/*
** Copyright (c) 2011 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/
**
*******************************************************************************
**  
** Code for the JSON API.
**
** For notes regarding the public JSON interface, please see:
**
** https://docs.google.com/document/d/1fXViveNhDbiXgCuE7QDXQOKeFzf2qNUkBEgiUvoqFN4/edit
**
**
** Notes for hackers...
**
** Here's how command/page dispatching works: json_page_top() (in HTTP mode) or
** json_cmd_top() (in CLI mode) catch the "json" path/command. Those functions then
** dispatch to a JSON-mode-specific command/page handler with the type fossil_json_f().
** See the API docs for that typedef (below) for the semantics of the callbacks.
**
**
*/
#include "config.h"
#include "VERSION.h"
#include "json.h"
#include <assert.h>
#include <time.h>

#if INTERFACE
#include "json_detail.h" /* workaround for apparent enum limitation in makeheaders */
#endif

const FossilJsonKeys_ FossilJsonKeys = {
  "anonymousSeed" /*anonymousSeed*/,
  "authToken"  /*authToken*/,
  "COMMAND_PATH" /*commandPath*/,
  "payload" /* payload */,
  "requestId" /*requestId*/,
  "resultCode" /*resultCode*/,
  "resultText" /*resultText*/,
  "timestamp" /*timestamp*/
};

/*
** Internal helpers to manipulate a byte array as a bitset. The B
** argument must be-a array at least (BIT/8+1) bytes long.
** The BIT argument is the bit number to query/set/clear/toggle.
*/
#define BITSET_BYTEFOR(B,BIT) ((B)[ BIT / 8 ])
#define BITSET_SET(B,BIT) ((BITSET_BYTEFOR(B,BIT) |= (0x01 << (BIT%8))),0x01)
#define BITSET_UNSET(B,BIT) ((BITSET_BYTEFOR(B,BIT) &= ~(0x01 << (BIT%8))),0x00)
#define BITSET_GET(B,BIT) ((BITSET_BYTEFOR(B,BIT) & (0x01 << (BIT%8))) ? 0x01 : 0x00)
#define BITSET_TOGGLE(B,BIT) (BITSET_GET(B,BIT) ? (BITSET_UNSET(B,BIT)) : (BITSET_SET(B,BIT)))


/* Timer code taken from sqlite3's shell.c, modified slightly.
   FIXME: move the timer into the fossil core API so that we can
   start the timer early on in the app init phase. Right now we're
   just timing the json ops themselves.
*/
#if !defined(_WIN32) && !defined(WIN32) && !defined(__OS2__) && !defined(__RTP__) && !defined(_WRS_KERNEL)
#include <sys/time.h>
#include <sys/resource.h>

/* Saved resource information for the beginning of an operation */
static struct rusage sBegin;

/*
** Begin timing an operation
*/
static void beginTimer(void){
  getrusage(RUSAGE_SELF, &sBegin);
}

/* Return the difference of two time_structs in milliseconds */
static double timeDiff(struct timeval *pStart, struct timeval *pEnd){
  return ((pEnd->tv_usec - pStart->tv_usec)*0.001 + 
          (double)((pEnd->tv_sec - pStart->tv_sec)*1000.0));
}

/*
** Print the timing results.
*/
static double endTimer(void){
  struct rusage sEnd;
  getrusage(RUSAGE_SELF, &sEnd);
  return timeDiff(&sBegin.ru_utime, &sEnd.ru_utime)
    + timeDiff(&sBegin.ru_stime, &sEnd.ru_stime);
#if 0
  printf("CPU Time: user %f sys %f\n",
         timeDiff(&sBegin.ru_utime, &sEnd.ru_utime),
         timeDiff(&sBegin.ru_stime, &sEnd.ru_stime));
#endif
}

#define BEGIN_TIMER beginTimer()
#define END_TIMER endTimer()
#define HAS_TIMER 1

#elif (defined(_WIN32) || defined(WIN32))

#include <windows.h>

/* Saved resource information for the beginning of an operation */
static HANDLE hProcess;
static FILETIME ftKernelBegin;
static FILETIME ftUserBegin;
typedef BOOL (WINAPI *GETPROCTIMES)(HANDLE, LPFILETIME, LPFILETIME, LPFILETIME, LPFILETIME);
static GETPROCTIMES getProcessTimesAddr = NULL;

/*
** Check to see if we have timer support.  Return 1 if necessary
** support found (or found previously).
*/
static int hasTimer(void){
  if( getProcessTimesAddr ){
    return 1;
  } else {
    /* GetProcessTimes() isn't supported in WIN95 and some other Windows versions.
    ** See if the version we are running on has it, and if it does, save off
    ** a pointer to it and the current process handle.
    */
    hProcess = GetCurrentProcess();
    if( hProcess ){
      HINSTANCE hinstLib = LoadLibrary(TEXT("Kernel32.dll"));
      if( NULL != hinstLib ){
        getProcessTimesAddr = (GETPROCTIMES) GetProcAddress(hinstLib, "GetProcessTimes");
        if( NULL != getProcessTimesAddr ){
          return 1;
        }
        FreeLibrary(hinstLib); 
      }
    }
  }
  return 0;
}

/*
** Begin timing an operation
*/
static void beginTimer(void){
  if( getProcessTimesAddr ){
    FILETIME ftCreation, ftExit;
    getProcessTimesAddr(hProcess, &ftCreation, &ftExit, &ftKernelBegin, &ftUserBegin);
  }
}

/* Return the difference of two FILETIME structs in milliseconds */
static double timeDiff(FILETIME *pStart, FILETIME *pEnd){
  sqlite_int64 i64Start = *((sqlite_int64 *) pStart);
  sqlite_int64 i64End = *((sqlite_int64 *) pEnd);
  return (double) ((i64End - i64Start) / 10000.0);
}

/*
** Print the timing results.
*/
static double endTimer(void){
  if(getProcessTimesAddr){
    FILETIME ftCreation, ftExit, ftKernelEnd, ftUserEnd;
    getProcessTimesAddr(hProcess, &ftCreation, &ftExit, &ftKernelEnd, &ftUserEnd);
    return timeDiff(&ftUserBegin, &ftUserEnd) +
      timeDiff(&ftKernelBegin, &ftKernelEnd);
  }
}

#define BEGIN_TIMER beginTimer()
#define END_TIMER endTimer()
#define HAS_TIMER hasTimer()

#else
#define BEGIN_TIMER 
#define END_TIMER 0.0
#define HAS_TIMER 0
#endif


/*
** Returns true if fossil is running in JSON mode and we are either
** running in HTTP mode OR g.json.post.o is not NULL (meaning POST
** data was fed in from CLI mode).
**
** Specifically, it will return false when any of these apply:
**
** a) Not running in JSON mode (via json command or /json path).
**
** b) We are running in JSON CLI mode, but no POST data has been fed
** in.
**
** Whether or not we need to take args from CLI or POST data makes a
** difference in argument/parameter handling in many JSON rountines.
*/
char fossil_is_json(){
  return g.json.isJsonMode && (g.isHTTP || g.json.post.o);
}

/*
** Placeholder /json/XXX page impl for NYI (Not Yet Implemented)
** (but planned) pages/commands.
*/
cson_value * json_page_nyi(){
  g.json.resultCode = FSL_JSON_E_NYI;
  return NULL;
}

/*
** Given a FossilJsonCodes value, it returns a string suitable for use
** as a resultText string. Returns some unspecified string if errCode
** is not one of the FossilJsonCodes values.
*/
char const * json_err_str( int errCode ){
  switch( errCode ){
    case 0: return "Success";
#define C(X,V) case FSL_JSON_E_ ## X: return V

    C(GENERIC,"Generic error");
    C(INVALID_REQUEST,"Invalid request");
    C(UNKNOWN_COMMAND,"Unknown Command");
    C(UNKNOWN,"Unknown error");
    C(TIMEOUT,"Timeout reached");
    C(ASSERT,"Assertion failed");
    C(ALLOC,"Resource allocation failed");
    C(NYI,"Not yet implemented");
    C(PANIC,"x");
    C(MANIFEST_READ_FAILED,"Reading artifact manifest failed.");
    C(FILE_OPEN_FAILED,"Opening file failed.");
    
    C(AUTH,"Authentication error");
    C(MISSING_AUTH,"Authentication info missing from request");
    C(DENIED,"Access denied");
    C(WRONG_MODE,"Request not allowed (wrong operation mode)");
    C(LOGIN_FAILED,"Login failed");
    C(LOGIN_FAILED_NOSEED,"Anonymous login attempt was missing password seed");
    C(LOGIN_FAILED_NONAME,"Login failed - name not supplied");
    C(LOGIN_FAILED_NOPW,"Login failed - password not supplied");
    C(LOGIN_FAILED_NOTFOUND,"Login failed - no match found");

