Geonames: The only terrible choice we have

Table of Contents

  1. Loading your data
  2. Associating your data
  3. The mysterious admin1Codes.txt

When it comes to getting an accurate, up to date database of cities, towns, and what have you, really the best we can hope for is Geonames. They manage to keep things updated fairly regularly, however there also seems to be a slight, just a slight, massive tidal wave of garbage all mixed in.

As they say on the BBC: I'm on a journey… to figure out how to get the cities, associated provinces, and postal codes from these data dumps, and associate them together. You'd think it'd be easy, you'd think the CSVs have all of the proper association information there, easy to manage, and of course that's what a sane person would think.

Well, I've got a surprise for you.

I'm not going to go super in-depth here all tutorial-style, but rather just outline some things I had to deal with and the solutions I came up with, in hopes they will help you, because they sure as hell were not easy to find.

Loading your data

Loading it is fairly straight forward for a lot of people, but some people it isn't quite clear. A few things need to be considered:

  1. Regular utf8 columns in MySQL will not allow you to import alternative names and some other stuff
  2. In some cases even utf8mb4 columns aren't "good enough" to import some alternative names because the alternative names themselves aren't "good enough" and are malformed Unicode.
  3. Most alternative names are either identical, misspellings, or bizarre stuff like airport codes, misspellings in other languages, what seems to be some sort of sub-typing (like one for Moscow was "wsa     MOW" , what the hell that means, I have no idea), and you'd think the city in the native language/script would come first, but it usually doesn't, if at all. Though I'm sure you'll enjoy all of the Gothic unicode script of names which are absolutely necessary since we live sometime in the Early Middle Ages. 
  4. My suggestion is to just ignore the alternative names completely, but if you insist on importing them, use varbinary or blob.
  5. There are tons and tons of duplicates, with our Moscow example alone there are currently 7 versions of the exact same city, in the same country, and yes the feature type is the same.

The easiest way to do this is create a table in MySQL identical to the schema they lay out for whatever dump you have, but maybe varbinary instead of varchar (as mentioned above). In some cases you may want to add a little on the end too, some columns go over the documented width.

From your MySQL client:

> use yourdatabase;
> create `yourtablewiththeschemathatsthesame`;
> load data infile '/~/allCountries.txt' INTO TABLE `whatevertable`

This general idea works with all of the dumps so long as the table is exactly the same.

Whoops, hold on, for some reason there are needless backslashes escaping the tabs from time to time, so you have to go in and replace those otherwise MySQL will freak out. Are these backslashes a part of the names of some cities or places, perhaps a part of a strange alphabet?

LOL, of course they're not, they're just randomly there.

Associating your data

With the locations, you'd think with name like geonameid that would mean other dumps like postal would match up with it in some way, or they'd easily reference administrative divisions by ID… come on, you know that wouldn't do something that logical.

Geonames has the most horrendously, implausibly baffling, most terribly undocumented method of associating data I've seen in years; it's a wonder anyone uses it. It's made worse by the fact that those who have figured it out (which to me seems on par with deciphering Egyptian Hieroglyphs) essentially keep it a secret, as if posting about it on a forum will cause the Stasi to burst in their door and take their families away.

How the data are associated is pretty damn goofy and poorly thought out:

  • Countries are referenced by their ISO-3166-1 code, at least this is somewhat consistent, in fact GB is used rather than UK, so don't confuse with the almost identical country top-level domains. SX and XK are also in use for the new nations of South Sudan and Kosovo, since as of writing this, they do not have codes yet.
  • Cities and states/provinces: You'd think they'd follow the logical geonameid integer logic, but at this point I guess I don't need to say it's far more moronic than that. Cities are associated with provinces by the `country` and `admin1` columns.  So what are these `admin1` values? In some cases like US states they make sense, they're the ISO standard abbreviations, in some cases they're area codes, and in some cases they're various other things. (See bottom of this post about where to get the admin1 values from).
  • Postal codes are matched primarily by country and admin1 , admin2 (when they're properly filled, sometimes not), and the "place_name," however if you're expecting place_name to match the city in spelling in the other dump (such as O'Brien in the locations dump to be also O'Brien in the postal codes dump), you are sorely mistaken, yet again. Instead spelling changes are haphazard, so it can be "O Brien" or "OBrien" but never the expected "O'Brien," I mean jeeze, it's almost like they're running a psychological experiment on us.
  • As for other stuff, I don't have any information, my only concern was getting associated cities, provinces/states, countries, and postal codes, but it doesn't take a massive leap of faith to guess they're all messed up too.
The mysterious admin1Codes.txt

So you want to associate your cities with the appropriate province/state? "lol, kiss my ass" say the administrators of Geonames, they make it basically impossible.

