where->offset) {
+           break;
+        } else if (line == pos + where->offset) {
+           return;
+        } else {
+           split(line - pos);
+           right();
+           return;
+        }
+      }
+      /* it goes before this patch */
+      insert(line-pos);
+   }
+
+   void new_change(void) { insert(where->offset); }
+
+   void insert(size_t offset)
+   {
+      assert(pos_is_okay());
+      assert(where == changes.end() || offset <= where->offset);
+      if (where != changes.end())
+        where->offset -= offset;
+      changes.insert(where, Change(offset));
+      --where;
+      assert(pos_is_okay());
+   }
+
+   void split(size_t offset)
+   {
+      assert(pos_is_okay());
+
+      assert(where->offset < offset);
+      assert(offset < where->offset + where->add_cnt);
+
+      size_t keep_lines = offset - where->offset;
+
+      Change before(*where);
+
+      where->del_cnt = 0;
+      where->offset = 0;
+      where->skip_lines(keep_lines);
+
+      before.add_cnt = keep_lines;
+      before.add_len -= where->add_len;
+
+      changes.insert(where, before);
+      --where;
+      assert(pos_is_okay());
+   }
+
+   void delete_lines(size_t cnt)
+   {
+      std::list<struct Change>::iterator x = where;
+      assert(pos_is_okay());
+      while (cnt > 0)
+      {
+        size_t del;
+        del = x->add_cnt;
+        if (del > cnt)
+           del = cnt;
+        x->skip_lines(del);
+        cnt -= del;
+
+        ++x;
+        if (x == changes.end()) {
+           del = cnt;
+        } else {
+           del = x->offset;
+           if (del > cnt)
+              del = cnt;
+           x->offset -= del;
+        }
+        where->del_cnt += del;
+        cnt -= del;
+      }
+      assert(pos_is_okay());
+   }
+
+   void left(void) {
+      assert(pos_is_okay());
+      --where;
+      pos -= where->offset + where->add_cnt;
+      assert(pos_is_okay());
+   }
+
+   void right(void) {
+      assert(pos_is_okay());
+      pos += where->offset + where->add_cnt;
+      ++where;
+      assert(pos_is_okay());
+   }
 };
-                                                                               /*}}}*/
-/** \brief Starter for the rred method (or its test method)                    {{{
- *
- *  Used without parameters is the normal behavior for methods for
- *  the APT acquire system. While this works great for the acquire system
- *  it is very hard to test the method and therefore the method also
- *  accepts one parameter which will switch it directly to debug test mode:
- *  The test mode expects that if "Testfile" is given as parameter
- *  the file "Testfile" should be ed-style patched with "Testfile.ed"
- *  and will write the result to "Testfile.result".
- */
-int main(int argc, char *argv[]) {
-       if (argc <= 1) {
-               RredMethod Mth;
-               return Mth.Run();
-       } else {
-               TestRredMethod Mth;
-               bool result = Mth.Run(argv[1]);
-               _error->DumpErrors();
-               return result;
-       }
+
+class Patch {
+   FileChanges filechanges;
+   MemBlock add_text;
+
+   static bool retry_fwrite(char *b, size_t l, FILE *f, Hashes *hash)
+   {
+      size_t r = 1;
+      while (r > 0 && l > 0)
+      {
+         r = fwrite(b, 1, l, f);
+        if (hash)
+           hash->Add((unsigned char*)b, r);
+        l -= r;
+        b += r;
+      }
+      return l == 0;
+   }
+
+   static void dump_rest(FILE *o, FILE *i, Hashes *hash)
+   {
+      char buffer[BLOCK_SIZE];
+      size_t l;
+      while (0 < (l = fread(buffer, 1, sizeof(buffer), i))) {
+        if (!retry_fwrite(buffer, l, o, hash))
+           break;
+      }
+   }
+
+   static void dump_lines(FILE *o, FILE *i, size_t n, Hashes *hash)
+   {
+      char buffer[BLOCK_SIZE];
+      while (n > 0) {
+        if (fgets(buffer, sizeof(buffer), i) == 0)
+           buffer[0] = '\0';
+        size_t const l = strlen(buffer);
+        if (l == 0 || buffer[l-1] == '\n')
+           n--;
+        retry_fwrite(buffer, l, o, hash);
+      }
+   }
+
+   static void skip_lines(FILE *i, int n)
+   {
+      char buffer[BLOCK_SIZE];
+      while (n > 0) {
+        if (fgets(buffer, sizeof(buffer), i) == 0)