    C(USAGE,"Usage error");
    C(INVALID_ARGS,"Invalid argument(s)");
    C(MISSING_ARGS,"Missing argument(s)");
    C(AMBIGUOUS_UUID,"Resource identifier is ambiguous");
    C(UNRESOLVED_UUID,"Provided uuid/tag/branch could not be resolved");
    C(RESOURCE_ALREADY_EXISTS,"Resource already exists");
    C(RESOURCE_NOT_FOUND,"Resource not found");

    C(DB,"Database error");
    C(STMT_PREP,"Statement preparation failed");
    C(STMT_BIND,"Statement parameter binding failed");
    C(STMT_EXEC,"Statement execution/stepping failed");
    C(DB_LOCKED,"Database is locked");
    C(DB_NEEDS_REBUILD,"Fossil repository needs to be rebuilt");
#undef C
    default:
      return "Unknown Error";
  }
}

/*
** Implements the cson_data_dest_f() interface and outputs the data to
** a fossil Blob object.  pState must be-a initialized (Blob*), to
** which n bytes of src will be appended.
**/
int cson_data_dest_Blob(void * pState, void const * src, unsigned int n){
  Blob * b = (Blob*)pState;
  blob_append( b, (char const *)src, (int)n ) /* will die on OOM */;
  return 0;
}

/*
** Implements the cson_data_source_f() interface and reads
** input from a fossil Blob object. pState must be-a (Blob*).
*/
int cson_data_src_Blob(void * pState, void * dest, unsigned int * n){
  Blob * b = (Blob*)pState;
  *n = blob_read( b, dest, *n );
  return 0;
}

/*
** Convenience wrapper around cson_output() which appends the output
** to pDest. pOpt may be NULL, in which case g.json.outOpt will be used.
*/
int cson_output_Blob( cson_value const * pVal, Blob * pDest, cson_output_opt const * pOpt ){
  return cson_output( pVal, cson_data_dest_Blob,
                      pDest, pOpt ? pOpt : &g.json.outOpt );
}

/*
** Convenience wrapper around cson_parse() which reads its input
** from pSrc. pSrc is rewound before parsing.
**
** pInfo may be NULL. If it is not NULL then it will contain details
** about the parse state when this function returns.
**
** On success a new JSON Object or Array is returned. On error NULL is
** returned.
*/
cson_value * cson_parse_Blob( Blob * pSrc, cson_parse_info * pInfo ){
  cson_value * root = NULL;
  blob_rewind( pSrc );
  cson_parse( &root, cson_data_src_Blob, pSrc, NULL, pInfo );
  return root;
}

/*
** Implements the cson_data_dest_f() interface and outputs the data to
** cgi_append_content(). pState is ignored.
**/
int cson_data_dest_cgi(void * pState, void const * src, unsigned int n){
  cgi_append_content( (char const *)src, (int)n );
  return 0;
}

/*
** Returns a string in the form FOSSIL-XXXX, where XXXX is a
** left-zero-padded value of code. The returned buffer is static, and
** must be copied if needed for later.  The returned value will always
** be 11 bytes long (not including the trailing NUL byte).
**
** In practice we will only ever call this one time per app execution
** when constructing the JSON response envelope, so the static buffer
** "shouldn't" be a problem.
**
*/
char const * json_rc_cstr( int code ){
  enum { BufSize = 12 };
  static char buf[BufSize] = {'F','O','S','S','I','L','-',0};
  assert((code >= 1000) && (code <= 9999) && "Invalid Fossil/JSON code.");
  sprintf(buf+7,"%04d", code);
  return buf;
}

/*
** Adds v to the API-internal cleanup mechanism. key must be a unique
** key for the given element. Adding another item with that key may
** free the previous one (depending on its reference count). If
** freeOnError is true then v is passed to cson_value_free() if the
** key cannot be inserted, otherweise ownership of v is not changed on
** error. Failure to insert a key may be caused by any of the
** following:
**
** - Allocation error.
** - g.json.gc.o is NULL
** - key is NULL or empty.
**
** Returns 0 on success.
**
** On success, ownership of v is transfered to (or shared with)
** g.json.gc, and v will be valid until that object is cleaned up or
** its key is replaced via another call to this function.
*/
int json_gc_add( char const * key, cson_value * v, char freeOnError ){
  int const rc = cson_object_set( g.json.gc.o, key, v );
  assert( NULL != g.json.gc.o );
  if( (0 != rc) && freeOnError ){
    cson_value_free( v );
  }
  return rc;
}


/*
** Returns the value of json_rc_cstr(code) as a new JSON
** string, which is owned by the caller and must eventually
** be cson_value_free()d or transfered to a JSON container.
*/
cson_value * json_rc_string( int code ){
  return cson_value_new_string( json_rc_cstr(code), 11 );
}

cson_value * json_new_string( char const * str ){
  return str
    ? cson_value_new_string(str,strlen(str))
    : NULL;
}

cson_value * json_new_int( int v ){
  return cson_value_new_integer((cson_int_t)v);
}

/*
** Gets a POST/POST.payload/GET/COOKIE/ENV value. The returned memory
** is owned by the g.json object (one of its sub-objects). Returns
** NULL if no match is found.
**
** ENV means the system environment (getenv()).
**
** Precedence: POST.payload, GET/COOKIE/non-JSON POST, JSON POST, ENV.
**
** FIXME: the precedence SHOULD be: GET, POST.payload, POST, COOKIE,
** ENV, but the amalgamation of the GET/POST vars makes it difficult
** for me to do that. Since fossil only uses one cookie, cookie
** precedence isn't a real/high-priority problem.
*/
cson_value * json_getenv( char const * zKey ){
  cson_value * rc;
  rc = g.json.reqPayload.o
    ? cson_object_get( g.json.reqPayload.o, zKey )
    : NULL;
  if(rc){
    return rc;
  }
  rc = cson_object_get( g.json.param.o, zKey );
  if( rc ){
    return rc;
  }
  rc = cson_object_get( g.json.post.o, zKey );
  if(rc){
    return rc;
  }else{
    char const * cv = PD(zKey,NULL);
    if(!cv){
      /* reminder to self: in CLI mode i'd like to try
         find_option(zKey,NULL,XYZ) here, but we don't have a sane
         default for the XYZ param here.
      */
      cv = getenv(zKey);
    }
    if(cv){/*transform it to JSON for later use.*/
      /* use sscanf() to figure out if it's an int,
         and transform it to JSON int if it is.
      */
      int intVal = -1;
      char endOfIntCheck;
      int const scanRc = sscanf(cv,"%d%c",&intVal, &endOfIntCheck)
        /* The %c bit there is to make sure that we don't accept 123x
          as a number. sscanf() returns the number of tokens
          successfully parsed, so an RC of 1 will be correct for "123"
          but "123x" will have RC==2.
        */
        ;
      if(1==scanRc){
        json_setenv( zKey, cson_value_new_integer(intVal) );
      }else{
        rc = cson_value_new_string(cv,strlen(cv));
        json_setenv( zKey, rc );
      }
      return rc;
    }
  }
  return NULL;
}

/*
** Wrapper around json_getenv() which...
**
** If it finds a value and that value is-a JSON number or is a string
** which looks like an integer or is-a JSON bool/null then it is
** converted to an int. If none of those apply then dflt is returned.
*/
int json_getenv_int(char const * pKey, int dflt ){
  cson_value const * v = json_getenv(pKey);
  if(!v){
    return dflt;
  }else if( cson_value_is_number(v) ){
    return (int)cson_value_get_integer(v);
  }else if( cson_value_is_string(v) ){
    char const * sv = cson_string_cstr(cson_value_get_string(v));
    assert( (NULL!=sv) && "This is quite unexpected." );
    return sv ? atoi(sv) : dflt;
  }else if( cson_value_is_bool(v) ){
    return cson_value_get_bool(v) ? 1 : 0;
  }else if( cson_value_is_null(v) ){
    return 0;
  }else{
    /* we should arguably treat JSON null as 0. */
    return dflt;
  }
}


/*
** Wrapper around json_getenv() which tries to evaluate a payload/env
** value as a boolean. Uses mostly the same logic as
** json_getenv_int(), with the addition that string values which
** either start with a digit 1..9 or the letters [tT] are considered
** to be true. If this function cannot find a matching key/value then
** dflt is returned. e.g. if it finds the key but the value is-a
** Object then dftl is returned.
**
** If an entry is found, this function guarantees that it will return
** either 0 or 1, as opposed to "0 or non-zero", so that clients can
** pass a different value as dflt. Thus they can use, e.g. -1 to know
** whether or not this function found a match (it will return -1 in
** that case).
*/
char json_getenv_bool(char const * pKey, char dflt ){
  cson_value const * v = json_getenv(pKey);
  if(!v){
    return dflt;
  }else if( cson_value_is_number(v) ){
    return cson_value_get_integer(v) ? 1 : 0;
  }else if( cson_value_is_string(v) ){
    char const * sv = cson_string_cstr(cson_value_get_string(v));
    if(!*sv || ('0'==*sv)){
      return 0;
    }else{
      return (('t'==*sv) || ('T'==*sv)
              || (('1'<=*sv) && ('9'>=*sv)))
        ? 1 : 0;
    }
  }else if( cson_value_is_bool(v) ){
    return cson_value_get_bool(v) ? 1 : 0;
  }else if( cson_value_is_null(v) ){
    return 0;
  }else{
    return dflt;
  }
}

