path: root/src/patch.c

/*
** Copyright (c) 2021 D. Richard Hipp
**
** This program is free software; you can redistribute it and/or
** modify it under the terms of the Simplified BSD License (also
** known as the "2-Clause License" or "FreeBSD License".)
**
** This program 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.
**
** Author contact information:
**   drh@hwaci.com
**   http://www.hwaci.com/drh/
**
*******************************************************************************
**
** This file contains code used to implement the "patch" command
*/
#include "config.h"
#include "patch.h"
#include <assert.h>

/*
** Try to compute the name of the computer on which this process
** is running.
*/
char *fossil_hostname(void){
  FILE *in;
  char zBuf[200];
  in = popen("hostname","r");
  if( in ){
    int n = fread(zBuf, 1, sizeof(zBuf)-1, in);
    while( n>0 && fossil_isspace(zBuf[n-1]) ){ n--; }
    if( n<0 ) n = 0;
    zBuf[n] = 0;
    pclose(in);
    return fossil_strdup(zBuf);
  }
  return 0;
}

/*
** Flags passed from the main patch_cmd() routine into subfunctions used
** to implement the various subcommands.
*/
#define PATCH_DRYRUN   0x0001
#define PATCH_VERBOSE  0x0002
#define PATCH_FORCE    0x0004
#define PATCH_RETRY    0x0008     /* Second attempt */

/*
** Implementation of the "readfile(X)" SQL function.  The entire content
** of the check-out file named X is read and returned as a BLOB.
*/
static void readfileFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  const char *zName;
  Blob x;
  sqlite3_int64 sz;
  (void)(argc);  /* Unused parameter */
  zName = (const char*)sqlite3_value_text(argv[0]);
  if( zName==0 || (zName[0]=='-' && zName[1]==0) ) return;
  sz = blob_read_from_file(&x, zName, RepoFILE);
  sqlite3_result_blob64(context, x.aData, sz, SQLITE_TRANSIENT);
  blob_reset(&x);
}

/*
** mkdelta(X,Y)
**
** X is an numeric artifact id.  Y is a filename.
**
** Compute a compressed delta that carries X into Y.  Or return
** and zero-length blob if X is equal to Y.
*/
static void mkdeltaFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  const char *zFile;
  Blob x, y, out;
  int rid;
  int nOut;
  sqlite3_int64 sz;

  rid = sqlite3_value_int(argv[0]);
  if( !content_get(rid, &x) ){
    sqlite3_result_error(context, "mkdelta(X,Y): no content for X", -1);
    return;
  }
  zFile = (const char*)sqlite3_value_text(argv[1]);
  if( zFile==0 ){
    sqlite3_result_error(context, "mkdelta(X,Y): NULL Y argument", -1);
    blob_reset(&x);
    return;
  }
  sz = blob_read_from_file(&y, zFile, RepoFILE);
  if( sz<0 ){
    sqlite3_result_error(context, "mkdelta(X,Y): cannot read file Y", -1);
    blob_reset(&x);
    return;
  }
  blob_init(&out, 0, 0);
  blob_resize(&out, sz+70);
  if( blob_size(&x)==blob_size(&y)
   && memcmp(blob_buffer(&x), blob_buffer(&y), blob_size(&x))==0
  ){
    blob_reset(&y);
    blob_reset(&x);
    sqlite3_result_blob64(context, "", 0, SQLITE_STATIC);
    return;
  }
  nOut = delta_create(blob_buffer(&x),blob_size(&x),
                      blob_buffer(&y),blob_size(&y), blob_buffer(&out));
  blob_resize(&out, nOut);
  blob_reset(&x);
  blob_reset(&y);
  blob_compress(&out, &out);
  sqlite3_result_blob64(context, blob_buffer(&out), blob_size(&out),
                        SQLITE_TRANSIENT);
  blob_reset(&out);
}


/*
** Generate a binary patch file and store it into the file
** named zOut.  Or if zOut is NULL, write it into out.
**
** Return the number of errors.
*/
void patch_create(unsigned mFlags, const char *zOut, FILE *out){
  int vid;
  char *z;

  if( zOut && file_isdir(zOut, ExtFILE)!=0 ){
    if( mFlags & PATCH_FORCE ){
      file_delete(zOut);
    }
    if( file_isdir(zOut, ExtFILE)!=0 ){
      fossil_fatal("patch file already exists: %s", zOut);
    }
  }
  add_content_sql_commands(g.db);
  deltafunc_init(g.db);
  sqlite3_create_function(g.db, "read_co_file", 1, SQLITE_UTF8, 0,
                          readfileFunc, 0, 0);
  sqlite3_create_function(g.db, "mkdelta", 2, SQLITE_UTF8, 0,
                          mkdeltaFunc, 0, 0);
  db_multi_exec("ATTACH %Q AS patch;", zOut ? zOut : ":memory:");
  db_multi_exec(
    "PRAGMA patch.journal_mode=OFF;\n"
    "PRAGMA patch.page_size=512;\n"
    "CREATE TABLE patch.chng(\n"
    "  pathname TEXT,\n" /* Filename */
    "  origname TEXT,\n" /* Name before rename.  NULL if not renamed */
    "  hash TEXT,\n"     /* Baseline hash.  NULL for new files. */
    "  isexe BOOL,\n"    /* True if executable */
    "  islink BOOL,\n"   /* True if is a symbolic link */
    "  delta BLOB\n"     /* compressed delta. NULL if deleted.
                         **    length 0 if unchanged */
    ");"
    "CREATE TABLE patch.cfg(\n"
    "  key TEXT,\n"
    "  value ANY\n"
    ");"
  );
  vid = db_lget_int("checkout", 0);
  vfile_check_signature(vid, CKSIG_ENOTFILE);
  user_select();
  db_multi_exec(
    "INSERT INTO patch.cfg(key,value)"
    "SELECT 'baseline',uuid FROM blob WHERE rid=%d "
    "UNION ALL"
    " SELECT 'ckout',rtrim(%Q,'/')"
    "UNION ALL"
    " SELECT 'repo',%Q "
    "UNION ALL"
    " SELECT 'user',%Q "
    "UNION ALL"
    " SELECT 'date',julianday('now')"
    "UNION ALL"
    " SELECT name,value FROM repository.config"
    "  WHERE name IN ('project-code','project-name') "
    "UNION ALL"
    " SELECT 'fossil-date',julianday('" MANIFEST_DATE "')"
    ";", vid, g.zLocalRoot, g.zRepositoryName, g.zLogin);
  z = fossil_hostname();
  if( z ){
    db_multi_exec(
       "INSERT INTO patch.cfg(key,value)VALUES('hostname',%Q)", z);
    fossil_free(z);
  }