+           buffer[0] = '\0';
+        size_t const l = strlen(buffer);
+        if (l == 0 || buffer[l-1] == '\n')
+           n--;
+      }
+   }
+
+   static void dump_mem(FILE *o, char *p, size_t s, Hashes *hash) {
+      retry_fwrite(p, s, o, hash);
+   }
+
+   public:
+
+   void read_diff(FileFd &f)
+   {
+      char buffer[BLOCK_SIZE];
+      bool cmdwanted = true;
+
+      Change ch(0);
+      while(f.ReadLine(buffer, sizeof(buffer)))
+      {
+        if (cmdwanted) {
+           char *m, *c;
+           size_t s, e;
+           s = strtol(buffer, &m, 10);
+           if (m == buffer) {
+              s = e = ch.offset + ch.add_cnt;
+              c = buffer;
+           } else if (*m == ',') {
+              m++;
+              e = strtol(m, &c, 10);
+           } else {
+              e = s;
+              c = m;
+           }
+           switch(*c) {
+              case 'a':
+                 cmdwanted = false;
+                 ch.add = NULL;
+                 ch.add_cnt = 0;
+                 ch.add_len = 0;
+                 ch.offset = s;
+                 ch.del_cnt = 0;
+                 break;
+              case 'c':
+                 cmdwanted = false;
+                 ch.add = NULL;
+                 ch.add_cnt = 0;
+                 ch.add_len = 0;
+                 ch.offset = s - 1;
+                 ch.del_cnt = e - s + 1;
+                 break;
+              case 'd':
+                 ch.offset = s - 1;
+                 ch.del_cnt = e - s + 1;
+                 ch.add = NULL;
+                 ch.add_cnt = 0;
+                 ch.add_len = 0;
+                 filechanges.add_change(ch);
+                 break;
+           }
+        } else { /* !cmdwanted */
+           if (buffer[0] == '.' && buffer[1] == '\n') {
+              cmdwanted = true;
+              filechanges.add_change(ch);
+           } else {
+              char *last = NULL;
+              char *add;
+              size_t l;
+              if (ch.add)
+                 last = ch.add + ch.add_len;
+              l = strlen(buffer);
+              add = add_text.add_easy(buffer, l, last);
+              if (!add) {
+                 ch.add_len += l;
+                 ch.add_cnt++;
+              } else {
+                 if (ch.add) {
+                    filechanges.add_change(ch);
+                    ch.del_cnt = 0;
+                 }
+                 ch.offset += ch.add_cnt;
+                 ch.add = add;
+                 ch.add_len = l;
+                 ch.add_cnt = 1;
+              }
+           }
+        }
+      }
+   }
+
+   void write_diff(FILE *f)
+   {
+      unsigned long long line = 0;
+      std::list<struct Change>::reverse_iterator ch;
+      for (ch = filechanges.rbegin(); ch != filechanges.rend(); ++ch) {
+        line += ch->offset + ch->del_cnt;
+      }
+
+      for (ch = filechanges.rbegin(); ch != filechanges.rend(); ++ch) {
+        std::list<struct Change>::reverse_iterator mg_i, mg_e = ch;
+        while (ch->del_cnt == 0 && ch->offset == 0)
+           ++ch;
+        line -= ch->del_cnt;
+        if (ch->add_cnt > 0) {
+           if (ch->del_cnt == 0) {
+              fprintf(f, "%llua\n", line);
+           } else if (ch->del_cnt == 1) {
+              fprintf(f, "%lluc\n", line+1);
+           } else {
+              fprintf(f, "%llu,%lluc\n", line+1, line+ch->del_cnt);
+           }
+
+           mg_i = ch;
+           do {
+              dump_mem(f, mg_i->add, mg_i->add_len, NULL);
+           } while (mg_i-- != mg_e);
+
+           fprintf(f, ".\n");
+        } else if (ch->del_cnt == 1) {
+           fprintf(f, "%llud\n", line+1);
+        } else if (ch->del_cnt > 1) {
+           fprintf(f, "%llu,%llud\n", line+1, line+ch->del_cnt);
+        }
+        line -= ch->offset;
+      }
+   }
+
+   void apply_against_file(FILE *out, FILE *in, Hashes *hash = NULL)
+   {
+      std::list<struct Change>::iterator ch;
+      for (ch = filechanges.begin(); ch != filechanges.end(); ++ch) {
+        dump_lines(out, in, ch->offset, hash);
+        skip_lines(in, ch->del_cnt);
+        dump_mem(out, ch->add, ch->add_len, hash);
+      }
+      dump_rest(out, in, hash);
+   }
+};
+
