There’s a lot here that confuses me.

O2 have not done the simple and obvious thing - delete the old address, store the new one - to change the address in their database. I don’t think they have a versioning scheme with effective dates or anything sophisticated like that. Instead, we have some peculiar piece by piece update, giving us this weird merged outcome.

Let’s annotate those addresses a little:^[2]

Some houses have names, some have numbers, some have both. Whoever wrote this wacky address update code knew to account for those various permutations. We need to know the difference between a name and number for example, if only to be able to format the address properly.

The new address has no house number, so the existing one was deleted. The new house name was written in. All looks good. Well done them.

On the other hand, they’ve completely fumbled the fact that the new address has no street name Well, maybe not completely fumbled. The address did change to include all the new bits. It didn’t complain or just plain break. That’s something.

But the choice, and this is a choice people actively made, of if there’s no street name then leave the street name unchanged I find very curious. It’s so obviously wrong. We’re changing the address - everything’s in play, we’ve paid attention the name/number thing - and yet we’re chosing to not change part of it. I wonder if the same applies if a new address lacks an area name, for instance^[4].

Somehow though, someone, somewhere, wrote this behaviour into the code. Nobody else in the organisation noticed it, or if they did it wasn’t challenged, and it made it out into public view. It must affect hundreds of their customers, but there it is. Very obvious, very wrong, and yet probably very easy to correct^[5].

I am now a fully enrolled O2 customer. I can log into the website, see my bills, change my plan, all that good stuff. But for all that, I’m still in that liminal space, hanging between Birmingham and Pembrokeshire.

Ossanna and Kernighan’s Troff User’s Manual does not say how it was typeset.

Thanks to Ben Deane for details of The Elements of Programming Style, and to Chris Oldwood for details of The Elements of Programming Style 2nd Edition.

For context, \<pattern>\ searches backwards, /<pattern>/ searches forward. \\ and // search backwards or forwards using the most recent pattern again. function and begin are Pascal keywords.

The key here is the difference between using , and ; as the separator between the line expressions.

A , is just a normal separator, but using ; updates the current line number to the value of the expression. That’s why the examples above do what they do. Without the ; the searches would just find the same line each time. With the power of ;, you can do all kinds of crazy stuff. I can’t think why you would, I mean I’ve lasted this long without knowing this stuff, but I’m already expecting to find out this is Turing complete^[1].

That’s it!

That’s the key. Multiple line expressions aren’t a quirk of Kernighan and Plauger’s little line editor, it’s a deliberate part of ed, and consequently ex, and vi, and vim, and presumably other editors too.

Bananas.

My editor didn’t support ; properly. You could pop one in as a separator, but I’m wasn’t doing anything special with it.

Kernighan and Plauger’s editor operates as it parses each command. If they see a ; then just update the current line right there and then. That’s a reasonable choice given their various constraints - writing didactic code in Pascal, in late 70s, for people who’ve never heard of Unix, etc, etc.

My code has a hard divide between command parsing and executing those commands. If you give my editor a command that makes no sense, then nothing should happen. Either it all works, or nothing works. What can I say? I’ve taken the Abrahams guarantees to soul of my being.

To evaluate the line expressions, I need to update the current line without updating the current line while parsing, then actually update the current line when executing.

Hey, what do you know? An extra level of indirection did it! 😀

5 commits, 3 files changed, 87 insertions, 48 deletions

It all went pretty well. The first four commits were gentle preparations to actually pulling the trigger, and implementing the new functionality, all three lines of it, in the last commit.

(And then fixing a classic C++ blunder (function argument evaluation order!) to actually making it work.)

Actually, I realised afterwards, I pulled the trigger on half the functionality - evaluating the line expressions - but not updating the current line during command execution.

Nevermind. Next time.