  /* New files */
  db_multi_exec(
    "INSERT INTO patch.chng(pathname,hash,isexe,islink,delta)"
    "  SELECT pathname, NULL, isexe, islink,"
    "         compress(read_co_file(%Q||pathname))"
    "    FROM vfile WHERE rid==0;",
    g.zLocalRoot
  );

  /* Deleted files */
  db_multi_exec(
    "INSERT INTO patch.chng(pathname,hash,isexe,islink,delta)"
    "  SELECT pathname, NULL, 0, 0, NULL"
    "    FROM vfile WHERE deleted;"
  );

  /* Changed files */
  db_multi_exec(
    "INSERT INTO patch.chng(pathname,origname,hash,isexe,islink,delta)"
    "  SELECT pathname, nullif(origname,pathname), blob.uuid, isexe, islink,"
            " mkdelta(blob.rid, %Q||pathname)"
    "    FROM vfile, blob"
    "   WHERE blob.rid=vfile.rid"
    "     AND NOT deleted AND (chnged OR origname<>pathname);",
    g.zLocalRoot
  );

  /* Merges */
  if( db_exists("SELECT 1 FROM localdb.vmerge WHERE id<=0") ){
    db_multi_exec(
      "CREATE TABLE patch.patchmerge(type TEXT,mhash TEXT);\n"
      "WITH tmap(id,type) AS (VALUES(0,'merge'),(-1,'cherrypick'),"
                                   "(-2,'backout'),(-4,'integrate'))"
      "INSERT INTO patch.patchmerge(type,mhash)"
      " SELECT tmap.type,vmerge.mhash FROM vmerge, tmap"
      "  WHERE tmap.id=vmerge.id;"
    );
  }

  /* Write the database to standard output if zOut==0 */
  if( zOut==0 ){
    sqlite3_int64 sz;
    unsigned char *pData;
    pData = sqlite3_serialize(g.db, "patch", &sz, 0);
    if( pData==0 ){
      fossil_fatal("out of memory");
    }
#ifdef _WIN32
    fflush(out);
    _setmode(_fileno(out), _O_BINARY);
#endif
    fwrite(pData, 1, sz, out);
    fflush(out);
    sqlite3_free(pData);
  }
  db_multi_exec("DETACH patch;");
}

/*
** Attempt to load and validate a patchfile identified by the first
** argument.
*/
void patch_attach(const char *zIn, FILE *in, int bIgnoreEmptyPatch){
  Stmt q;
  if( g.db==0 ){
    sqlite3_open(":memory:", &g.db);
  }
  if( zIn==0 ){
    Blob buf;
    int rc;
    int sz;
    unsigned char *pData;
    blob_init(&buf, 0, 0);
#ifdef _WIN32
    _setmode(_fileno(in), _O_BINARY);
#endif
    sz = blob_read_from_channel(&buf, in, -1);
    pData = (unsigned char*)blob_buffer(&buf);
    if( sz<512 ){
      blob_reset(&buf);
      if( bIgnoreEmptyPatch ) return;
      fossil_fatal("input is too small to be a patch file");
    }
    db_multi_exec("ATTACH ':memory:' AS patch");
    if( g.fSqlTrace ){
      fossil_trace("-- deserialize(\"patch\", pData, %lld);\n", sz);
    }
    rc = sqlite3_deserialize(g.db, "patch", pData, sz, sz, 0);
    if( rc ){
      fossil_fatal("cannot open patch database: %s", sqlite3_errmsg(g.db));
    }
  }else if( !file_isfile(zIn, ExtFILE) ){
    fossil_fatal("no such file: %s", zIn);
  }else{
    db_multi_exec("ATTACH %Q AS patch", zIn);
  }
  db_prepare(&q, "PRAGMA patch.quick_check");
  while( db_step(&q)==SQLITE_ROW ){
    if( fossil_strcmp(db_column_text(&q,0),"ok")!=0 ){
      fossil_fatal("file %s is not a well-formed Fossil patchfile", zIn);
    }
  }
  db_finalize(&q);
}