/*
** Returns the string form of a json_getenv() value, but ONLY If that
** value is-a String. Non-strings are not converted to strings for
** this purpose. Returned memory is owned by g.json or fossil and is
** valid until end-of-app or the given key is replaced in fossil's
** internals via cgi_replace_parameter() and friends or json_setenv().
*/
char const * json_getenv_cstr( char const * zKey ){
  return cson_value_get_cstr( json_getenv(zKey) );
}

/*
** An extended form of find_option() which tries to look up a combo
** GET/POST/CLI argument.
**
** zKey must be the GET/POST parameter key. zCLILong must be the "long
** form" CLI flag (NULL means to use zKey). zCLIShort may be NUL or
** the "short form" CLI flag.
**
** On error (no match found) NULL is returned.
**
** This ONLY works for String JSON/GET/CLI values, not JSON
** booleans and whatnot.
*/
char const * json_find_option_cstr(char const * zKey,
                                   char const * zCLILong,
                                   char const * zCLIShort){
  char const * rc = NULL;
  assert(NULL != zKey);
  if(!g.isHTTP){
    rc = find_option(zCLILong ? zCLILong : zKey,
                     zCLIShort, 1);
  }
  if(!rc && fossil_is_json()){
    rc = json_getenv_cstr(zKey);
  }
  return rc;
}

/*
** The boolean equivalent of json_find_option_cstr().
** If the option is not found, dftl is returned.
*/
char json_find_option_bool(char const * zKey,
                           char const * zCLILong,
                           char const * zCLIShort,
                           char dflt ){
  char rc = -1;
  if(!g.isHTTP){
    if(NULL != find_option(zCLILong ? zCLILong : zKey,
                           zCLIShort, 0)){
      rc = 1;
    }
  }
  if((-1==rc) && fossil_is_json()){
    rc = json_getenv_bool(zKey,-1);
  }
  return (-1==rc) ? dflt : rc;
}

/*
** The integer equivalent of json_find_option_cstr().
** If the option is not found, dftl is returned.
*/
int json_find_option_int(char const * zKey,
                         char const * zCLILong,
                         char const * zCLIShort,
                         int dflt ){
  enum { Magic = -947 };
  int rc = Magic;
  if(!g.isHTTP){
    char const * opt = find_option(zCLILong ? zCLILong : zKey,
                                   zCLIShort, 1);
    if(NULL!=opt){
      rc = atoi(opt);
    }
  }
  if(Magic==rc){
    rc = json_getenv_int(zKey,Magic);
  }
  return (Magic==rc) ? dflt : rc;
}


/*
** Adds v to g.json.param.o using the given key. May cause any prior
** item with that key to be destroyed (depends on current reference
** count for that value). On success, transfers (or shares) ownership
** of v to (or with) g.json.param.o. On error ownership of v is not
** modified.
*/
int json_setenv( char const * zKey, cson_value * v ){
  return cson_object_set( g.json.param.o, zKey, v );
}

/*
** Guesses a RESPONSE Content-Type value based (primarily) on the
** HTTP_ACCEPT header.
**
** It will try to figure out if the client can support
** application/json or application/javascript, and will fall back to
** text/plain if it cannot figure out anything more specific.
**
** Returned memory is static and immutable.
*/
char const * json_guess_content_type(){
  char const * cset;
  char doUtf8;
  cset = PD("HTTP_ACCEPT_CHARSET",NULL);
  doUtf8 = ((NULL == cset) || (NULL!=strstr("utf-8",cset)))
    ? 1 : 0;
  if( g.json.jsonp ){
    return doUtf8
      ? "application/javascript; charset=utf-8"
      : "application/javascript";
  }else{
    /*
      Content-type
      
      If the browser does not sent an ACCEPT for application/json
      then we fall back to text/plain.
    */
    char const * cstr;
    cstr = PD("HTTP_ACCEPT",NULL);
    if( NULL == cstr ){
      return doUtf8
        ? "application/json; charset=utf-8"
        : "application/json";
    }else{
      if( strstr( cstr, "application/json" )
          || strstr( cstr, "*/*" ) ){
        return doUtf8
          ? "application/json; charset=utf-8"
          : "application/json";
      }else{
        return "text/plain";
      }
    }
  }
}

/*
** Sends pResponse to the output stream as the response object.  This
** function does no validation of pResponse except to assert() that it
** is not NULL. The caller is responsible for ensuring that it meets
** API response envelope conventions.
**
** In CLI mode pResponse is sent to stdout immediately. In HTTP
** mode pResponse replaces any current CGI content but cgi_reply()
** is not called to flush the output.
**
** If g.json.jsonp is not NULL then the content type is set to
** application/javascript and the output is wrapped in a jsonp
** wrapper.
*/
void json_send_response( cson_value const * pResponse ){
  assert( NULL != pResponse );
  if( g.isHTTP ){
    cgi_reset_content();
    if( g.json.jsonp ){
      cgi_printf("%s(",g.json.jsonp);
    }
    cson_output( pResponse, cson_data_dest_cgi, NULL, &g.json.outOpt );
    if( g.json.jsonp ){
      cgi_append_content(")",1);
    }
  }else{/*CLI mode*/
    if( g.json.jsonp ){
      fprintf(stdout,"%s(",g.json.jsonp);
    }
    cson_output_FILE( pResponse, stdout, &g.json.outOpt );
    if( g.json.jsonp ){
      fwrite(")\n", 2, 1, stdout);
    }
  }
}

/*
** Returns the current request's JSON authentication token, or NULL if
** none is found. The token's memory is owned by (or shared with)
** g.json.
**
** If an auth token is found in the GET/POST request data then fossil
** is given that data for use in authentication for this
** session. i.e. the GET/POST data overrides fossil's authentication
** cookie value (if any) and also works with clients which do not
** support cookies.
**
** Must be called once before login_check_credentials() is called or
** we will not be able to replace fossil's internal idea of the auth
** info in time (and future changes to that state may cause unexpected
** results).
**
** The result of this call are cached for future calls.
*/
cson_value * json_auth_token(){
  if( !g.json.authToken ){
    /* Try to get an authorization token from GET parameter, POSTed
       JSON, or fossil cookie (in that order). */
    g.json.authToken = json_getenv(FossilJsonKeys.authToken);
    if(g.json.authToken
       && cson_value_is_string(g.json.authToken)
       && !PD(login_cookie_name(),NULL)){
      /* tell fossil to use this login info.

      FIXME?: because the JSON bits don't carry around
      login_cookie_name(), there is a potential login hijacking
      window here. We may need to change the JSON auth token to be
      in the form: login_cookie_name()=...

      Then again, the hardened cookie value helps ensure that
      only a proper key/value match is valid.
      */
      cgi_replace_parameter( login_cookie_name(), cson_value_get_cstr(g.json.authToken) );
    }else if( g.isHTTP ){
      /* try fossil's conventional cookie. */
      /* Reminder: chicken/egg scenario regarding db access in CLI
         mode because login_cookie_name() needs the db. CLI
         mode does not use any authentication, so we don't need
         to support it here.
      */
      char const * zCookie = P(login_cookie_name());
      if( zCookie && *zCookie ){
        /* Transfer fossil's cookie to JSON for downstream convenience... */
        cson_value * v = cson_value_new_string(zCookie, strlen(zCookie));
        if(0 == json_gc_add( FossilJsonKeys.authToken, v, 1 )){
          g.json.authToken = v;
        }
      }
    }
  }
  return g.json.authToken;
}

/*
** IFF json.reqPayload.o is not NULL then this returns
** cson_object_get(json.reqPayload.o,pKey), else it returns NULL.
**
** The returned value is owned by (or shared with) json.reqPayload.v.
*/
cson_value * json_req_payload_get(char const *pKey){
  return g.json.reqPayload.o
    ? cson_object_get(g.json.reqPayload.o,pKey)
    : NULL;
}

/*
** Initializes some JSON bits which need to be initialized relatively
** early on. It should only be called from cgi_init() or
** json_cmd_top() (early on in those functions).
**
** Initializes g.json.gc and g.json.param. This code does not (and
** must not) rely on any of the fossil environment having been set
** up. e.g. it must not use cgi_parameter() and friends because this
** must be called before those data are initialized.
*/
void json_main_bootstrap(){
  cson_value * v;
  assert( (NULL == g.json.gc.v) && "cgi_json_bootstrap() was called twice!" );

