Cubus Arts Blog

@ Mihai Târnovan
Mihai Târnovan

Installing mysql gem with Percona XtraDB

Mihai Târnovan

After installing the mysql gem, we were seeing uninitialized constant MysqlCompat::MysqlRes errors all over the place, although the gem installed fine.

This was on FreeBSD 8.1. Turns out Percona XtraDB binary doesn’t include the mysql client libs, so you need to install those as well. On FreeBSD just do

pkg_add -r mysql51-client

Then reinstall the gem:

env ARCHFLAGS="-arch x86_64" gem i mysql -- --with-mysql-config=/usr/local/percona-xtradb/bin/mysql_config

Mihai Târnovan

Ruby Eval stacktrace position information

Mihai Târnovan

We’ve been often annoyed by warnings or exceptions like this: (eval):231: some error or warning. They make it very difficult to pinpoint the exact location of the problem.

The problem is usually cause by module_eval being called without file and line information. There’s an interesting write-up about eval and position information on Ola Bini’s blog. One of his readers observed that when using module_eval like this you’ll get the incorrect line number:

However, I noticed that if you write the heredoc inline you’ll get the correct line in the backtrace:

__LINE__ + 1 accounts for the line with the module_eval itself. I tested this with Ruby 1.8.6 and Ruby 1.9.1 and it applies for both, tough 1.9 at least puts more lines in the backtrace with the position of the module_eval call.

Passing __FILE__ and __LINE__ is better than skipping them, but can lead to false line numbers unless you use the 2nd form with the inlined heredoc.

Mihai Târnovan

Dear typographers

Mihai Târnovan

The romanian language has some specific diacritics which can radically change the meaning of a word.

The problem

For example țânc vs. tanc, tata vs. țâța vs. țața (kid vs. tank, father vs. breast vs. aunt). The problem with this characters and their implementation in keyboard layouts, charsets and fonts, is well known and has been explained in almost academic detail and extremely well documented manner by Cristian Secară.

Microsoft’s “solution” and its consequences

When Microsoft implemented the first romanian keyboard layout, they mapped the keys which should have produced these characters to similar turkish characters instead. They could have fixed this as early as Windows 2000 (around then the romanian standard SR 13411:1999 was already in place). Instead, they did nothing until 2007.

By then, their core fonts were already the de facto standard for web typography and everyone had fonts installed on their computer which lacked characters for lowercase and uppercase ș and ț (Unicode code points U+0218, U+0219, U+021A and U+021B).

Fixing the problem and causing new ones

It took Microsoft until 2006 to get it right. They released a patch which added the missing characters to Times New Roman, Arial and Verdana. When they released Windows Vista, they still didn’t get it right for all fonts – Verdana was only fixed by a later patch.

Ok, so Microsoft release a patch for XP and did the right thing with Vista. So everything’s fine, right ? Well, not quite. If you use the Romanian keyboard layout on Vista or MacOS, you will write text containing the proper Unicode characters. But a vast majority of romanian users don’t have these characters in their fonts, because they didn’t apply the font patch. So for them, a ș written with the proper romanian character appears as a square.

Possible solutions

  1. Use fonts that never had this problem in the first place.
    • Palatino Linotype
    • Tahoma

    This is of course too limiting for practical purposes, especially in web design.

  2. Use old keyboard layout.

    I still haven’t figured out how to do this for MacOS without custom keyboard layouts, but it should be fairly easy to do this for Vista and Linux (use Romanian legacy or a similar keyboard layout). You’ll write text with turkish diacritics, but at least all your visitors will be able to display them correctly.

  3. Serve two versions of content, one for Windows XP (and older) users with turkish characters, and one for all the others.

    Due to the complexity of this solution, I’m not sure this is a good idea. (we do this for travelbox with a Rack filter — but that was implemented more to learn about Rack than as a long term solution for this problem)