/*
** Show a summary of the content of a patch on standard output
*/
void patch_view(unsigned mFlags){
  Stmt q;
  db_prepare(&q,
    "WITH nmap(nkey,nm) AS (VALUES"
       "('baseline','BASELINE'),"
       "('project-name','PROJECT-NAME'))"
    "SELECT nm, value FROM nmap, patch.cfg WHERE nkey=key;"
  );
  while( db_step(&q)==SQLITE_ROW ){
    fossil_print("%-12s %s\n", db_column_text(&q,0), db_column_text(&q,1));
  }
  db_finalize(&q);
  if( mFlags & PATCH_VERBOSE ){
    db_prepare(&q,
      "WITH nmap(nkey,nm,isDate) AS (VALUES"
         "('project-code','PROJECT-CODE',0),"
         "('date','TIMESTAMP',1),"
         "('user','USER',0),"
         "('hostname','HOSTNAME',0),"
         "('ckout','CHECKOUT',0),"
         "('repo','REPOSITORY',0))"
      "SELECT nm, CASE WHEN isDate THEN datetime(value) ELSE value END"
      "  FROM nmap, patch.cfg WHERE nkey=key;"
    );
    while( db_step(&q)==SQLITE_ROW ){
      fossil_print("%-12s %s\n", db_column_text(&q,0), db_column_text(&q,1));
    }
    db_finalize(&q);
  }
  if( db_table_exists("patch","patchmerge") ){
    db_prepare(&q, "SELECT upper(type),mhash FROM patchmerge");
    while( db_step(&q)==SQLITE_ROW ){
      fossil_print("%-12s %s\n",
        db_column_text(&q,0),
        db_column_text(&q,1));
    }
    db_finalize(&q);
  }
  db_prepare(&q,
    "SELECT pathname,"                            /* 0: new name */
          " hash IS NULL AND delta IS NOT NULL,"  /* 1: isNew */
          " delta IS NULL,"                       /* 2: isDeleted */
          " origname"                             /* 3: old name or NULL */
    "  FROM patch.chng ORDER BY 1");
  while( db_step(&q)==SQLITE_ROW ){
    const char *zClass = "EDIT";
    const char *zName = db_column_text(&q,0);
    const char *zOrigName = db_column_text(&q, 3);
    if( db_column_int(&q, 1) && zOrigName==0 ){
      zClass = "NEW";
    }else if( db_column_int(&q, 2) ){
      zClass = zOrigName==0 ? "DELETE" : 0;
    }
    if( zOrigName!=0 && zOrigName[0]!=0 ){
      fossil_print("%-12s %s -> %s\n", "RENAME",zOrigName,zName);
    }
    if( zClass ){
      fossil_print("%-12s %s\n", zClass, zName);
    }
  }
  db_finalize(&q);
}

/*
** Apply the patch currently attached as database "patch".
**
** First update the check-out to be at "baseline".  Then loop through
** and update all files.
*/
void patch_apply(unsigned mFlags){
  Stmt q;
  Blob cmd;

  blob_init(&cmd, 0, 0);
  if( unsaved_changes(0) ){
    if( (mFlags & PATCH_FORCE)==0 ){
      fossil_fatal("Cannot apply patch: there are unsaved changes "
                   "in the current check-out");
    }else{
      blob_appendf(&cmd, "%$ revert --noundo", g.nameOfExe);
      if( mFlags & PATCH_DRYRUN ){
        fossil_print("%s\n", blob_str(&cmd));
      }else{
        int rc = fossil_system(blob_str(&cmd));
        if( rc ){
          fossil_fatal("unable to revert preexisting changes: %s",
                       blob_str(&cmd));
        }
      }
      blob_reset(&cmd);
    }
  }
  file_chdir(g.zLocalRoot, 0);
  db_prepare(&q,
    "SELECT patch.cfg.value"
    "  FROM patch.cfg, localdb.vvar"
    " WHERE patch.cfg.key='baseline'"
    "   AND localdb.vvar.name='checkout-hash'"
    "   AND patch.cfg.key<>localdb.vvar.name"
  );
  if( db_step(&q)==SQLITE_ROW ){
    blob_append_escaped_arg(&cmd, g.nameOfExe, 1);
    blob_appendf(&cmd, " update %s", db_column_text(&q, 0));
    if( mFlags & PATCH_VERBOSE ){
      fossil_print("%-10s %s\n", "BASELINE", db_column_text(&q,0));
    }
  }
  db_finalize(&q);
  if( blob_size(&cmd)>0 ){
    if( mFlags & PATCH_DRYRUN ){
      fossil_print("%s\n", blob_str(&cmd));
    }else{
      int rc = fossil_system(blob_str(&cmd));
      if( rc ){
        fossil_fatal("unable to update to the baseline check-out: %s",
                     blob_str(&cmd));
      }
    }
  }
  blob_reset(&cmd);
  if( db_table_exists("patch","patchmerge") ){
    int nMerge = 0;
    db_prepare(&q,
      "SELECT type, mhash, upper(type) FROM patch.patchmerge"
      " WHERE type IN ('merge','cherrypick','backout','integrate')"
      "   AND mhash NOT GLOB '*[^a-fA-F0-9]*';"
    );
    while( db_step(&q)==SQLITE_ROW ){
      const char *zType = db_column_text(&q,0);
      blob_append_escaped_arg(&cmd, g.nameOfExe, 1);
      if( strcmp(zType,"merge")==0 ){
        blob_appendf(&cmd, " merge --noundo --nosync %s\n",
                     db_column_text(&q,1));
      }else{
        blob_appendf(&cmd, " merge --%s --noundo --nosync %s\n",
                     zType, db_column_text(&q,1));
      }
      nMerge++;
      if( mFlags & PATCH_VERBOSE ){
        fossil_print("%-10s %s\n", db_column_text(&q,2),
                    db_column_text(&q,0));
      }
    }
    db_finalize(&q);
    if( mFlags & PATCH_DRYRUN ){
      fossil_print("%s", blob_str(&cmd));
    }else{
      int rc = fossil_unsafe_system(blob_str(&cmd));
      if( rc ){
        fossil_fatal("unable to do merges:\n%s",
                     blob_str(&cmd));
      }
    }
    blob_reset(&cmd);

