Osm Admin

I stopped working on the Osm Admin project, for financial reasons. Did it hurt? A lot. Was it necessary? Unfortunately, yes. This post is a short retrospection on this matter.

I waited for this day too long. Today, I've tried out GitPod, and made it work for a project based on Osm Admin.

Today, I reiterated on what's left in this iteration, and finished the side menu view, and created a trivial home page for the admin area.

After getting the main branch all-green (tests pass, CLI and UI works as expected), I decided that from now on, I'll keep it always green, and use feature branches for all major development.

Then, I started implementing the main menu of the admin area.

Yesterday, I created a template for new projects powered by Osm Admin.

This way, you can create and publish a project in minutes, just follow the README. Yay!

Yesterday, I tested the instructions for installing Osm Admin locally as a contributor.

Yesterday:

• Osm Admin in-browser functionality was fully restored after creating search index in the indexer.
• All test suites started passing again, and being run automatically on git push.

Yesterday started as usual, planning new iteration, and then BAM! - one folk - Saif - joined the project. It made me correct the course a bit, and take care of potential contributors:

• README explains what the project is about, and how to get started as a contributor;
• I'm fixing things to make the project actually runnable just after installing it.

Yay! After implementing invalid data conversion, I finished the iteration #18 dedicated to diff-based migrations!

I must say, it's a bit disturbing to implement the same thing for the third time. Still, the goal is to get it finished, and have it sustainable.

After the effort, the code has become su much easier to read!

I continued working on property diff algorithm that plans all the migration details.

Yesterday, I finished implementing data conversion for int and string property types.

Then, I started refactoring it.

Last time, I pushed through string property migrations and created a migration log.

Today, I continued solving data conversion issues.

• I finished implementing string property migrations.
• Migrations have got a meaningful log explaining what migrations take place, and why.
• Along the way, I fixed numerous issues.

Yesterday, I finished writing int property migrations. True, testing it is still a todo.

The major part of the code (type change, nullability and other attribute handling) will be reused in other property types.

I refactored Property::migrate() using additional Migration classes. It's so much more convenient to compare property definition versions and generate migration SQLs!

New code structure already handles property explicitness changes.

After enumerating what kind of changes can happen to a property, I started implementing the most hard one - changing property type.

The Query::bulkUpdate() method is implemented in the TDD way.

Implemented in Osm Admin:

• new ALTER TABLE algorithm
• NULL to non-NULL data conversion
• new query test suite

This time, I tackled adding an explicit property to an existing table, and then making it non-nullable.

It raised the need for data conversions - additional handling of the existing data - or otherwise, the database engine fails, or the data becomes invalid.

And it made me split migrations in two parts - the one that runs before the data conversion, and the other one that runs afterwards.

While working on diff algorithm I noticed a certain pattern that I use over an over again.

Using this pattern, I implemented notification table diff.

Finally, I rearranged diff/migration code for better code readability.

Yesterday, I moved the table creation code into new schema diff migration engine.

Now, working on notification table migrations.

This time:

• I found out that the whole search index creation should be done during indexing, not migrations.
• I refactored diff algorithm and made it much easier to read.

Before digging into property diff algorithm, let's dig into how exactly it's used, and what other property-related logic is there.

Yesterday, I continued working on schema migrations:

• implemented table diff algorithm;
• took into consideration table renames.

After a system update, I've got some problem with my laptop, and one thing to check was system log files.

This got me thinking: what makes a good log for my own application? And I think that application logs are as useful as they are able to answer potential questions I may have when an application is in production.

Let's think what kind of questions I might have, and what information might be useful to answer them.

Recently, I fixed numerous issues to make the first migration tests work.

Then, I started implementing the schema diff algorithm - the basis for schema migrations.

More dots got connected in the schema migration design, see this post for details.

Then, I tried to return to TDDing it, but strange errors met me there, so I left them to be resolved in the next piece.

Schema migrations are going to be generated and executed automatically under gulp watch. It will result in lots of tiny migrations: add a column, change its type, make it not nullable, and so on.

It's convenient during development, but it might be not optimal to reapply in production. Indeed, it's better to convert a large table to new structure all at once rather than running a conversion for each of the tiny migration.

This observation contradicts the previously stated migration workflow, so let's update it.

In spirit of TDD, I started with a meaningful, but failing test that is reasonably fast. Then, I worked to make it green:

• implemented the loading of schema fixtures from the codebase;
• worked on generating migration files.

After drafting some ideas of how the schema migrations should actually work, let's start implementing it in TDD way.

The goal of the current iteration is to adjust the database schema and preserve/convert existing data according to any changes in schema classes, grids, forms or indexers.

This post presents some thoughts how it might work with class definitions.

After finishing the iteration "#17 Minimum UI", I revisited the goals and the scope of the minimum viable product, and picked the most pressing task for the next iteration - "#18 Database Schema Changes".

I implemented the last user action that is typical to a CRUD application - object deletion.

Done:

• routes responsible for creating new products, and modifying existing products
• incremental indexing
• cloning query objects
• INSERT INTO ... SELECT ... statements

My current goal is to get some minimum Osm Admin user interface and API working - just for one property, one data type, one UI control type, and then improve.

And there is a lot of space for improvement. The main areas are listed in this document.

I'd really appreciate a helping hand, so if you are into this sort of things, DM me on Twitter.

I feel that both attributes decorating data classes, and object model parsed from it are too complex.

Let's have a fresh look at Osm Admin. If I had to start from scratch, how would it be?

Today, I finished implementing an object creation form.

It works according to the specs on Interfaces and Forms. Currently, it only supports string and integer input fields; later, there will be a lot more.

Let's review how it works.

In Osm Admin, a form is an integral part of the class user interface. It's a very specific kind of HTML form.

In this article, I'll review what's in a class form, and how it's reflected from the application code.

Yesterday, I described a user interface pattern for managing data of a certain data class using a grid/form pair. Let's call this grid/form pair class interface in the admin area.

In this article, I'll examine what moving parts are inside a class interface.

So far, I've been concerned about how application data is modeled, stored and handled internally.

Yet, the final objective a user interface for viewing and editing it visually.

Let's dream a bit about how it might look like.

Yay, indexing works!

It's changed quite a bit, so let's review how it works from the very beginning.

I'm still working on indexing, and it seems to get one more refactoring. But more on that later.

Meanwhile, let's review Osm Admin modules, their responsibilities and dependencies.