Nearly 15 years ago, fresh out of college, Giuseppe was hired on at a mobile networking company. The company wasn't a software company, but since they were heavy in networking and could handle all sorts of weird technical problems there, software must basically be the same thing, so they also started producing applications to manage those networks.

It didn't take them long to discover that "building networks" and "writing software" are different skillsets, and so they did what every company with some room in the budget does: they hire a pack of Highly Paid Consultants to get their project on track.

Giuseppe joined the team just as the budget for consultants ran out. They "knowledge transfered" "everything" in a whirlwind of incomprehensible PowerPoint presentations, and then vanished back into whatever lairs HPCs come from, and left Giuseppe and the other new hires to figure out what was going on.

When the HPCs were hired, the company had extensively specified the requirements. Each requirement was expressed in terms like:

  • 12.2.23.a.1 The user will be able to reverse their journey through the app using the back button.
  • 12.2.23.a.2 The app will remember the last 5 screens the user has visited.

There was an elaborate tracking system, with multiple signoffs, to confirm who tested what and when. That specific requirement, 12.2.23.a.2, was signed as "passed" by the original HPC. Then it was passed off to a QA test, which also signed off. Then it went to the users, who submitted a defect: the back button often doesn't behave as expected. It drifted back to the original developer, who closed the defect as "can not replicate", QA signed off again, the user re-opened the ticket claiming the issue was still there, and round and round until the budget for consulting ran out and it was Giuseppe's turn to figure out what was actually going on.

There was no automated testing, and no formal test plan, but it didn't take Giuseppe long to figure out what the bug was. In fact, it took slightly more than 6 navigation actions, to discover that he could replicate the bug consistently. It also explained why the HPC couldn't replicate the bug: they navigated through exactly six pages, and confirmed that if they hit the back button 5 times, they got the same pages they'd been through. If they'd gone to one more page, they would have seen that it didn't work. And the offending code was easy to spot:

private void OnPageNavigated(Page pageFrom)
{
	if(m_backStack.Count < 5)
	{
		m_backStack.Push(pageFrom);
	}
}

Instead of tracking the last five pages the user viewed, this tracked the first five pages they viewed, and ignored every navigation thereafter.

This was just one of many bugs around the back button alone, and additional confusion about the specs. Many pages might open a modal dialog- should that be in the history? If it is, does it count against the 5 pages total? It ended up creating situations where the stack of navigation events needed to track more than 5 items, because of the fuzzy definition of what was a "page", but that created additional bugs because if you could navigate backwards more than five times, you were technically in violation of the spec.

Giuseppe adds:

As I've moved up the software engineering ladder I cringe about how the project was run and managed. Things like unit testing or CI didnt exist, all testing was extremely manual, performed by a dedicated team of 3 testers to an exhaustive specification. … As a [junior] developer nobody ever performed a code review on what I wrote, it went straight into the app. In the whole time I worked there only one person gave me feedback on my code when I screwed up OTA updates that amounted to 'dont do it again'. I was there for over 2 years, and it should be no surprise that the product still hadn't been signed off when I left.

It wasn't signed off when he left, and it probably hasn't been signed off to this day.

[Advertisement] Utilize BuildMaster to release your software with confidence, at the pace your business demands. Download today!