Friday, 30 November 2012

Link: Why engineers are grumpy

There have been many great articles on how to deal with IT-folks. This is another of them: The care and feeding of software engineers (or, why engineers are grumpy)

It explains: what makes engineers tick, what makes them grumpy and why, and also solutions to make it all Just Work. It covers both the strengths and weaknesses of engineers, and how to build on the former, while avoiding the latter.

I'll leave you with a few quotes:

In almost every other industry where things are built, it is expected that all requirements and details are agreed upon and finalized before building commences. Except in software. In software there’s “not enough time” to gather all the requirements ahead of time. The importance of moving quickly is hammered into us from day one. And so engineers learn to fill in the gaps left by product managers just to keep the project going. Product managers, of course, also have the reputation for changing their minds frequently, which means engineers assumptions are often invalidated partway through the process. Is it any wonder that software engineers tend to burn out quickly and change jobs frequently?
True priorities aren’t transient, they are static. The frequency with which people above us change their minds is incredibly frustrating for software engineers. We frequently stand ready to march into battle and just want a direction to march in. But if you tell us one day that we’re building a house and the next day that we’re building a car, you should expect some dissension in the ranks.
There are few phrases that anger software engineers more than, “I used to code.” ... If I were to ask LeBron James how much time he needs to prepare for a game, I’m sure he’d be amused if I disagreed because I played basketball in high school. Software engineers get the equivalent all the time.
We software engineers are an interesting bunch. There’s a definite personality that comes along with us, and we really do want to make the best thing possible. If you stop treating us like short-order cooks and start treating us like part of the creative process, you are likely to get much farther, faster than you would otherwise. The teams on which I’ve worked have all had varying degrees of friction caused by not understanding the engineers’ mindset and what drives them. It’s my sincere hope that this article will lead to better communication between engineers and those that they work with. It’s really not that hard. We’re all just people who want to feel like a part of the solution rather than a worker bee.

Saturday, 24 November 2012

Deleting un-named foreign keys in a migration

So, we use foreign-key constraints in our db. It's pretty annoying to get Rails to work nicely with that, but we have some basic helper methods (which I didn't write, so I cant share).

However I recently had some trouble dropping an old table. I kept getting these errors:

Mysql::Error: Error on rename of './mydbname/#sql-1ca8_f7842' to 
   './mydbname/my_table_name' (errno: 150): DROP INDEX `my_table_name_idx3` 
   ON my_table_name

I discovered that this was because it had foreign-key constraints. The awful error message is itself listed as a bug on mysql... but the real problem is that it doesn't want to drop an index I've asked it - because there's a foreign-key constraint on the column referenced by that index*.

Fair enough.

Unfortunately, the "remove_foreign_key" code we have assumes the foreign-key has been named in a certain way (table_name_column_name) but in this case it wasn't. So Trying a drop index on that caused it to give an equally unhelpful message along the same lines...

This foreign key has been hanging around since the dawn-o-time, and it has one of those automagically-generated constraint-names built by the db itself... something like: my_table_name_ibfk_16.

Now that'd be fine to drop if we just had one client. with one database... you can easily put the following into a migration:

   execute "ALTER TABLE `my_table_name` DROP FOREIGN KEY `my_table_name_ibfk_16`" 

Unfortunately for us, we have 150 clients - each with their own db... and it looks like that constraint-name differs depending on which order the foreign-keys got created. ie sometimes it's my_table_name_ibfk_16 and sometimes it's my_table_name_ibfk_3 - and if you drop it by number - you could be killing the WRONG foreign-key constraint... which would just be embarrassing.

The way out of this quandary is to query the information_schema table to find the foreign-key's actual constraint-name. and here it is for your amusement. (note: put this into initializers eg by saving it as config/initializers/migrations.rb

class ActiveRecord::Migration
  # grab the db-name out of the connection and persist it
  # it's not going to change over the course of a single migration
  def self.fetch_database_name
    @@database_name ||= connection.database_name

  # Use this if the foreign-key was created without an explicit name - 
  # and has one of the automatically-generated constraint-names.
  # This method queries the information-schema table to fetch out the key
  # name before continuing to drop the foreign-key
  # Use this in your migrations with:
  #    remove_legacy_foreign_key :table_name, :field_name
  # eg:
  #    remove_legacy_foreign_key :widgets, :wodget_id
  def self.remove_legacy_foreign_key(table, column_name)
    # first pull the foreign-key name from the information schema
    result = execute "select constraint_name from information_schema.key_column_usage as ke where ke.table_schema = '#{fetch_database_name}' and ke.table_name = '#{table}' and ke.column_name = '#{column_name}';"
    name = result.fetch_hash['constraint_name']
    raise "Got no foreign key by that name" unless name.present?
    execute "ALTER TABLE `#{table}` DROP FOREIGN KEY `#{name}`"

[*] Note: mysql also gives a similar error if you're dropping a foreign-key constraint that doesn't exist at all by the name you give it:

ERROR 1025 (HY000): Error on rename of './mydbname/my_table_name' to 
   './mydbname/#sql2-1ca8-f6cc5' (errno: 152)

Saturday, 17 November 2012

Kiva is addictive

No IT angle today...

I have a mixed approach to giving to charity. I have a number of regular direct debits for various causes, and I also give sporadically in lump sums when inspiration strikes me.

My latest of these is Kiva (here's a Kiva invite) - which is a site that does micro-loans to people in developing countries.

Kiva has been on my radar for a while now, but I'd been putting it off because I had a few misconceptions about the whole process and I finally got down and actually did some reading on what it's all about.

I was mainly concerned by the fact that in most cases, the loans have already been given out, and you are paying for something that is already a fait accompli - which I thought was weird, and that it meant that you weren't really lending the money to the people you had chosen.

Turns out to make a bit more sense. These people need loans immediately, so the partner in that country does so, and then posts the loan up to Kiva. When you choose to "fund" that loan, your money then underwrites it. You are taking on the risk for the loan.

Now I've had it explained, it makes sense to me, so I've been having fun donating... it's a bit more addictive than I expected. I've already made five loans in as many countries, three of which have now been fully funded - one of which I was the final person that funded it (and that feels pretty good). We'll see how they go.

Thursday, 8 November 2012

Link: Unit tests don't find bugs

A quick article called unit tests don't find bugs points out what unit tests are really for, and advises us not to forget that unit tests cannot be used as a substitute for QA-style testing.

Friday, 2 November 2012

ruby gotcha: beware trailing commas

I had something like the following (though with more values):

 h =
 h[:a] = {:a => 1}
 h[:b] = {:b => 2}, # note the trailing comma - oops
 h[:c] = {:c => 3}

and started getting errors telling me that h[:b] was an array, when I was expecting it to be a hash...

Here's what's actually happening (via irb):

>> a = {:a => 1}
=> {:a=>1}
>> b = {:b => 2},
?> c = {:c => 3}         # still part of the previous statement
=> [{:a=>1}, {:c=>3}] # so we end up with an array

Note to self: beware trailing commas