  /* g.json.gc is our "garbage collector" - where we put JSON values
     which need a long lifetime but don't have a logical parent to put
     them in.
  */
  v = cson_value_new_object();
  g.json.gc.v = v;
  g.json.gc.o = cson_value_get_object(v);

  /*
    g.json.param holds the JSONized counterpart of fossil's
    cgi_parameter_xxx() family of data. We store them as JSON, as
    opposed to using fossil's data directly, because we can retain
    full type information for data this way (as opposed to it always
    being of type string).
  */
  v = cson_value_new_object();
  g.json.param.v = v;
  g.json.param.o = cson_value_get_object(v);
  json_gc_add("$PARAMS", v, 1);
}

/*
** Appends a warning object to the (pending) JSON response.
**
** Code must be a FSL_JSON_W_xxx value from the FossilJsonCodes enum.
**
** A Warning object has this JSON structure:
**
** { "code":integer, "text":"string" }
**
** But the text part is optional.
**
** If msg is non-NULL and not empty then it is used as the "text"
** property's value. It is copied, and need not refer to static
** memory.
**
** CURRENTLY this code only allows a given warning code to be
** added one time, and elides subsequent warnings. The intention
** is to remove that burden from loops which produce warnings.
**
** FIXME: if msg is NULL then use a standard string for
** the given code. If !*msg then elide the "text" property,
** for consistency with how json_err() works.
*/
void json_warn( int code, char const * msg ){
  cson_value * objV = NULL;
  cson_object * obj = NULL;
  assert( (code>FSL_JSON_W_START)
          && (code<FSL_JSON_W_END)
          && "Invalid warning code.");
  if(!g.json.warnings.v){
    g.json.warnings.v = cson_value_new_array();
    assert((NULL != g.json.warnings.v) && "Alloc error.");
    g.json.warnings.a = cson_value_get_array(g.json.warnings.v);
    json_gc_add("$WARNINGS",g.json.warnings.v,0);
  }
  objV = cson_value_new_object();
  assert((NULL != objV) && "Alloc error.");
  cson_array_append(g.json.warnings.a, objV);
  obj = cson_value_get_object(objV);
  cson_object_set(obj,"code",cson_value_new_integer(code));
  if(msg && *msg){
    /* FIXME: treat NULL msg as standard warning message for
       the code, but we don't have those yet.
    */
    cson_object_set(obj,"text",cson_value_new_string(msg,strlen(msg)));
  }
}

/*
** Splits zStr (which must not be NULL) into tokens separated by the
** given separator character. If doDeHttp is true then each element
** will be passed through dehttpize(), otherwise they are used
** as-is. Note that tokenization happens before dehttpize(),
** which is significant if the ENcoded tokens might contain the
** separator character.
**
** Each new element is appended to the given target array object,
** which must not be NULL and ownership of it is not changed by this
** call.
**
** On success, returns the number of tokens _encountered_. On error a
** NEGATIVE number is returned - its absolute value is the number of
** tokens encountered (i.e. it reveals which token in zStr was
** problematic).
**
** Achtung: leading and trailing whitespace of elements are elided.
**
** Achtung: empty elements will be skipped, meaning consecutive empty
** elements are collapsed.
*/
int json_string_split( char const * zStr,
                       char separator,
                       char doDeHttp,
                       cson_array * target ){
  char const * p = zStr /* current byte */;
  char const * head  /* current start-of-token */;
  unsigned int len = 0   /* current token's length */;
  int rc = 0   /* return code (number of added elements)*/;
  char skipWs = fossil_isspace(separator) ? 0 : 1;
  assert( zStr && target );
  while( fossil_isspace(*p) ){
    ++p;
  }
  head = p;
  for( ; ; ++p){
    if( !*p || (separator == *p) ){
      if( len ){/* append head..(head+len) as next array
                   element. */
        cson_value * part = NULL;
        char * zPart = NULL;
        ++rc;
        assert( head != p );
        zPart = (char*)malloc(len+1);
        assert( (zPart != NULL) && "malloc failure" );
        memcpy(zPart, head, len);
        zPart[len] = 0;
        if(doDeHttp){
          dehttpize(zPart);
        }
        if( *zPart ){ /* should only fail if someone manages to url-encoded a NUL byte */
          part = cson_value_new_string(zPart, strlen(zPart));
          if( 0 != cson_array_append( target, part ) ){
            cson_value_free(part);
            rc = -rc;
            break;
          }
        }else{
          assert(0 && "i didn't think this was possible!");
          fprintf(stderr,"%s:%d: My God! It's full of stars!\n",
                  __FILE__, __LINE__);
          fossil_exit(1)
            /* Not fossil_panic() b/c this code needs to be able to
              run before some of the fossil/json bits are initialized,
              and fossil_panic() calls into the JSON API.
            */
            ;
        }
        free(zPart);
        len = 0;
      }
      if( !*p ){
        break;
      }
      head = p+1;
      while( *head && fossil_isspace(*head) ){
        ++head;
        ++p;
      }
      if(!*head){
        break;
      }
      continue;
    }
    ++len;
  }
  return rc;
}

/*
** Wrapper around json_string_split(), taking the same first 3
** parameters as this function, but returns the results as a JSON
** Array (if splitting produced tokens) or NULL (if splitting failed
** in any way or produced no tokens).
**
** The returned value is owned by the caller. If not NULL then it
** _will_ have a JSON type of Array or Null.
*/
cson_value * json_string_split2( char const * zStr,
                                 char separator,
                                 char doDeHttp ){
  cson_value * v = cson_value_new_array();
  cson_array * a = cson_value_get_array(v);
  int rc = json_string_split( zStr, separator, doDeHttp, a );
  if( 0 == rc ){
    cson_value_free(v);
    v = NULL;
  }else if(rc<0){
    cson_value_free(v);
    v = NULL;
  }
  return v;
}


/*
** Performs some common initialization of JSON-related state.  Must be
** called by the json_page_top() and json_cmd_top() dispatching
** functions to set up the JSON stat used by the dispatched functions.
**
** Implicitly sets up the login information state in CGI mode, but
** does not perform any permissions checking. It _might_ (haven't
** tested this) die with an error if an auth cookie is malformed.
**
** This must be called by the top-level JSON command dispatching code
** before they do any work.
**
** This must only be called once, or an assertion may be triggered.
*/
static void json_mode_bootstrap(){
  static char once = 0  /* guard against multiple runs */;
  char const * zPath = P("PATH_INFO");
  cson_value * pathSplit = NULL;
  assert( (0==once) && "json_mode_bootstrap() called too many times!");
  if( once ){
    return;
  }else{
    once = 1;
  }
  g.json.isJsonMode = 1;
  g.json.resultCode = 0;
  g.json.cmd.offset = -1;
  g.json.jsonp = PD("jsonp",NULL)
    /* FIXME: do some sanity checking on g.json.jsonp and ignore it
       if it is not halfway reasonable.
    */
    ;
  if( !g.isHTTP && g.fullHttpReply ){
    /* workaround for server mode, so we see it as CGI mode. */
    g.isHTTP = 1;
  }

  if(g.isHTTP){
    cgi_set_content_type(json_guess_content_type())
      /* reminder: must be done after g.json.jsonp is initialized */
      ;
#if 0
    /* Calling this seems to trigger an SQLITE_MISUSE warning???
       Maybe it's not legal to set the logger more than once?
    */
    sqlite3_config(SQLITE_CONFIG_LOG, NULL, 0)
        /* avoids debug messages on stderr in JSON mode */
        ;
#endif
  }

  g.json.cmd.v = cson_value_new_array();
  g.json.cmd.a = cson_value_get_array(g.json.cmd.v);
  json_gc_add( FossilJsonKeys.commandPath, g.json.cmd.v, 1 );
  /*
    The following if/else block translates the PATH_INFO path (in
    CLI/server modes) or g.argv (CLI mode) into an internal list so
    that we can simplify command dispatching later on.