    /* 2024-12-16 https://fossil-scm.org/home/forumpost/51a37054
    ** If one or more merge operations occurred in the patch and there are
    ** files that are marked as "chnged' in the local VFILE but which
    ** are not mentioned as having been modified in the patch, then
    ** revert those files.
    */
    if( nMerge ){
      int vid = db_lget_int("checkout", 0);
      int nRevert = 0;
      blob_append_escaped_arg(&cmd, g.nameOfExe, 1);
      blob_appendf(&cmd, " revert --noundo ");
      db_prepare(&q,
        "SELECT pathname FROM vfile WHERE vid=%d AND chnged "
        "EXCEPT SELECT pathname FROM chng",
        vid
      );
      while( db_step(&q)==SQLITE_ROW ){
        blob_append_escaped_arg(&cmd, db_column_text(&q,0), 1);
        nRevert++;
      }
      db_finalize(&q);
      if( nRevert ){
        if( mFlags & PATCH_DRYRUN ){
          fossil_print("%s", blob_str(&cmd));
        }else{
          int rc = fossil_unsafe_system(blob_str(&cmd));
          if( rc ){
            fossil_fatal("unable to do reverts:\n%s",
                         blob_str(&cmd));
          }
        }
      }
      blob_reset(&cmd);
    }
  }

  /* Deletions */
  db_prepare(&q, "SELECT pathname FROM patch.chng"
                 " WHERE origname IS NULL AND delta IS NULL");
  while( db_step(&q)==SQLITE_ROW ){
    if( blob_size(&cmd)==0 ){
      blob_append_escaped_arg(&cmd, g.nameOfExe, 1);
      blob_appendf(&cmd, " rm --hard");
    }
    blob_appendf(&cmd, " %$", db_column_text(&q,0));
    if( mFlags & PATCH_VERBOSE ){
      fossil_print("%-10s %s\n", "DELETE", db_column_text(&q,0));
    }
  }
  db_finalize(&q);
  if( blob_size(&cmd)>0 ){
    blob_appendf(&cmd, "\n");
    if( mFlags & PATCH_DRYRUN ){
      fossil_print("%s", blob_str(&cmd));
    }else{
      int rc = fossil_unsafe_system(blob_str(&cmd));
      if( rc ){
        fossil_fatal("unable to do merges:\n%s",
                     blob_str(&cmd));
      }
    }
    blob_reset(&cmd);
  }

  /* Renames */
  db_prepare(&q,
    "SELECT origname, pathname FROM patch.chng"
    " WHERE origname IS NOT NULL"
    "   AND origname<>pathname"
  );
  while( db_step(&q)==SQLITE_ROW ){
    blob_append_escaped_arg(&cmd, g.nameOfExe, 1);
    blob_appendf(&cmd, " mv --hard %$ %$\n",
        db_column_text(&q,0), db_column_text(&q,1));
    if( mFlags & PATCH_VERBOSE ){
      fossil_print("%-10s %s -> %s\n", "RENAME",
         db_column_text(&q,0), db_column_text(&q,1));
    }
  }
  db_finalize(&q);
  if( blob_size(&cmd)>0 ){
    if( mFlags & PATCH_DRYRUN ){
      fossil_print("%s", blob_str(&cmd));
    }else{
      int rc = fossil_unsafe_system(blob_str(&cmd));
      if( rc ){
        fossil_print("%-10s unable to rename files:\n%s", "WARNING!",
                       blob_str(&cmd));
      }
    }
    blob_reset(&cmd);
  }

  /* Edits and new files */
  db_prepare(&q,
    "SELECT pathname, hash, isexe, islink, delta FROM patch.chng"
    " WHERE delta IS NOT NULL"
  );
  while( db_step(&q)==SQLITE_ROW ){
    const char *zPathname = db_column_text(&q,0);
    const char *zHash = db_column_text(&q,1);
    int isExe = db_column_int(&q,2);
    int isLink = db_column_int(&q,3);
    Blob data;

    blob_init(&data, 0, 0);
    db_ephemeral_blob(&q, 4, &data);
    if( blob_size(&data) ){
      blob_uncompress(&data, &data);
    }
    if( blob_size(&data)==0 ){
      /* No changes to the file */
      continue;
    }else if( zHash ){
      Blob basis;
      int rid = fast_uuid_to_rid(zHash);
      int outSize, sz;
      char *aOut;
      if( rid==0 ){
        fossil_fatal("cannot locate basis artifact %s for %s",
                     zHash, zPathname);
      }
      if( !content_get(rid, &basis) ){
        fossil_fatal("cannot load basis artifact %d for %s", rid, zPathname);
      }
      outSize = delta_output_size(blob_buffer(&data),blob_size(&data));
      if( outSize<=0 ){
        fossil_fatal("malformed delta for %s", zPathname);
      }
      aOut = sqlite3_malloc64( outSize+1 );
      if( aOut==0 ){
        fossil_fatal("out of memory");
      }
      sz = delta_apply(blob_buffer(&basis), blob_size(&basis),
                       blob_buffer(&data), blob_size(&data), aOut);
      if( sz<0 ){
        fossil_fatal("malformed delta for %s", zPathname);
      }
      blob_reset(&basis);
      blob_reset(&data);
      blob_append(&data, aOut, sz);
      sqlite3_free(aOut);
      if( mFlags & PATCH_VERBOSE ){
        fossil_print("%-10s %s\n", "EDIT", zPathname);
      }
    }else{
      blob_append_escaped_arg(&cmd, g.nameOfExe, 1);
      blob_appendf(&cmd, " add %$\n", zPathname);
      if( mFlags & PATCH_VERBOSE ){
        fossil_print("%-10s %s\n", "NEW", zPathname);
      }
    }
    if( (mFlags & PATCH_DRYRUN)==0 ){
      if( isLink ){
        symlink_create(blob_str(&data), zPathname);
      }else{
        blob_write_to_file(&data, zPathname);
      }
      file_setexe(zPathname, isExe);
      blob_reset(&data);
    }
  }
  db_finalize(&q);
  if( blob_size(&cmd)>0 ){
    if( mFlags & PATCH_DRYRUN ){
      fossil_print("%s", blob_str(&cmd));
    }else{
      int rc = fossil_unsafe_system(blob_str(&cmd));
      if( rc ){
        fossil_fatal("unable to add new files:\n%s",
                     blob_str(&cmd));
      }
    }
    blob_reset(&cmd);
  }
}