How you say?

So where do they keep the states/provinces? No where, they deleted them. Apparently they felt it was "confusing for a lot of users" because of various issues, so instead of fixing those, they simply remove admin1Codes.txt from their web site, but continue to reference it in the official readme.txt for how to import the data.

Brilliant! That's exactly how not to confuse people, reference a non-existent file which contains important information on which cities belong to which provinces/states in countries, just delete it and replace it with nothing, because having anything at all would be confusing.

I don't recommend using the one linked by "marc," it's missing a lot and it's badly encoded.

I went ahead and recreated a new admin1Codes.txt, I based it on combining several versions of admin1Codes.txt I found, being certain to actually use the same encoding type through the whole file and making sure all of the cities listed have a code.

admin1Codes.txt with names in "plain-text" Unicode
admin1Codes.txt with names in hex (for easy transport, but they're still Unicode bytes)

If you import this, be certain to use utf8mb4_unicode_ci columns, and I explain at great length in my post: Better Unicode support for MySQL (including emoji).

Disclaimer: I can't vouch for the accuracy of the names or whether or not obsolete ones still exist, but hell, it's better than providing you with nothing.

In case you are interested in contributing to this list, an example of the missing codes can be found here: missing.txt, if you manage to match them up, let me know. A few I started to manually do, but since I spent so much time on this already, I stopped. I'm also considering starting an API service like Geonames that uses proper association. I realize the job of getting accurate data isn't easy, but if they've got people manually entering things to properly set it (and they do) then why in the hell don't they maintain a logical, static association?

Better Unicode support for MySQL (including emoji)

When it comes to character support I think the only thing that should ever be used is Unicode. That's right, I said it. However, when it comes to support in MySQL, things get a little bit murky.

I never had too much of an issue using plain ol' utf8_general_ci, however when trying to add language support for Gothic (tested because it's rare), I ran into a serious issue:

Incorrect string value: '\xF0\x90\x8C\xB0\xF0\x90...'
Source: brainstuck.com
Source: brainstuck.com

Son of a …

This issue is caused by the fact UTF-8 in MySQL isn't fully supported by utf8 the character set, it only supports a maximum of 3-byte characters. If you want something more realistic you're going to need to have at least MySQL 5.5.3 and you're going to have to use utf8mb4 not regular utf8. Yes, seriously.

Make sure you read through all this before trying anything, because there are edge issues, especially with indices (indexes) which you may need to consider.

Also back up your data first.

Updating Database

I'll be working under the assumption that you want your entire database to be utf8mb4, but if you don't then you'll have to adjust a bit, but seriously reconsider joining the 21st century if you're not using unicode. I'm also assuming you want case insensitive text, and if you don't, replace utf8mb4_unicode_ci with utf8mb4_bin — most people want case insensitive text in most cases.

Update the default character set and collation for your database:

ALTER DATABASE `mydatabase` CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;
Updating Tables

First we need to change the default character set, this way when you add new columns in the future, or whatever, you don't need to worry about adding all of the character set specification:

ALTER TABLE `mytable1` DEFAULT CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;
Updating Table Columns

Now, you can convert one column at a time, and this may be what you wish to do if you require different character sets for your CHAR, VARCHAR, and TEXT columns, here's how you do that:

ALTER TABLE `mytable1` CHANGE `mycolumn` `mycolumn` VARCHAR(64) CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci NOT NULL DEFAULT  '';

Now obviously you're going to want to make sure that you're converting to the same column type and length, etc, the above is for example only and if you copy/paste it, you may screw up your column schema. Essentially you're just using ALTER TABLE CHANGE on the column in order to change the character set to utf8mb4 and collation to utf8mb4_unicode_ci.

If on the other hand you just want to change the entire the entire table at once, you can do:

ALTER TABLE `mytable1` CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;
Updating Table Indices

When changing the character type, you may run into this on InnoDB:

ERROR 1071 (42000): Specified key was too long; max key length is 767 bytes

or this on MyISAM:

ERROR 1071 (42000): Specified key was too long; max key length is 1000 bytes

Ah, CRAP!