    Note that translating g.argv this way is overkill but allows us to
    avoid CLI-only special-case handling in other code, e.g.
    json_command_arg().
  */
  if( zPath ){/* Either CGI or server mode... */
    /* Translate PATH_INFO into JSON array for later convenience. */
    json_string_split(zPath, '/', 1, g.json.cmd.a);
  }else{/* assume CLI mode */
    int i;
    char const * arg;
    cson_value * part;
    for(i = 1/*skip argv[0]*/; i < g.argc; ++i ){
      arg = g.argv[i];
      if( !arg || !*arg ){
        continue;
      }
      if('-' == *arg){
        /* workaround to skip CLI args so that
           json_command_arg() does not see them.
           This assumes that all arguments come LAST
           on the command line.
        */
        break;
      }
      part = cson_value_new_string(arg,strlen(arg));
      cson_array_append(g.json.cmd.a, part);
    }
  }

  while(!g.isHTTP){
    /* Simulate JSON POST data via input file.  Pedantic reminder:
       error handling does not honor user-supplied g.json.outOpt
       because outOpt cannot (generically) be configured until after
       POST-reading is finished.
    */
    FILE * inFile = NULL;
    char const * jfile = find_option("json-input",NULL,1);
    if(!jfile || !*jfile){
      break;
    }
    inFile = (0==strcmp("-",jfile))
      ? stdin
      : fopen(jfile,"rb");
    if(!inFile){
      g.json.resultCode = FSL_JSON_E_FILE_OPEN_FAILED;
      fossil_fatal("Could not open JSON file [%s].",jfile)
        /* Does not return. */
        ;
    }
    cgi_parse_POST_JSON(inFile, 0);
    if( stdin != inFile ){
      fclose(inFile);
    }
    break;
  }
  
  /* g.json.reqPayload exists only to simplify some of our access to
     the request payload. We currently only use this in the context of
     Object payloads, not Arrays, strings, etc.
  */
  g.json.reqPayload.v = cson_object_get( g.json.post.o, FossilJsonKeys.payload );
  if( g.json.reqPayload.v ){
    g.json.reqPayload.o = cson_value_get_object( g.json.reqPayload.v )
        /* g.json.reqPayload.o may legally be NULL, which means only that
           g.json.reqPayload.v is-not-a Object.
        */;
  }

  /* Anything which needs json_getenv() and friends should go after
     this point.
  */

  if(1 == cson_array_length_get(g.json.cmd.a)){
    /* special case: if we're at the top path, look for
       a "command" request arg which specifies which command
       to run.
    */
    char const * cmd = json_getenv_cstr("command");
    if(cmd){
      json_string_split(cmd, '/', 0, g.json.cmd.a);
      g.json.cmd.commandStr = cmd;
    }
  }

  
  if(!g.json.jsonp){
    g.json.jsonp = json_find_option_cstr("jsonp",NULL,NULL);
  }
  if(!g.isHTTP){
    g.json.errorDetailParanoia = 0 /*disable error code dumb-down for CLI mode*/;
  }

  {/* set up JSON output formatting options. */
    int indent = -1;
    char const * indentStr = NULL;
    indent = json_find_option_int("indent",NULL,"I",-1);
    g.json.outOpt.indentation = (0>indent)
      ? (g.isHTTP ? 0 : 1)
      : (unsigned char)indent;
    g.json.outOpt.addNewline = g.isHTTP
      ? 0
      : (g.json.jsonp ? 0 : 1);
  }

  if( g.isHTTP ){
    json_auth_token()/* will copy our auth token, if any, to fossil's
                        core, which we need before we call
                        login_check_credentials(). */;
    login_check_credentials()/* populates g.perm */;
  }
  else{
    db_find_and_open_repository(OPEN_ANY_SCHEMA,0);
  }
}

/*
** Returns the ndx'th item in the "command path", where index 0 is the
** position of the "json" part of the path. Returns NULL if ndx is out
** of bounds or there is no "json" path element.
**
** In CLI mode the "path" is the list of arguments (skipping argv[0]).
** In server/CGI modes the path is taken from PATH_INFO.
**
** The returned bytes are owned by g.json.cmd.v and _may_ be
** invalidated if that object is modified (depending on how it is
** modified).
**
** Note that CLI options are not included in the command path. Use
** find_option() to get those.
**
*/
char const * json_command_arg(unsigned char ndx){
  cson_array * ar = g.json.cmd.a;
  assert((NULL!=ar) && "Internal error. Was json_mode_bootstrap() called?");
  assert((g.argc>1) && "Internal error - we never should have gotten this far.");
  if( g.json.cmd.offset < 0 ){
    /* first-time setup. */
    short i = 0;
#define NEXT cson_string_cstr(          \
                 cson_value_get_string( \
                   cson_array_get(ar,i) \
                   ))
    char const * tok = NEXT;
    while( tok ){
      if( !g.isHTTP/*workaround for "abbreviated name" in CLI mode*/
          ? (0==strcmp(g.argv[1],tok))
          : (0==strncmp("json",tok,4))
          ){
        g.json.cmd.offset = i;
        break;
      }
      ++i;
      tok = NEXT;
    }
  }
#undef NEXT
  if(g.json.cmd.offset < 0){
    return NULL;
  }else{
    ndx = g.json.cmd.offset + ndx;
    return cson_string_cstr(cson_value_get_string(cson_array_get( ar, g.json.cmd.offset + ndx )));
  }
}

/*
** If g.json.reqPayload.o is NULL then NULL is returned, else the
** given property is searched for in the request payload.  If found it
** is returned. The returned value is owned by (or shares ownership
** with) g.json, and must NOT be cson_value_free()'d by the
** caller.
*/
cson_value * json_payload_property( char const * key ){
  return g.json.reqPayload.o ?
    cson_object_get( g.json.reqPayload.o, key )
    : NULL;
}


/* Returns the C-string form of json_auth_token(), or NULL
** if json_auth_token() returns NULL.
*/
char const * json_auth_token_cstr(){
  return cson_value_get_cstr( json_auth_token() );
}

/*
** Returns the JsonPageDef with the given name, or NULL if no match is
** found.
**
** head must be a pointer to an array of JsonPageDefs in which the
** last entry has a NULL name.
*/
JsonPageDef const * json_handler_for_name( char const * name, JsonPageDef const * head ){
  JsonPageDef const * pageDef = head;
  assert( head != NULL );
  if(name && *name) for( ; pageDef->name; ++pageDef ){
    if( 0 == strcmp(name, pageDef->name) ){
      return pageDef;
    }
  }
  return NULL;
}

/*
** Given a Fossil/JSON result code, this function "dumbs it down"
** according to the current value of g.json.errorDetailParanoia. The
** dumbed-down value is returned.
**
** This function assert()s that code is in the inclusive range 0 to
** 9999.
**
** Note that WARNING codes (1..999) are never dumbed down.
**
*/
static int json_dumbdown_rc( int code ){
  if(!code || ((code>FSL_JSON_W_START) && (code>FSL_JSON_W_END))){
    return code;
  }else{
    int modulo = 0;
    assert((code >= 1000) && (code <= 9999) && "Invalid Fossil/JSON code.");
    switch( g.json.errorDetailParanoia ){
      case 1: modulo = 10; break;
      case 2: modulo = 100; break;
      case 3: modulo = 1000; break;
      default: break;
    }
    if( modulo ) code = code - (code % modulo);
    return code;
  }
}

/*
** Convenience routine which converts a Julian time value into a Unix
** Epoch timestamp. Requires the db, so this cannot be used before the
** repo is opened (will trigger a fatal error in db_xxx()).
*/
cson_value * json_julian_to_timestamp(double j){
  return cson_value_new_integer((cson_int_t)
           db_int64(0,"SELECT strftime('%%s',%lf)",j)
                                );
}

/*
** Returns a timestamp value.
*/
cson_int_t json_timestamp(){
  return (cson_int_t)time(0);
}
/*
** Returns a new JSON value (owned by the caller) representing
** a timestamp. If timeVal is < 0 then time(0) is used to fetch
** the time, else timeVal is used as-is
*/
cson_value * json_new_timestamp(cson_int_t timeVal){
  return cson_value_new_integer((timeVal<0) ? (cson_int_t)time(0) : timeVal);
}

/*
** Creates a new Fossil/JSON response envelope skeleton.  It is owned
** by the caller, who must eventually free it using cson_value_free(),
** or add it to a cson container to transfer ownership. Returns NULL
** on error.
**
** If payload is not NULL and resultCode is 0 then it is set as the
** "payload" property of the returned object.  If resultCode is
** non-zero and payload is not NULL then this function calls
** cson_value_free(payload) and does not insert the payload into the
** response. In either case, onwership of payload is transfered to
** this function.
**
** pMsg is an optional message string property (resultText) of the
** response. If resultCode is non-0 and pMsg is NULL then
** json_err_str() is used to get the error string. The caller may
** provide his own or may use an empty string to suppress the
** resultText property.
**
*/
cson_value * json_create_response( int resultCode,
                                   char const * pMsg,
                                   cson_value * payload){
  cson_value * v = NULL;
  cson_value * tmp = NULL;
  cson_object * o = NULL;
  int rc;
  resultCode = json_dumbdown_rc(resultCode);
  v = cson_value_new_object();
  o = cson_value_get_object(v);
  if( ! o ) return NULL;
#define SET(K) if(!tmp) goto cleanup; \
  rc = cson_object_set( o, K, tmp ); \
  if(rc) do{\
    cson_value_free(tmp); \
    tmp = NULL; \
    goto cleanup; \
  }while(0)

  tmp = cson_value_new_string(MANIFEST_UUID,strlen(MANIFEST_UUID));
  SET("fossil");

