LOAD DATA LOCAL INFILE implemetation

[anton-kotenko]

1. Make queries like "load data local infile" to work in their normal way (data is taken from file on client's machine and sent to mysql server)
2. Make possible to use instance of nodejs's Buffer class instead of file, as source of rows.
   (feature similar to postgresql's COPY FROM STDIN). Sometimes it's more convenient that using temporary file or pipe.

This feature requires update in query and querySync interface (without breaking compatibility)
Examples:

var buf = new Buffer('1,2,3\n4,5,6');
conn.querySync('LOAD DATA LOCAL INFILE \'some_meaningless_name\' INTO TABLE test_table FIELDS TERMINATED BY \',\'  LINES TERMINATED BY \'\n''', buf);

[Sannis]

Thanks! I will merge this after fix failed tests in master.

And I prefer to cherry-pick commits manually to avoid duplications from your fork. You can help me with this if you have some time.

[Sannis]

Мы не должны вызвать mysql_thread_init(); перед этой строкой?

[Sannis]

Хм, всё равно это восстановление и если дальше будет использоваться стандартный обработчик - в его инициализации функция всё равно вызовется. Но всё равно странный порядок вызовов:

mysql_set_local_infile_default(conn);
...
mysql_thread_init();
mysql_set_local_infile_default(conn);

Так что пусть будет.

[anton-kotenko]

В случае стандартного load infile (без буфера), последовательность становиться

mysql_thread_init();
mysql_set_local_infile_default(conn);
mysql_thread_end(); //чтобы не было memory leak  если в libuv в пуле потоков  какой то закроется

в случае наличия буфера последовательность

mysql_set_local_infile_handler(...) 
// не зависит от mysql_thread_init
mysql_set_local_infile_default(conn); 
// в принципе не очень хорошо, так как стандартная реализация зависит от
// mysql_thread_init(), но не очень и плохо, так как в следующем запросе, 
// в случае надобности, будет вызван mysql_thread_init

[anton-kotenko]

As I understand, you are busy with some bugs with Travis, but if there will be any problems with this pull request, please, write. I would like to help, as this feature is critical for me (it makes massive inserts into database 3-5 times faster that simple inserts)

[Sannis]

As I understand, you are busy with some bugs with Travis

Done!

but if there will be any problems with this pull request, please, write.

Can you rebase your branch onto upstream master?
I'm not commit-counter hunter, but it will be great to keep history more linear for such simple project.
I hope to have some time for this next evening.

[Sannis]

Now this is second call in this function that causes dead lock.

[Sannis]

Good work, Anton!

[anton-kotenko]

Thanks, don't forget to update code, because I've just fixed bug (my, incorrect rebase) with locks (double lock in EIO_Query);

[Sannis]

Good to merge — The Travis build passed

Now passes :)