/*
** This routine processes the
**
**   ...  [--dir64 DIR64] [DIRECTORY] FILENAME
**
** part of various "fossil patch" subcommands.
**
** Find and return the filename of the patch file to be used by
** "fossil patch apply" or "fossil patch create".  Space to hold
** the returned name is obtained from fossil_malloc() and should
** be freed by the caller.
**
** If the name is "-" return NULL.  The caller will interpret this
** to mean the patch is coming in over stdin or going out over
** stdout.
**
** If there is a prior DIRECTORY argument, or if
** the --dir64 option is present, first chdir to the specified
** directory, and adjust the path of FILENAME as appropriate so
** that it still points to the same file.
**
** The --dir64 option is undocumented.  The argument to --dir64
** is a base64-encoded directory name.  The --dir64 option is used
** to transmit the directory as part of the command argument to
** a "ssh" command without having to worry about quoting
** any special characters in the filename.
**
** The returned name is obtained from fossil_malloc() and should
** be freed by the caller.
*/
static char *patch_find_patch_filename(const char *zCmdName){
  const char *zDir64 = find_option("dir64",0,1);
  const char *zDir = 0;
  const char *zBaseName;
  char *zToFree = 0;
  char *zPatchFile = 0;
  if( zDir64 ){
    int n = 0;
    zToFree = decode64(zDir64, &n);
    zDir = zToFree;
  }
  verify_all_options();
  if( g.argc!=4 && g.argc!=5 ){
    usage(mprintf("%s [DIRECTORY] FILENAME", zCmdName));
  }
  if( g.argc==5 ){
    zDir = g.argv[3];
    zBaseName = g.argv[4];
  }else{
    zBaseName = g.argv[3];
  }
  if( fossil_strcmp(zBaseName, "-")==0 ){
    zPatchFile = 0;
  }else if( zDir ){
    zPatchFile = file_canonical_name_dup(g.argv[4]);
  }else{
    zPatchFile = fossil_strdup(g.argv[3]);
  }
  if( zDir && file_chdir(zDir,0) ){
    fossil_fatal("cannot change to directory \"%s\"", zDir);
  }
  fossil_free(zToFree);
  return zPatchFile;
}

/*
** Resolves a patch-command remote system name, accounting for patch
** aliases.
**
** If a CONFIG table entry matching name='patch-alias:$zKey' is found,
** the corresponding value is returned, else a fossil_strdup() of zKey
** is returned. The caller is responsible for passing the resulting
** string to fossil_free().
*/
static char *patch_resolve_remote(const char *zKey){
  char *zAlias = db_text(0, "SELECT value FROM config "
                            "WHERE name = 'patch-alias:%q'",
                            zKey);
  return zAlias ? zAlias : fossil_strdup(zKey);
}

/*
** Create a FILE* that will execute the remote side of a push or pull
** using ssh (probably) or fossil for local pushes and pulls.  Return
** a FILE* obtained from popen() into which we write the patch, or from
** which we read the patch, depending on whether this is a push or pull.
*/
static FILE *patch_remote_command(
  unsigned mFlags,             /* flags */
  const char *zThisCmd,        /* "push" or "pull" */
  const char *zRemoteCmd,      /* "apply" or "create" */
  const char *zFossilCmd,      /* Name of "fossil" on remote system */
  const char *zRW              /* "w" or "r" */
){
  char *zRemote = 0;
  char *zDir = 0;
  Blob cmd;
  FILE *f = 0;
  Blob flgs;
  char *zForce = 0;
  int isRetry = (mFlags & PATCH_RETRY)!=0;

  blob_init(&flgs, 0, 0);
  blob_init(&cmd, 0, 0);
  if( mFlags & PATCH_FORCE )  blob_appendf(&flgs, " -f");
  if( mFlags & PATCH_VERBOSE )  blob_appendf(&flgs, " -v");
  if( mFlags & PATCH_DRYRUN )  blob_appendf(&flgs, " -n");
  zForce = blob_size(&flgs)>0 ? blob_str(&flgs) : "";
  if( g.argc!=4 ){
    usage(mprintf("%s [USER@]HOST:DIRECTORY", zThisCmd));
  }
  zRemote = patch_resolve_remote(g.argv[3]);
  zDir = (char*)file_skip_userhost(zRemote);
  if( zDir==0 ){
    if( isRetry ) goto remote_command_error;
    zDir = zRemote;
    blob_append_escaped_arg(&cmd, g.nameOfExe, 1);
    blob_appendf(&cmd, " patch %s%s %$ -", zRemoteCmd, zForce, zDir);
  }else{
    Blob remote;
    *(char*)(zDir-1) = 0;
    transport_ssh_command(&cmd);
    blob_appendf(&cmd, " -T");
    blob_append_escaped_arg(&cmd, zRemote, 0);
    blob_init(&remote, 0, 0);
    if( zFossilCmd==0 ){
      if( ssh_needs_path_argument(zRemote,-1) ^ isRetry ){
        ssh_add_path_argument(&cmd);
      }
      zFossilCmd = "fossil";
    }else if( mFlags & PATCH_RETRY ){
      goto remote_command_error;
    }
    blob_appendf(&remote, "%$ patch %s%s --dir64 %z -",
                 zFossilCmd, zRemoteCmd, zForce, encode64(zDir, -1));
    blob_append_escaped_arg(&cmd, blob_str(&remote), 0);
    blob_reset(&remote);
  }
  if( isRetry ){
    fossil_print("First attempt to run \"fossil\" on %s failed\n"
                 "Retry: ", zRemote);
  }
  fossil_print("%s\n", blob_str(&cmd));
  fflush(stdout);
  f = popen(blob_str(&cmd), zRW);
  if( f==0 ){
    fossil_fatal("cannot run command: %s", blob_str(&cmd));
  }
remote_command_error:
  fossil_free(zRemote);
  blob_reset(&cmd);
  blob_reset(&flgs);
  return f;
}

