Walking on the Sun

In 1992, I worked at a shop that was all SunOS. Most people had a Sparc-1. Production boxes were the mighty Sparc-2, and secretaries had the lowly Sun 360. Somewhat typical hardware for the day.

Sun was giving birth to their brand spanking new Solaris, and was pushing everyone to convert from SunOS. As with any OS change in a large shop, it doesn't just happen; migration planning needs to occur. All of our in-house software needed to be ported to the new Operating System.

This planning boiled down to: assign it to snoofle; let him figure it out.

This was before Sun made OpCom available to help people do their migrations.

I took the latest official code, opened an editor, grepped through the include files and compiled, for each OS. Then I went into a nine month long compile-edit-build cycle, noting the specifics of each item that required different include files/syntax/whatever. Basically, Sun had removed the Berkeley libraries when they first put out Solaris, so everything signal or messaging related had to change.

Finally, I naively thought the pain was over; it compiled. I had coalesced countless functions that had nearly identical multiple versions, deleted numerous blocks of dead code, and reduced 1.4 million LOC to about 700K. Then began the debugging cycle. That took about 3 weeks.

Then I was told not to merge it because another subteam in our group was doing a 9-month sub-project and couldn't be interrupted. Naturally, they were working in the main branch, which forced me to keep pulling and porting their code into mine several times a week, for months. Ironically, they were constantly changing dead code as part of trying to fix their own code.

You can only do this for so long before getting fed up; I'd had it and let it be known to boss+1 (who was pushing for Solaris) that this had to end. He set a date three months out, at which time I would do the merge and commit; other tasks be damned! The subteam was repeatedly informed of this drop-dead date.

So I put up with it for 3 months, then did the final merge; over 3,500 diffs. I went through them all, praying the power wouldn't cut out. After fixing a few typos and running the cursory test, I held my breath and committed. Then I told everyone to pull and merge.

It turns out that I missed 3 little bugs, but they were suffiently visible that it prevented the application from doing anything useful. The manager of the sub-team ordered me to roll it back because they were busy. I handed her the written memo from B+1 ordering me to do it on this date and told her to suck it up and give me a chance to debug it.

An hour later, it was working and committed.

I instructed everyone to pull and build, and to follow the instructions in my handout for coding going forward. Anything that broke the Solaris build would be summarily rolled back per orders from B+1.

It took a few months and hundreds of rollbacks for them to start to follow my instructions, but when they finally did, the problems ceased.

Then the managers from the other teams took my instructions and all my global edit scripts (it wasn't a perfect parser, but it at least left syntax errors if it tried to change code that was really badly formatted, so you could trivially find them and fix them very quickly).

Using my scripts and cheat sheets, my peers on the other projects managed to do their ports in just a couple of hours, and mercilessly rode me about it for the next 3 years.

[Advertisement] Otter - Provision your servers automatically without ever needing to log-in to a command prompt. Get started today!