One goal of Inweb is to scale from tiny to enormous webs without too much effort. A project can begin as a single-file web, grow to the point where it needs to be broken up into multiple sections and given a contents page, and grow further until the sections need to be divided up into chapters.

But what then? The Inform compiler has 793 sections, divided into 146 chapters. No single book should have 146 chapters.

The answer is that Inweb supports a still larger-scale grouping, module. A module is just a web, but one which is not tangled independently: it doesn't make a stand-alone program, it provides a component for other webs to use. Inform is therefore broken up into 27 different webs, one of which is inform7, the web itself, and the other 26 of which are its modules. So the average module has about six chapters, and the average chapter has about six sections.

For a more modest example, suppose we want smorgasbord to use a module called random-arrays which might, say, provide random lists of numbers to try sorting. The contents page might then read:

Title: Sorting Smorgasbord
Author: Various Artists
Notation: MarkdownCode
Language: Python
Version Number: 3.0.1

Import: random-arrays

Sections
	Counting Sort
	Quick Sort

The line Import: random-arrays says that this second web provides a block of material forming part of the program. This second web is exactly like a regular web, though it has to be a multi-file one with a contents page, except that its directory name must end -module. In this case, then, Inweb looks for a directory called random-arrays-module.

How does Inweb know where to find random-arrays-module? It tries three locations in turn when looking for a module:

• the inside of the web making the import, that is, its own directory;
• the outside of that web, that is, the directory containing it;
• Inweb's own interior, wherever that is on the user's computer, but only if the module asked for was called foundation or literate.

In this case, then, if smorgasbord is inside the directory programs, then Inweb would first try programs/smorgasbord/random-arrays-module, and then, if that failed, programs/random-arrays-module.

And the third possibility does not arise. Inweb contains just two modules, foundation and literate, and both are written in a dialect of C called InC. One is a general-purpose library wrapping and considerably extending the C standard library, and the other contains literate-programming functions which are essentially the whole Inweb engine. (Inweb itself is really just a command-line interface to this module.)

Two notes for power users of this feature:

• The import location can be a path, as in Import: services/words. This is then applied relative to the locations given above.

• Modules can themselves import other modules, and so on: they form a dependency tree. But less is more. It doesn't seem to be very helpful to have elaborate import pathways.