  {
    tmp = json_new_timestamp(-1);
    SET(FossilJsonKeys.timestamp);
  }

  if( 0 != resultCode ){
    if( ! pMsg ){
      pMsg = g.zErrMsg;
      if(!pMsg){
        pMsg = json_err_str(resultCode);
      }
    }
    tmp = json_new_string(json_rc_cstr(resultCode));
    SET(FossilJsonKeys.resultCode);
  }

  if( pMsg && *pMsg ){
    tmp = cson_value_new_string(pMsg, strlen(pMsg));
    SET(FossilJsonKeys.resultText);
  }

  if(g.json.cmd.commandStr){
    tmp = json_new_string(g.json.cmd.commandStr);
    SET("command");
  }
  
  tmp = json_getenv(FossilJsonKeys.requestId);
  if( tmp ) cson_object_set( o, FossilJsonKeys.requestId, tmp );

  if(0){/* these are only intended for my own testing...*/
    if(g.json.cmd.v){
      tmp = g.json.cmd.v;
      SET("$commandPath");
    }
    if(g.json.param.v){
      tmp = g.json.param.v;
      SET("$params");
    }
    if(0){/*Only for debuggering, add some info to the response.*/
      tmp = cson_value_new_integer( g.json.cmd.offset );
      cson_object_set( o, "cmd.offset", tmp );
      cson_object_set( o, "isCGI", cson_value_new_bool( g.isHTTP ) );
    }
  }

  if(HAS_TIMER){
    /* This is, philosophically speaking, not quite the right place
       for ending the timer, but this is the one function which all of
       the JSON exit paths use (and they call it after processing,
       just before they end).
    */
    double span;
    span = END_TIMER;
    /* i'm actually seeing sub-ms runtimes in some tests, but a time of
       0 is "just wrong", so we'll bump that up to 1ms.
    */
    cson_object_set(o,"procTimeMs", cson_value_new_integer((cson_int_t)((span>0.0)?span:1)));
  }
  if(g.json.warnings.v){
    tmp = g.json.warnings.v;
    SET("warnings");
  }
  
  /* Only add the payload to SUCCESS responses. Else delete it. */
  if( NULL != payload ){
    if( resultCode ){
      cson_value_free(payload);
      payload = NULL;
    }else{
      tmp = payload;
      SET(FossilJsonKeys.payload);
    }
  }

#undef SET
  goto ok;
  cleanup:
  cson_value_free(v);
  v = NULL;
  ok:
  return v;
}

/*
** Outputs a JSON error response to either the cgi_xxx() family of
** buffers (in CGI/server mode) or stdout (in CLI mode). If rc is 0
** then g.json.resultCode is used. If that is also 0 then the "Unknown
** Error" code is used.
**
** If g.isHTTP then the generated JSON error response object replaces
** any currently buffered page output. Because the output goes via
** the cgi_xxx() family of functions, this function inherits any
** compression which fossil does for its output.
**
** If alsoOutput is true AND g.isHTTP then cgi_reply() is called to
** flush the output (and headers). Generally only do this if you are
** about to call exit().
**
** !g.isHTTP then alsoOutput is ignored and all output is sent to
** stdout immediately.
**
** For generating the resultCode property: if msg is not NULL then it
** is used as-is. If it is NULL then g.zErrMsg is checked, and if that
** is NULL then json_err_str(code) is used.
*/
void json_err( int code, char const * msg, char alsoOutput ){
  int rc = code ? code : (g.json.resultCode
                          ? g.json.resultCode
                          : FSL_JSON_E_UNKNOWN);
  cson_value * resp = NULL;
  rc = json_dumbdown_rc(rc);
  if( rc && !msg ){
    msg = g.zErrMsg;
    if(!msg){
      msg = json_err_str(rc);
    }
  }
  resp = json_create_response(rc, msg, NULL);
  if(!resp){
    /* about the only error case here is out-of-memory. DO NOT
       call fossil_panic() here because that calls this function.
    */
    fprintf(stderr, "%s: Fatal error: could not allocate "
            "response object.\n", fossil_nameofexe());
    fossil_exit(1);
  }
  if( g.isHTTP ){
    if(alsoOutput){
      json_send_response(resp);
    }else{
      /* almost a duplicate of json_send_response() :( */
      cgi_reset_content();
      if( g.json.jsonp ){
        cgi_printf("%s(",g.json.jsonp);
      }
      cson_output( resp, cson_data_dest_cgi, NULL, &g.json.outOpt );
      if( g.json.jsonp ){
        cgi_append_content(")",1);
      }
    }
  }else{
    json_send_response(resp);
  }
  cson_value_free(resp);
}

/*
** Sets g.json.resultCode and g.zErrMsg, but does not report the error
** via json_err(). Returns the code passed to it.
**
** code must be in the inclusive range 1000..9999.
*/
int json_set_err( int code, char const * fmt, ... ){
  assert( (code>=1000) && (code<=9999) );
  free(g.zErrMsg);
  g.json.resultCode = code;
  if(!fmt || !*fmt){
    g.zErrMsg = mprintf("%s", json_err_str(code));
  }else{
    va_list vargs;
    va_start(vargs,fmt);
    char * msg = vmprintf(fmt, vargs);
    va_end(vargs);
    g.zErrMsg = msg;
  }
  return code;
}

/*
** Iterates through a prepared SELECT statement and converts each row
** to a JSON object. If pTgt is not NULL then it must be-a Array
** object and this function will return pTgt. If pTgt is NULL then a
** new Array object is created and returned (owned by the
** caller). Each row of pStmt is converted to an Object and appended
** to the array.
*/
cson_value * json_stmt_to_array_of_obj(Stmt *pStmt,
                                       cson_value * pTgt){
  cson_value * v = pTgt ? pTgt : cson_value_new_array();
  cson_array * a = cson_value_get_array(pTgt ? pTgt : v);
  char const * warnMsg = NULL;
  assert( NULL != a );
  while( (SQLITE_ROW==db_step(pStmt)) ){
    cson_value * row = cson_sqlite3_row_to_object(pStmt->pStmt);
    if(!row && !warnMsg){
      warnMsg = "Could not convert at least one result row to JSON.";
      continue;
    }
    cson_array_append(a, row);
  }
  if(warnMsg){
    json_warn( FSL_JSON_W_ROW_TO_JSON_FAILED, warnMsg );
  }
  return v;  
}


/*
** If the given rid has any tags associated with it, this function
** returns a JSON Array containing the tag names, else it returns
** NULL.
**
** See info_tags_of_checkin() for more details (this is simply a JSON
** wrapper for that function).
*/
cson_value * json_tags_for_rid(int rid, char propagatingOnly){
  cson_value * v = NULL;
  char * tags = info_tags_of_checkin(rid, propagatingOnly);
  if(tags){
    if(*tags){
      v = json_string_split2(tags,',',0);
    }
    free(tags);
  }
  return v;  
}


/*
** /json/version implementation.
**
** Returns the payload object (owned by the caller).
*/
cson_value * json_page_version(){
  cson_value * jval = NULL;
  cson_object * jobj = NULL;
  jval = cson_value_new_object();
  jobj = cson_value_get_object(jval);
#define FSET(X,K) cson_object_set( jobj, K, cson_value_new_string(X,strlen(X)))
  FSET(MANIFEST_UUID,"manifestUuid");
  FSET(MANIFEST_VERSION,"manifestVersion");
  FSET(MANIFEST_DATE,"manifestDate");
  FSET(MANIFEST_YEAR,"manifestYear");
  FSET(RELEASE_VERSION,"releaseVersion");
#undef FSET
  cson_object_set( jobj, "releaseVersionNumber",
                   cson_value_new_integer(RELEASE_VERSION_NUMBER) );
  cson_object_set( jobj, "resultCodeParanoiaLevel",
                   cson_value_new_integer(g.json.errorDetailParanoia) );
  return jval;
}


/*
** Implementation for /json/cap
**
** Returned object contains details about the "capabilities" of the
** current user (what he may/may not do).
**
** This is primarily intended for debuggering, but may have
** a use in client code. (?)
*/
cson_value * json_page_cap(){
  cson_value * payload = cson_value_new_object();
  cson_value * sub = cson_value_new_object();
  Stmt q;
  cson_object * obj = cson_value_get_object(payload);
  db_prepare(&q, "SELECT login, cap FROM user WHERE uid=%d", g.userUid);
  if( db_step(&q)==SQLITE_ROW ){
    /* reminder: we don't use g.zLogin because it's 0 for the guest
       user and the HTML UI appears to currently allow the name to be
       changed (but doing so would break other code). */
    char const * str = (char const *)sqlite3_column_text(q.pStmt,0);
    if( str ){
      cson_object_set( obj, "name",
                       cson_value_new_string(str,strlen(str)) );
    }
    str = (char const *)sqlite3_column_text(q.pStmt,1);
    if( str ){
      cson_object_set( obj, "capabilities",
                       cson_value_new_string(str,strlen(str)) );
    }
  }
  db_finalize(&q);
  cson_object_set( obj, "permissionFlags", sub );
  obj = cson_value_get_object(sub);