/*
** Toggle the use-path-for-ssh setting for the remote host defined
** by g.argv[3].
*/
static void patch_toggle_ssh_needs_path(void){
  char *zRemote = patch_resolve_remote(g.argv[3]);
  char *zDir = (char*)file_skip_userhost(zRemote);
  if( zDir ){
    *(char*)(zDir - 1) =  0;
    ssh_needs_path_argument(zRemote, 99);
  }
  fossil_free(zRemote);
}

/*
** Show a diff for the patch currently loaded into database "patch".
*/
static void patch_diff(
  unsigned mFlags,         /* Patch flags.  only -f is allowed */
  DiffConfig *pCfg         /* Diff options */
){
  int nErr = 0;
  Stmt q;
  int bWebpage = (pCfg->diffFlags & DIFF_WEBPAGE)!=0;
  Blob empty;
  blob_zero(&empty);

  if( (mFlags & PATCH_FORCE)==0 ){
    /* Check to ensure that the patch is against the repository that
    ** we have opened.
    **
    ** To do: If there is a mismatch, should we scan all of the repositories
    ** listed in the global_config table looking for a match?
    */
    if( db_exists(
        "SELECT 1 FROM patch.cfg"
        " WHERE cfg.key='baseline'"
        "   AND NOT EXISTS(SELECT 1 FROM blob WHERE uuid=cfg.value)"
    )){
      char *zBaseline;
      db_prepare(&q,
         "SELECT config.value, cfg.value FROM config, cfg"
         " WHERE config.name='project-name'"
         "   AND cfg.key='project-name'"
         "   AND config.value<>cfg.value"
      );
      if( db_step(&q)==SQLITE_ROW ){
        char *zRepo = fossil_strdup(db_column_text(&q,0));
        char *zPatch = fossil_strdup(db_column_text(&q,1));
        db_finalize(&q);
        fossil_fatal("the patch is against project \"%z\" but you are using "
                     "project \"%z\"", zPatch, zRepo);
      }
      db_finalize(&q);
      zBaseline = db_text(0, "SELECT value FROM patch.cfg"
                             " WHERE key='baseline'");
      if( zBaseline ){
        fossil_fatal("the baseline of the patch (check-in %S) is not found "
                     "in the %s repository", zBaseline, g.zRepositoryName);
      }
    }
  }

  diff_begin(pCfg);
  db_prepare(&q,
     "SELECT"
       " (SELECT blob.rid FROM blob WHERE blob.uuid=chng.hash),"
       " pathname,"    /* 1: new pathname */
       " origname,"    /* 2: original pathname.  Null if not renamed */
       " delta,"       /* 3: delta.  NULL if deleted.  empty is no change */
       " hash"         /* 4: baseline hash */
     " FROM patch.chng"
     " ORDER BY pathname"
  );
  while( db_step(&q)==SQLITE_ROW ){
    int rid;
    const char *zName;
    Blob a, b;

    if( db_column_type(&q,0)!=SQLITE_INTEGER
     && db_column_type(&q,4)==SQLITE_TEXT
    ){
      char *zUuid = fossil_strdup(db_column_text(&q,4));
      char *zName = fossil_strdup(db_column_text(&q,1));
      if( mFlags & PATCH_FORCE ){
        fossil_print("ERROR cannot find base artifact %S for file \"%s\"\n",
                     zUuid, zName);
        nErr++;
        fossil_free(zUuid);
        fossil_free(zName);
        continue;
      }else{
        db_finalize(&q);
        fossil_fatal("base artifact %S for file \"%s\" not found",
                     zUuid, zName);
      }
    }
    zName = db_column_text(&q, 1);
    rid = db_column_int(&q, 0);

    pCfg->diffFlags &= (~DIFF_FILE_MASK);
    if( db_column_type(&q,3)==SQLITE_NULL ){
      if( !bWebpage ) fossil_print("DELETE %s\n", zName);
      pCfg->diffFlags |= DIFF_FILE_DELETED;
      diff_print_index(zName, pCfg, 0);
      content_get(rid, &a);
      diff_file_mem(&a, &empty, zName, pCfg);
    }else if( rid==0 ){
      db_ephemeral_blob(&q, 3, &a);
      blob_uncompress(&a, &a);
      if( !bWebpage ) fossil_print("ADDED %s\n", zName);
      pCfg->diffFlags |= DIFF_FILE_ADDED;
      diff_print_index(zName, pCfg, 0);
      diff_file_mem(&empty, &a, zName, pCfg);
      blob_reset(&a);
    }else if( db_column_bytes(&q, 3)>0 ){
      Blob delta;
      db_ephemeral_blob(&q, 3, &delta);
      blob_uncompress(&delta, &delta);
      content_get(rid, &a);
      blob_delta_apply(&a, &delta, &b);
      diff_file_mem(&a, &b, zName, pCfg);
      blob_reset(&a);
      blob_reset(&b);
      blob_reset(&delta);
    }
  }
  db_finalize(&q);
  diff_end(pCfg, nErr);
  if( nErr ) fossil_fatal("abort due to prior errors");
}