Oh snippy snap, snap, there are solutions:

If you're using InnoDB on MySQL 5.6.3 or higher you can enable innodb_large_prefix in your MySQL config file (more information in manual here), but if you aren't you can take a few steps to work it out the old way:

  1. Make note of the conflicting index, it's going to likely be one which is something like VARCHAR(255) or an index across multiple columns which includes VARCHAR. Make note of the index name, type, and which column(s) it crosses.
  2. In my own scenario, I had a lot of columns which included some sort of VARCHAR(254) and ID which was binary(20).  Now it seems like 254+20 = 274, and hey that's less than 767 (or 1000) so what's the deal?Well, not so fast there, Professor.MySQL doesn't count literal bytes in VARCHAR when it comes to Unicode, rather potential Unicode bytes are themselves counted as a byte (wait, what?).So if the column is 254 and it's utf8 that means the actual potential length is literally (254 * 3) bytes, and with utf8mb4 it's (254 * 4). So really the length of the key you're trying to create is ((254 * 4) + 20).InnoDB only allows a maximum of 255 bytes for the column in an index with utf8 and 191 bytes for utf8mb4.So if you need the entire column indexed, you aren't going to want to change the character set for that column(s), and instead I recommend changing all others one by one (as seen in the Table Columns section) rather than trying to convert the entire table. However if you do not need the entire column to be index, and in certain cases I did not.Drop the index:
    ALTER TABLE `mytable1` DROP INDEX `theindex`;

    Then recreate it with the offending column(s) limited to 191:

    ALTER TABLE `mytable1` ADD INDEX `theindex` (`mycolumn` (191));

    or if across multiple columns (assuming mycolumnb is not utf8 for example):

    ALTER TABLE `mytable1` ADD INDEX `theindex` (`mycolumn` (191),`mycolumnb`);

As long as the indices are the same, and in the same column order, you should receive the same benefits for the indices without worrying about redoing your queries.

Additional Notes and Considerations

If a column is not being used for search and case insensitivity isn't an issue, instead of using CHAR or VARCHAR, I suggest using BINARY and VARBINARY. Not only is comparison vastly faster, but also there's less to worry about as far as character set issues go, i.e. they don't matter. Further also VARBINARY is literal length so the UTF-8 limitations described in the index section of this post do not apply, so you can get the full width for your index.

Additionally instead of using TEXT, use BLOB, for the same reasons, but also realise the same limitations apply, such as no fulltext searching.

In summation, if you don't need case sensitivity and you don't need fulltext search, consider BINARY, VARBINARY, and BLOB over CHAR, VARCHAR, and TEXT, it'll be a lot easier to deal with when it comes to Unicode.

You can learn more about this on my MySQL performance, using case insensitive columns post.

Database Connections

Depending on your programming language, you may need to specify when connecting which chartype to use (you can also, in most cases, specify this on configuration, see that section at the bottom), this usually can be done by sending this query  right after connection:

SET NAMES utf8mb4;
Configuration

You can edit your my.cnf (or my.ini on Windows) and make these changes to the appropriate sections of the configuration file (applicable to MySQL 5.6, older versions may need adjusted configuration):

[client]
default-character-set = utf8mb4

[mysql]
default-character-set = utf8mb4

[mysqld]
character-set-client-handshake = FALSE
character-set-server = utf8mb4
collation-server = utf8mb4_unicode_ci

MySQL performance, using case insensitive columns

Fast_train_(4712207733)When it comes to matching in queries, if you don't need to do case insensitive comparison (and not search) you should definitely consider using binary columns. They're much faster. Let's just jump right into benchmarks, then I'll talk more about why this is good and scenarios where it can be done (using MySQL 5.6.17 x64 Linux).

In this example, I use utf8mb4, which is what you probably should use if you wish to have better Unicode support (learn more here). Keep in mind though that the character set can be essentially anything, and these results are always in favor of comparing against case sensitive (or binary) columns.

Case Sensitive
SELECT BENCHMARK(500000000, _utf8mb4'foo' < _utf8mb4'zot' COLLATE utf8mb4_bin);

Result

1 row in set (5.88 sec)
Case Insensitive
SELECT BENCHMARK(500000000, _utf8mb4'foo' < _utf8mb4'zot' COLLATE utf8mb4_unicode_ci);

Result

1 row in set (16.27 sec)

As you can see, though, case sensitive was nearly three (2.77) times faster than case insensitive.

My suggestion to you is that unless you need the column to be case insensitive for comparison or search, you should certainly consider using the case insensitive character set, for example utf8_unicode_ci would be utf8_bin and utf8mb4_unicode_ci would be utf8mb4_bin.

For CHAR and VARCHAR other character sets also have case insensitive versions, and if not you can alter the column to be VARCHAR BINARY:

ALTER TABLE `mytable1` CHANGE `mycolumn` `mycolumn` VARCHAR(254) BINARY NOT NULL;

However, I don't think that's ideal, instead use VARBINARY, and in fact here's a little chart for which to change to:

CHAR BINARY
VARCHAR VARBINARY
TEXT BLOB
MEDIUMTEXT MEDIUMBLOB
LONGTEXT LONGBLOB

If you're not using fulltext search on MyISAM you might just want to consider switching your TEXT columns to BLOB anyway to avoid character set issues with Unicode.