#define ADD(X,K) cson_object_set(obj, K, cson_value_new_bool(g.perm.X))
  ADD(Setup,"setup");
  ADD(Admin,"admin");
  ADD(Delete,"delete");
  ADD(Password,"password");
  ADD(Query,"query"); /* don't think this one is actually used */
  ADD(Write,"checkin");
  ADD(Read,"checkout");
  ADD(History,"history");
  ADD(Clone,"clone");
  ADD(RdWiki,"readWiki");
  ADD(NewWiki,"createWiki");
  ADD(ApndWiki,"appendWiki");
  ADD(WrWiki,"editWiki");
  ADD(RdTkt,"readTicket");
  ADD(NewTkt,"createTicket");
  ADD(ApndTkt,"appendTicket");
  ADD(WrTkt,"editTicket");
  ADD(Attach,"attachFile");
  ADD(TktFmt,"createTicketReport");
  ADD(RdAddr,"readPrivate");
  ADD(Zip,"zip");
  ADD(Private,"xferPrivate");
#undef ADD
  return payload;
}

/*
** Implementation of the /json/stat page/command.
**
*/
cson_value * json_page_stat(){
  i64 t, fsize;
  int n, m;
  int full;
  const char *zDb;
  enum { BufLen = 1000 };
  char zBuf[BufLen];
  cson_value * jv = NULL;
  cson_object * jo = NULL;
  cson_value * jv2 = NULL;
  cson_object * jo2 = NULL;
  if( !g.perm.Read ){
    json_set_err(FSL_JSON_E_DENIED,
                 "Requires 'o' permissions.");
    return NULL;
  }
  full = json_find_option_bool("full",NULL,"f",0);
#define SETBUF(O,K) cson_object_set(O, K, cson_value_new_string(zBuf, strlen(zBuf)));

  jv = cson_value_new_object();
  jo = cson_value_get_object(jv);

  sqlite3_snprintf(BufLen, zBuf, db_get("project-name",""));
  SETBUF(jo, "projectName");
  /* FIXME: don't include project-description until we ensure that
     zBuf will always be big enough. We "should" replace zBuf
     with a blob for this purpose.
  */
  fsize = file_size(g.zRepositoryName);
  cson_object_set(jo, "repositorySize", cson_value_new_integer((cson_int_t)fsize));

  if(full){
    n = db_int(0, "SELECT count(*) FROM blob");
    m = db_int(0, "SELECT count(*) FROM delta");
    cson_object_set(jo, "blobCount", cson_value_new_integer((cson_int_t)n));
    cson_object_set(jo, "deltaCount", cson_value_new_integer((cson_int_t)m));
    if( n>0 ){
      int a, b;
      Stmt q;
      db_prepare(&q, "SELECT total(size), avg(size), max(size)"
                 " FROM blob WHERE size>0");
      db_step(&q);
      t = db_column_int64(&q, 0);
      cson_object_set(jo, "uncompressedArtifactSize",
                      cson_value_new_integer((cson_int_t)t));
      cson_object_set(jo, "averageArtifactSize",
                      cson_value_new_integer((cson_int_t)db_column_int(&q, 1)));
      cson_object_set(jo, "maxArtifactSize",
                      cson_value_new_integer((cson_int_t)db_column_int(&q, 2)));
      db_finalize(&q);
      if( t/fsize < 5 ){
        b = 10;
        fsize /= 10;
      }else{
        b = 1;
      }
      a = t/fsize;
      sqlite3_snprintf(BufLen,zBuf, "%d:%d", a, b);
      SETBUF(jo, "compressionRatio");
    }
    n = db_int(0, "SELECT count(distinct mid) FROM mlink /*scan*/");
    cson_object_set(jo, "checkinCount", cson_value_new_integer((cson_int_t)n));
    n = db_int(0, "SELECT count(*) FROM filename /*scan*/");
    cson_object_set(jo, "fileCount", cson_value_new_integer((cson_int_t)n));
    n = db_int(0, "SELECT count(*) FROM tag  /*scan*/"
               " WHERE +tagname GLOB 'wiki-*'");
    cson_object_set(jo, "wikiPageCount", cson_value_new_integer((cson_int_t)n));
    n = db_int(0, "SELECT count(*) FROM tag  /*scan*/"
               " WHERE +tagname GLOB 'tkt-*'");
    cson_object_set(jo, "ticketCount", cson_value_new_integer((cson_int_t)n));
  }/*full*/
  n = db_int(0, "SELECT julianday('now') - (SELECT min(mtime) FROM event)"
                " + 0.99");
  cson_object_set(jo, "ageDays", cson_value_new_integer((cson_int_t)n));
  cson_object_set(jo, "ageYears", cson_value_new_double(n/365.24));
  sqlite3_snprintf(BufLen, zBuf, db_get("project-code",""));
  SETBUF(jo, "projectCode");
  sqlite3_snprintf(BufLen, zBuf, db_get("server-code",""));
  SETBUF(jo, "serverCode");
  cson_object_set(jo, "compiler", cson_value_new_string(COMPILER_NAME, strlen(COMPILER_NAME)));

  jv2 = cson_value_new_object();
  jo2 = cson_value_get_object(jv2);
  cson_object_set(jo, "sqlite", jv2);
  sqlite3_snprintf(BufLen, zBuf, "%.19s [%.10s] (%s)",
                   SQLITE_SOURCE_ID, &SQLITE_SOURCE_ID[20], SQLITE_VERSION);
  SETBUF(jo2, "version");
  zDb = db_name("repository");
  cson_object_set(jo2, "pageCount", cson_value_new_integer((cson_int_t)db_int(0, "PRAGMA %s.page_count", zDb)));
  cson_object_set(jo2, "pageSize", cson_value_new_integer((cson_int_t)db_int(0, "PRAGMA %s.page_size", zDb)));
  cson_object_set(jo2, "freeList", cson_value_new_integer((cson_int_t)db_int(0, "PRAGMA %s.freelist_count", zDb)));
  sqlite3_snprintf(BufLen, zBuf, "%s", db_text(0, "PRAGMA %s.encoding", zDb));
  SETBUF(jo2, "encoding");
  sqlite3_snprintf(BufLen, zBuf, "%s", db_text(0, "PRAGMA %s.journal_mode", zDb));
  cson_object_set(jo2, "journalMode", *zBuf ? cson_value_new_string(zBuf, strlen(zBuf)) : cson_value_null());
  return jv;
#undef SETBUF
}



static cson_value * json_user_list();
static cson_value * json_user_get();
#if 0
static cson_value * json_user_create();
static cson_value * json_user_edit();
#endif

/*
** Mapping of /json/user/XXX commands/paths to callbacks.
*/
static const JsonPageDef JsonPageDefs_User[] = {
{"create", json_page_nyi, 1},
{"edit", json_page_nyi, 1},
{"get", json_user_get, 0},
{"list", json_user_list, 0},
/* Last entry MUST have a NULL name. */
{NULL,NULL,0}
};


cson_value * json_page_dispatch_helper(JsonPageDef const * pages){
  JsonPageDef const * def;
  char const * cmd = json_command_arg(1+g.json.dispatchDepth);
  assert( NULL != pages );
  if( ! cmd ){
    json_set_err(FSL_JSON_E_MISSING_ARGS, "No subcommand specified.");
    return NULL;
  }
  def = json_handler_for_name( cmd, pages );
  if(!def){
    g.json.resultCode = FSL_JSON_E_UNKNOWN_COMMAND;
    return NULL;
  }
  else{
    ++g.json.dispatchDepth;
    return (*def->func)();
  }
}

/*
** Implements the /json/user family of pages/commands.
**
*/
static cson_value * json_page_user(){
  return json_page_dispatch_helper(&JsonPageDefs_User[0]);
}


/*
** Impl of /json/rebuild. Requires admin previleges.
*/
static cson_value * json_page_rebuild(){
  if( !g.perm.Admin ){
    json_set_err(FSL_JSON_E_DENIED,"Requires 'a' privileges.");
    return NULL;
  }else{
  /* Reminder: the db_xxx() ops "should" fail via the fossil core
     error handlers, which will cause a JSON error and exit(). i.e. we
     don't handle the errors here. TODO: confirm that all these db
     routine fail gracefully in JSON mode.