/*
** COMMAND: patch
**
** Usage: %fossil patch SUBCOMMAND ?ARGS ..?
**
** This command is used to create, view, and apply Fossil binary patches.
** A Fossil binary patch is a single (binary) file that captures all of the
** uncommitted changes of a check-out.  Use Fossil binary patches to transfer
** proposed or incomplete changes between machines for testing or analysis.
**
** > fossil patch alias add|rm|ls|list ?ARGS?
**
**       Manage remote-name aliases, which act as short-form
**       equivalents to REMOTE-CHECKOUT strings. Aliases are local to
**       a given repository and do not sync. Subcommands:
**
**         ... add ALIAS REMOTE-CHECKOUT       Add ALIAS as an alias
**                                             for REMOTE-CHECKOUT.
**         ... ls|list                         List all local aliases.
**         ... rm ALIAS [ALIAS...]             Remove named aliases
**         ... rm --all                        Remove all aliases
**
** > fossil patch create [DIRECTORY] PATCHFILE
**
**       Create a new binary patch in PATCHFILE that captures all uncommitted
**       changes in the check-out at DIRECTORY, or the current directory if
**       DIRECTORY is omitted.  If PATCHFILE is "-" then the binary patch
**       is written to standard output.
**
**       Options:
**           -f|--force     Overwrite an existing patch with the same name
**
** > fossil patch apply [DIRECTORY] PATCHFILE
**
**       Apply the changes in PATCHFILE to the check-out at DIRECTORY, or
**       in the current directory if DIRECTORY is omitted.
**
**       Options:
**           -f|--force     Apply the patch even though there are unsaved
**                          changes in the current check-out.  Unsaved changes
**                          are reverted and permanently lost.
**           -n|--dry-run   Do nothing, but print what would have happened
**           -v|--verbose   Extra output explaining what happens
**
** > fossil patch diff [DIRECTORY] PATCHFILE
** > fossil patch gdiff [DIRECTORY] PATCHFILE
**
**       Show a human-readable diff for the patch in PATCHFILE and associated
**       with the repository checked out in DIRECTORY.  The current
**       directory is used if DIRECTORY is omitted. All the usual
**       diff flags described at "fossil help diff" apply. With gdiff,
**       gdiff-command is used instead of internal diff logic.  In addition:
**
**           -f|--force     Continue trying to perform the diff even if
**                          baseline information is missing from the current
**                          repository
**
** > fossil patch push REMOTE-CHECKOUT
**
**       Create a patch for the current check-out, transfer that patch to
**       a remote machine (using ssh) and apply the patch there.  The
**       REMOTE-CHECKOUT is in one of the following formats:
**
**           *   DIRECTORY
**           *   HOST:DIRECTORY
**           *   USER@HOST:DIRECTORY
**
**       The name of the fossil executable on the remote host is specified
**       by the --fossilcmd option, or if there is no --fossilcmd, it first
**       tries "$HOME/bin/fossil" and if not found there it searches for any
**       executable named "fossil" on the default $PATH set by SSH on the
**       remote.
**
**       Command-line options:
**
**           -f|--force         Apply the patch even though there are unsaved
**                              changes in the current check-out.  Unsaved
**                              changes will be reverted and then the patch is
**                              applied.
**           --fossilcmd EXE    Name of the "fossil" executable on the remote
**           -n|--dry-run       Do nothing, but print what would have happened
**           -v|--verbose       Extra output explaining what happens
**
**
** > fossil patch pull REMOTE-CHECKOUT
**
**       Like "fossil patch push" except that the transfer is from remote
**       to local.  All the same command-line options apply.
**
** > fossil patch view PATCHFILE
**
**       View a summary of the changes in the binary patch in PATCHFILE.
**       Use "fossil patch diff" for detailed patch content.
**
**           -v|--verbose       Show extra detail about the patch
**
*/
void patch_cmd(void){
  const char *zCmd;
  size_t n;
  if( g.argc<3 ){
    patch_usage:
    usage("alias|apply|create|diff|gdiff|pull|push|view");
  }
  zCmd = g.argv[2];
  n = strlen(zCmd);
  if( strncmp(zCmd, "alias", n)==0 ){
    const char * zArg = g.argc>3 ? g.argv[3] : 0;
    db_must_be_within_tree();
    if( 0==zArg ){
      goto usage_patch_alias;
    }else if( 0==strcmp("ls",zArg) || 0==strcmp("list",zArg) ){
      /* alias ls|list */
      Stmt q;
      int nAlias = 0;