+class RredMethod : public pkgAcqMethod {
+   private:
+      bool Debug;
+
+   protected:
+      virtual bool Fetch(FetchItem *Itm) {
+        Debug = _config->FindB("Debug::pkgAcquire::RRed", false);
+        URI Get = Itm->Uri;
+        std::string Path = Get.Host + Get.Path; // rred:/path - no host
+
+        FetchResult Res;
+        Res.Filename = Itm->DestFile;
+        if (Itm->Uri.empty())
+        {
+           Path = Itm->DestFile;
+           Itm->DestFile.append(".result");
+        } else
+           URIStart(Res);
+
+        std::vector<std::string> patchpaths;
+        Patch patch;
+
+        if (FileExists(Path + ".ed") == true)
+           patchpaths.push_back(Path + ".ed");
+        else
+        {
+           _error->PushToStack();
+           std::vector<std::string> patches = GetListOfFilesInDir(flNotFile(Path), "gz", true, false);
+           _error->RevertToStack();
+
+           std::string const baseName = Path + ".ed.";
+           for (std::vector<std::string>::const_iterator p = patches.begin();
+                 p != patches.end(); ++p)
+              if (p->compare(0, baseName.length(), baseName) == 0)
+                 patchpaths.push_back(*p);
+        }
+
+        std::string patch_name;
+        for (std::vector<std::string>::iterator I = patchpaths.begin();
+              I != patchpaths.end();
+              ++I)
+        {
+           patch_name = *I;
+           if (Debug == true)
+              std::clog << "Patching " << Path << " with " << patch_name
+                 << std::endl;
+
+           FileFd p;
+           // all patches are compressed, even if the name doesn't reflect it
+           if (p.Open(patch_name, FileFd::ReadOnly, FileFd::Gzip) == false) {
+              std::cerr << "Could not open patch file " << patch_name << std::endl;
+              _error->DumpErrors(std::cerr);
+              abort();
+           }
+           patch.read_diff(p);
+           p.Close();
+        }
+
+        if (Debug == true)
+           std::clog << "Applying patches against " << Path
+              << " and writing results to " << Itm->DestFile
+              << std::endl;
+
+        FILE *inp = fopen(Path.c_str(), "r");
+        FILE *out = fopen(Itm->DestFile.c_str(), "w");
+
+        Hashes hash;
+
+        patch.apply_against_file(out, inp, &hash);
+
+        fclose(out);
+        fclose(inp);
+
+        if (Debug == true) {
+           std::clog << "rred: finished file patching of " << Path  << "." << std::endl;
+        }
+
+        struct stat bufbase, bufpatch;
+        if (stat(Path.c_str(), &bufbase) != 0 ||
+              stat(patch_name.c_str(), &bufpatch) != 0)
+           return _error->Errno("stat", _("Failed to stat"));
+
+        struct timeval times[2];
+        times[0].tv_sec = bufbase.st_atime;
+        times[1].tv_sec = bufpatch.st_mtime;
+        times[0].tv_usec = times[1].tv_usec = 0;
+        if (utimes(Itm->DestFile.c_str(), times) != 0)
+           return _error->Errno("utimes",_("Failed to set modification time"));
+
+        if (stat(Itm->DestFile.c_str(), &bufbase) != 0)
+           return _error->Errno("stat", _("Failed to stat"));
+
+        Res.LastModified = bufbase.st_mtime;
+        Res.Size = bufbase.st_size;
+        Res.TakeHashes(hash);
+        URIDone(Res);
+
+        return true;
+      }
+
+   public:
+      RredMethod() : pkgAcqMethod("2.0",SingleInstance | SendConfig), Debug(false) {}
+};
+
+int main(int argc, char **argv)
+{
+   int i;
+   bool just_diff = true;
+   Patch patch;
+
+   if (argc <= 1) {
+      RredMethod Mth;
+      return Mth.Run();
+   }
+
+   if (argc > 1 && strcmp(argv[1], "-f") == 0) {
+      just_diff = false;
+      i = 2;
+   } else {
+      i = 1;
+   }
+
+   for (; i < argc; i++) {
+      FileFd p;
+      if (p.Open(argv[i], FileFd::ReadOnly) == false) {
+        _error->DumpErrors(std::cerr);
+        exit(1);
+      }
+      patch.read_diff(p);
+   }
+
+   if (just_diff) {
+      patch.write_diff(stdout);
+   } else {
+      FILE *out, *inp;
+      out = stdout;
+      inp = stdin;
+
+      patch.apply_against_file(out, inp);
+   }
+   return 0;
 }
-                                                                               /*}}}*/