All entries for Wednesday 16 January 2019

January 16, 2019

Sandcastles

A super quick post, mainly to point at somebody else's post from a while ago, namely http://kera.name/articles/2013/01/the-dynamic-allocation-of-sandcastles/

In case the link disappears, which it looks like the original post already has, the idea is as follows. This is something between a paraphrase and a direct quote, and all credit goes to the Original Poster quoted in the link above, but not named.

The memory space in the computer is like an area of beach, marked off for the building of sandcastles. To strain the analogy to breaking point, other parts of said beach might be marked off for other purposes (e.g. sunbathing, which has no useful relation to program space, and there is far less import to leaving your towel around).

When you want to build a sandcastle (create any object or variable) you ask for an area of beach, and one is selected and given to you.
What this area contains is a load of lumpy sand, usually containing the dregs of other peoples castles, possibly cutting across multiple. It's a mess, basically.
You, usually, either flatten the sand out, or immediately build a castle there, in both cases ignoring what might be there already.
When you're done, you surrender the area, but usually don't remove your castle.
The organisers are then free to give that area to somebody else.

So, what does this mean? Well:

If you go back to your area (keep around a pointer, reference etc to it), your castle may well still be there, in all its glory. But somebody else can come along at any time and kick it down, and beach security are going to support their right to do so. So, you can't trust the shape of any of it.
You also can't TELL whether anything has happened to your castle by examining it.
If you accidentally released the area (delete, dealloc etc) and you've forgotten that, you will, sooner or later, find a bit that's not how you left it, or get your castle kicked down at a crucial moment. Do NOT DO THIS.
You can't even trust what seems to be a patch of flat sand, unless you flattened it, or requested it pre-flattened (e.g. calloc in c).
If you inherit somebody else's area, there might be a castle there! Since you can never trust the castle, you can't use it. Somebody might have already been along and undermined all the turrets. If you try and use it as a castle, all kinds of strangeness can ensue.
BUT suppose somebody has a secret technique for building castles. If they leave one un-destroyed, you might be able to find out all kinds of "secret" things. So, any secret data has to be carefully destroyed. It's not enough just to surrender control of the area. This is just like the need to carefully destroy data on an old hard drive - by actually over-writing it, not just deleting it.
If you, or somebody else, spills out of their designated area, you generally end up mucking up bits of each-other's castles, usually in such a way as to ruin them. DO NOT DO THIS either.

It's a simple analogy, but it is perfect to explain why it's so easy to miss uninitialised variables, and premature free/delete/dealloc operations. Often things still look perfectly fine! Often all the areas close to yours are unallocated, so you can apparently spill out of your area without consequence, until management changes policy on picking areas and suddenly things go wrong.

Heather Ratcliffe : 16 Jan 2019 12:29 | Comments (0) | Close comments | Report a problem