Tag Archives: NexJ Server Express

SQLite: Can Using Unicode Collations be Improved? Attempt 2

In my quest to improve using Unicode collations in SQLite, I figured I needed to accomplish two things first. First, persist the locale for the DB between closing and opening it again. Second, change the default collation from BINARY (ASCII) to the a collator for the locale.

Here is the logic that I used for my implementation:
1. try to load the ICU collation for the given locale or use the system’s default locale
2. check if any tables other than the “locale table” exist. If they exist, exit
3. update the “locale table” with the name of the locale. (This allows reloading the collation when opening the DB)
4. using the ICU collation,  create an sqlite3 collation and make it the default collation for the DB

This is some of the source code for implementing the logic above:

int setDBDefaultLocale(sqlite3 * db, const char * const zLocale)
{
  int rc;
  sqlite3_stmt* stmt;
  UErrorCode status = U_ZERO_ERROR;
  UCollator *pUCollator= ucol_open(zLocale ? zLocale : uloc_getDefault(), &status);

  if(!U_SUCCESS(status))
  {
    return SQLITE_ERROR;
  }
  // check if any tables other than the "locale table" have been created.
  // if they exist, just quit, can not change default locale if a another table has already been created
  if ((rc = sqlite3_prepare_v2(db, "select * from sqlite_master where name != 'locale_table' and type = 'table'", 76, &stmt, 0)) != SQLITE_OK ||
      (rc = sqlite3_step(stmt)) != SQLITE_ROW)
  {
     sqlite3_finalize(stmt);

      return rc;
  }

  if (sqlite3_column_int(stmt, 0))
  {
    sqlite3_finalize(stmt);
    localeErrMsg(vdbe);

    return SQLITE_MISUSE;
  }
  // creates a locale table if one doesn't exist and updates it with the value of zLocale
  updateLocaleTable(zLocale ? zLocale : uloc_getDefault());
  // create the collation
  if ((rc = sqlite3_create_collation_v2(db, "LOCALE", SQLITE_UTF16,
  (void *)pUCollator, icuCollationColl, icuCollationDel)) != SQLITE_OK)
  {
    ucol_close(pUCollator);

    return rc;
  }

  // set the default collating sequence to the locale collation
  db->pDfltColl = sqlite3FindCollSeq(db, SQLITE_UTF16, "LOCALE", 0);

  return db->pDfltColl ? SQLITE_OK : SQLITE_ERROR;
}

With the implementation above, I was able to accomplish the first task i.e. persist the locale between connections. But when I tried to change the locale, it would throw an unexpected exception. After much digging, I figured out that the call to sqlite3FindCollseq was returning a buggy collating sequence. There was a mismatch between the DB encoding and the collating sequence encoding. That was solved by querying the encoding of the db object and passing that into the function call:

...
  db->pDfltColl = sqlite3FindCollSeq(db, ENC(db), "LOCALE", 0);
...

But as I thought about it, because this implementation always creates a table, this breaks SQLite’s functionality that depends on having an empty database. (e.g. setting the database encoding, pragma encoding = ‘UTF-16’)

As such, I am now investigating other ways to persist the locale. I am thinking of actually appending the locale to the header of the database file. One of my colleagues suggested persisting the locale outside of the database, using either another DB instance or a file. I am now pursuing the file header option. So far I have found out that I may need to modify the Pager (pager.c) implementation, to add more space to store the locale and the Btree (btree.c) implementation, specifically the sqlite3BtreeOpen method to both read in the locale. From what I found out so far, seems like nothing has to change to persist the locale to the disk.

So now the fun begins to implement the above. Here goes…

Advertisements
Tagged , , , , , , ,

NexJ Server Express: JSON Adapter

So my project has now been defined.
I will be working on extending the capabilities of NexJ Express Server: a J2EE server software that provides cost effective services for deploying highly-scalable, fault tolerant, distributed enterprise solutions. The server framework commonly referred to as “OpenCore” makes use of several open source projects like: Jboss, Apache Xalan, Apache Xerces, JavaMail, JMS and others.

So for my project, for the reminder of the summer, I am to develop a JSON Adapter to expose NexJ server Objects in JSON via a RESTful web service. Have a look at the project landing page on CDOT’s wiki.

current implementation:
server/nexj/xml/Person/104C083E8A4E8A4AD2BAF3011B6A393C97

<?xml version="1.0" encoding="utf-8"?>
<_Collection xmlns="http://www.nexj.com/xml" xmlns:xs="http://www.w3.org/2001/XMLSchema"
 xmlns:xi="http://www.w3.org/2001/XMLSchema-instance" xmlns:ns="http://www.nexj.com/xml">
 xi:type="Person">
 <_version>0</_version>
 <_oid>104C083E8A4E8A4AD2BAF3011B6A393C97</_oid>
 <_keys>primaryTelcom</_keys>
 <_keys>primaryLanguage</_keys>
 <_keys>initials</_keys>
 <_values xi:nil="true"/>
 <_values xi:nil="true"/>
 <_values xi:nil="true"/>
 <lastName>Bar
 <locking>0</locking>
 <firstName>Foo
 <company>
 <_version>0</_version>
 <_oid>10209394A0D92641BC9E0FCF36DDAD1C23</_oid>
 </company>
 </item>
</_Collection>

JSON implementation:
server/nexj/json/Person/104C083E8A4E8A4AD2BAF3011B6A393C97

{
“_Collection”: {
“item”: {
“_keys”: [“primaryTelcom”, “primaryLanguage”, “initials”],
“type”: “Person”,
“locking”: 0,
“_oid”: “104C083E8A4E8A4AD2BAF3011B6A393C97”,
“_version”: 0,
“_values”: [
{“nil”: true},
{“nil”: true},
{“nil”: true}
],
“firstName”: “Foo”,
“company”: {
“_oid”: “10209394A0D92641BC9E0FCF36DDAD1C23”,
“_version”: 0
},
“lastName”: “Bar”
},
}
}

Tagged ,