      verify_all_options();
      db_prepare(&q, "SELECT substr(name,13), value FROM config "
                 "WHERE name GLOB 'patch-alias:*' ORDER BY name");
      while( SQLITE_ROW==db_step(&q) ){
        const char *zName = db_column_text(&q, 0);
        const char *zVal = db_column_text(&q, 1);
        ++nAlias;
        fossil_print("%s = %s\n", zName, zVal);
      }
      db_finalize(&q);
      if( 0==nAlias ){
        fossil_print("No patch aliases defined\n");
      }
    }else if( 0==strcmp("add", zArg) ){
      /* alias add localName remote */
      verify_all_options();
      if( 6!=g.argc ){
        usage("alias add localName remote");
      }
      db_unprotect(PROTECT_CONFIG);
      db_multi_exec("REPLACE INTO config (name, value, mtime) "
                    "VALUES ('patch-alias:%q', %Q, unixepoch())",
                    g.argv[4], g.argv[5]);
      db_protect_pop();
    }else if( 0==strcmp("rm", zArg) ){
      /* alias rm */
      const int fAll = 0!=find_option("all", 0, 0);
      if( fAll ? g.argc<4 : g.argc<5 ){
        usage("alias rm [-all] [aliasGlob [...aliasGlobN]]");
      }
      verify_all_options();
      db_unprotect(PROTECT_CONFIG);
      if( 0!=fAll ){
        db_multi_exec("DELETE FROM config WHERE name GLOB 'patch-alias:*'");
      }else{
        Stmt q;
        int i;
        db_prepare(&q, "DELETE FROM config WHERE name "
                   "GLOB 'patch-alias:' || :pattern");
        for(i = 4; i < g.argc; ++i){
          db_bind_text(&q, ":pattern", g.argv[i]);
          db_step(&q);
          db_reset(&q);
        }
        db_finalize(&q);
      }
      db_protect_pop();
    }else{
    usage_patch_alias:
      usage("alias ls|list|add|rm ...");
    }
  }else
  if( strncmp(zCmd, "apply", n)==0 ){
    char *zIn;
    unsigned flags = 0;
    if( find_option("dry-run","n",0) )  flags |= PATCH_DRYRUN;
    if( find_option("verbose","v",0) )  flags |= PATCH_VERBOSE;
    if( find_option("force","f",0) )    flags |= PATCH_FORCE;
    zIn = patch_find_patch_filename("apply");
    db_must_be_within_tree();
    patch_attach(zIn, stdin, 0);
    patch_apply(flags);
    fossil_free(zIn);
  }else
  if( strncmp(zCmd, "create", n)==0 ){
    char *zOut;
    unsigned flags = 0;
    if( find_option("force","f",0) )    flags |= PATCH_FORCE;
    zOut = patch_find_patch_filename("create");
    verify_all_options();
    db_must_be_within_tree();
    patch_create(flags, zOut, stdout);
    fossil_free(zOut);
  }else
  if( (strncmp(zCmd, "diff", n)==0) || (strncmp(zCmd, "gdiff", n)==0) ){
    char *zIn;
    unsigned flags = 0;
    DiffConfig DCfg;

    if( find_option("tk",0,0)!=0 ){
      db_close(0);
      diff_tk("patch diff", 3);
      return;
    }
    db_find_and_open_repository(0, 0);
    if( gdiff_using_tk(zCmd[0]=='g') ){
      diff_tk("patch diff", 3);
      return;
    }
    if( find_option("force","f",0) )    flags |= PATCH_FORCE;
    diff_options(&DCfg, zCmd[0]=='g', 0);
    verify_all_options();
    zIn = patch_find_patch_filename("diff");
    patch_attach(zIn, stdin, 0);
    patch_diff(flags, &DCfg);
    fossil_free(zIn);
  }else
  if( strncmp(zCmd, "pull", n)==0 ){
    FILE *pIn = 0;
    unsigned flags = 0;
    const char *zFossilCmd = find_option("fossilcmd",0,1);
    if( find_option("dry-run","n",0) )  flags |= PATCH_DRYRUN;
    if( find_option("verbose","v",0) )  flags |= PATCH_VERBOSE;
    if( find_option("force","f",0) )    flags |= PATCH_FORCE;
    db_must_be_within_tree();
    verify_all_options();
    pIn = patch_remote_command(flags & (~PATCH_FORCE),
                 "pull", "create", zFossilCmd, "r");
    if( pIn ){
      patch_attach(0, pIn, 1);
      if( pclose(pIn) ){
        flags |= PATCH_RETRY;
        pIn = patch_remote_command(flags & (~PATCH_FORCE),
                     "pull", "create", zFossilCmd, "r");
        if( pIn ){
          patch_attach(0, pIn, 0);
          if( pclose(pIn)==0 ){
            patch_toggle_ssh_needs_path();
          }
        }
      }
      patch_apply(flags);
    }
  }else
  if( strncmp(zCmd, "push", n)==0 ){
    FILE *pOut = 0;
    unsigned flags = 0;
    const char *zFossilCmd = find_option("fossilcmd",0,1);
    if( find_option("dry-run","n",0) )  flags |= PATCH_DRYRUN;
    if( find_option("verbose","v",0) )  flags |= PATCH_VERBOSE;
    if( find_option("force","f",0) )    flags |= PATCH_FORCE;
    db_must_be_within_tree();
    verify_all_options();
    pOut = patch_remote_command(flags, "push", "apply", zFossilCmd, "w");
    if( pOut ){
      patch_create(0, 0, pOut);
      if( pclose(pOut)!=0 ){
        flags |= PATCH_RETRY;
        pOut = patch_remote_command(flags, "push", "apply", zFossilCmd, "w");
        if( pOut ){
          patch_create(0, 0, pOut);
          if( pclose(pOut)==0 ){
            patch_toggle_ssh_needs_path();
          }
        }
      }
    }
  }else
  if( strncmp(zCmd, "view", n)==0 ){
    const char *zIn;
    unsigned int flags = 0;
    if( find_option("verbose","v",0) )  flags |= PATCH_VERBOSE;
    verify_all_options();
    if( g.argc!=4 ){
      usage("view FILENAME");
    }
    zIn = g.argv[3];
    if( fossil_strcmp(zIn, "-")==0 ) zIn = 0;
    patch_attach(zIn, stdin, 0);
    patch_view(flags);
  }else
  {
    goto patch_usage;
  }
}