     On large repos (e.g. fossil's) this operation is likely to take
     longer than the client timeout, which will cause it to fail (but
     it's sqlite3, so it'll fail gracefully).
  */
    db_close(1);
    db_open_repository(g.zRepositoryName);
    db_begin_transaction();
    rebuild_db(0, 0, 0);
    db_end_transaction(0);
    return NULL;
  }
}

/*
** Impl of /json/user/list. Requires admin rights.
*/
static cson_value * json_user_list(){
  cson_value * payV = NULL;
  Stmt q;
  if(!g.perm.Admin){
    g.json.resultCode = FSL_JSON_E_DENIED;
    return NULL;
  }
  db_prepare(&q,"SELECT uid AS uid,"
             " login AS name,"
             " cap AS capabilities,"
             " info AS info,"
             " mtime AS mtime"
             " FROM user ORDER BY login");
  payV = json_stmt_to_array_of_obj(&q, NULL);
  db_finalize(&q);
  if(NULL == payV){
    json_set_err(FSL_JSON_E_UNKNOWN,
                 "Could not convert user list to JSON.");
  }
  return payV;  
}

/*
** Impl of /json/user/get. Requires admin rights.
*/
static cson_value * json_user_get(){
  cson_value * payV = NULL;
  char const * pUser = NULL;
  Stmt q;
  if(!g.perm.Admin){
    g.json.resultCode = FSL_JSON_E_DENIED;
    return NULL;
  }
  pUser = json_command_arg(g.json.dispatchDepth+1);
  if( g.isHTTP && (!pUser || !*pUser) ){
    pUser = json_getenv_cstr("name")
      /* ACHTUNG: fossil apparently internally sets name=user/get/XYZ
         if we pass the name as part of the path, so we check the path
         _before_ checking for name=XYZ.
      */;
  }
  if(!pUser || !*pUser){
    json_set_err(FSL_JSON_E_MISSING_ARGS,"Missing 'name' property.");
    return NULL;
  }
  db_prepare(&q,"SELECT uid AS uid,"
             " login AS name,"
             " cap AS capabilities,"
             " info AS info,"
             " mtime AS mtime"
             " FROM user"
             " WHERE login=%Q",
             pUser);
  if( (SQLITE_ROW == db_step(&q)) ){
    payV = cson_sqlite3_row_to_object(q.pStmt);
    if(!payV){
      json_set_err(FSL_JSON_E_UNKNOWN,"Could not convert user row to JSON.");
    }
  }else{
    json_set_err(FSL_JSON_E_RESOURCE_NOT_FOUND,"User not found.");
  }
  db_finalize(&q);
  return payV;  
}

/* Impl in json_login.c. */
cson_value * json_page_anon_password();
/* Impl in json_login.c. */
cson_value * json_page_login();
/* Impl in json_login.c. */
cson_value * json_page_logout();
/* Impl in json_artifact.c. */
cson_value * json_page_artifact();
/* Impl in json_branch.c. */
cson_value * json_page_branch();

/*
** Mapping of names to JSON pages/commands.  Each name is a subpath of
** /json (in CGI mode) or a subcommand of the json command in CLI mode
*/
static const JsonPageDef JsonPageDefs[] = {
/* please keep alphabetically sorted (case-insensitive) for maintenance reasons. */
{"anonymousPassword",json_page_anon_password, 1},
{"artifact", json_page_artifact, 0},
{"branch", json_page_branch,0},
{"cap", json_page_cap, 0},
{"dir", json_page_nyi, 0},
{"HAI",json_page_version,0},
{"login",json_page_login,1},
{"logout",json_page_logout,1},
{"rebuild",json_page_rebuild,0},
{"stat",json_page_stat,0},
{"tag", json_page_nyi,0},
{"ticket", json_page_nyi,0},
{"timeline", json_page_timeline,0},
{"user",json_page_user,0},
{"version",json_page_version,0},
{"whoami",json_page_whoami,0/*FIXME: work in CLI mode*/},
{"wiki",json_page_wiki,0},
/* Last entry MUST have a NULL name. */
{NULL,NULL,0}
};


/*
** Mapping of /json/ticket/XXX commands/paths to callbacks.
*/
static const JsonPageDef JsonPageDefs_Ticket[] = {
{"get", json_page_nyi, 0},
{"list", json_page_nyi, 0},
{"save", json_page_nyi, 1},
{"create", json_page_nyi, 1},
/* Last entry MUST have a NULL name. */
{NULL,NULL,0}
};

/*
** Mapping of /json/artifact/XXX commands/paths to callbacks.
*/
static const JsonPageDef JsonPageDefs_Artifact[] = {
{"vinfo", json_page_nyi, 0},
{"finfo", json_page_nyi, 0},
/* Last entry MUST have a NULL name. */
{NULL,NULL,0}
};

/*
** Mapping of /json/tag/XXX commands/paths to callbacks.
*/
static const JsonPageDef JsonPageDefs_Tag[] = {
{"list", json_page_nyi, 0},
{"create", json_page_nyi, 1},
/* Last entry MUST have a NULL name. */
{NULL,NULL,0}
};


/*
** WEBPAGE: json
**
** Pages under /json/... must be entered into JsonPageDefs.
** This function dispatches them, and is the HTTP equivalent of
** json_cmd_top().
*/
void json_page_top(void){
  int rc = FSL_JSON_E_UNKNOWN_COMMAND;
  char const * cmd;
  cson_value * payload = NULL;
  JsonPageDef const * pageDef = NULL;
  BEGIN_TIMER;
  json_mode_bootstrap();
  cmd = json_command_arg(1);
  /*cgi_printf("{\"cmd\":\"%s\"}\n",cmd); return;*/
  pageDef = json_handler_for_name(cmd,&JsonPageDefs[0]);
  if( ! pageDef ){
    json_err( FSL_JSON_E_UNKNOWN_COMMAND, NULL, 0 );
    return;
  }else if( pageDef->runMode < 0 /*CLI only*/){
    rc = FSL_JSON_E_WRONG_MODE;
  }else{
    rc = 0;
    g.json.dispatchDepth = 1;
    payload = (*pageDef->func)();
  }
  if( g.json.resultCode ){
    json_err(g.json.resultCode, NULL, 0);
  }else{
    cson_value * root = json_create_response(rc, NULL, payload);
    json_send_response(root);
    cson_value_free(root);
  }
}

/*
** This function dispatches json commands and is the CLI equivalent of
** json_page_top().
**
** COMMAND: json
**
** Usage: %fossil json SUBCOMMAND
**
** The commands include:
**
**   branch
**   cap
**   stat
**   timeline
**   version (alias: HAI)
**   wiki
**
**
** TODOs:
**
**   tag
**   ticket
**   ...
**
*/
void json_cmd_top(void){
  char const * cmd = NULL;
  int rc = FSL_JSON_E_UNKNOWN_COMMAND;
  cson_value * payload = NULL;
  JsonPageDef const * pageDef;
  BEGIN_TIMER;
  memset( &g.perm, 0xff, sizeof(g.perm) )
    /* In CLI mode fossil does not use permissions
       and they all default to false. We enable them
       here because (A) fossil doesn't use them in local
       mode but (B) having them set gives us one less
       difference in the CLI/CGI/Server-mode JSON
       handling.
    */
    ;
  json_main_bootstrap();
  json_mode_bootstrap();
  if( 2 > cson_array_length_get(g.json.cmd.a) ){
    goto usage;
  }
  db_find_and_open_repository(0, 0);
#if 0
  json_warn(FSL_JSON_W_ROW_TO_JSON_FAILED, "Just testing.");
  json_warn(FSL_JSON_W_ROW_TO_JSON_FAILED, "Just testing again.");
#endif
  cmd = json_command_arg(1);
  if( !cmd || !*cmd ){
    goto usage;
  }
  pageDef = json_handler_for_name(cmd,&JsonPageDefs[0]);
  if( ! pageDef ){
    json_err( FSL_JSON_E_UNKNOWN_COMMAND, NULL, 1 );
    return;
  }else if( pageDef->runMode > 0 /*HTTP only*/){
    rc = FSL_JSON_E_WRONG_MODE;
  }else{
    rc = 0;
    g.json.dispatchDepth = 1;
    payload = (*pageDef->func)();
  }
  if( g.json.resultCode ){
    json_err(g.json.resultCode, NULL, 1);
  }else{
    payload = json_create_response(rc, NULL, payload);
    json_send_response(payload);
    cson_value_free( payload );
    if((0 != rc) && !g.isHTTP){
      /* FIXME: we need a way of passing this error back
         up to the routine which called this callback.
         e.g. add g.errCode.
      */
      fossil_exit(1);
    }
  }
  return;
  usage:
  usage("subcommand");
}

#undef BITSET_BYTEFOR
#undef BITSET_SET
#undef BITSET_UNSET
#undef BITSET_GET
#undef BITSET_TOGGLE