# Linda
**Linda** is a coordination model, introduced by David Gelernter in the early 1980s, built around a shared associative memory called a **tuple space**. Unlike [[sync-csp|CSP]], where processes send messages directly to each other, Linda processes never address one another at all. They only interact with the tuple space: writing tuples into it and reading or removing tuples that match a pattern.

```
out("task", 42)          // write a tuple into the tuple space
in("task", ?x)            // remove a matching tuple, blocking until one exists
rd("task", ?x)             // like in, but leaves the tuple in the space
```

## The three operations

 - **out** writes a tuple into the shared space. It never blocks.
 - **in** searches for a tuple matching a given pattern, removes it, and blocks if no match currently exists.
 - **rd** is like `in` but leaves the matched tuple in place, so other processes can still read (or take) it too.

Pattern matching is by shape and value: a query like `in("task", ?x)` matches any tuple that is a 2-element tuple whose first field is the string `"task"`, binding `x` to whatever the second field holds.

## Decoupling in space and time

Linda's key property is that producers and consumers are decoupled in both **space** (neither needs to know the other exists, let alone its address) and **time** (a tuple written before any consumer exists just sits in the tuple space until one shows up and reads it). This is a strictly looser coupling than [[sync-mbox|mailboxes]], which are still addressed to a specific recipient even though sender and receiver don't rendezvous synchronously the way [[sync-csp|CSP]] channels do. The tradeoff is that a shared associative space is harder to implement efficiently at scale than point-to-point channels, since every `in`/`rd` is conceptually a search over everything currently in the space.