Nuweb

Nuweb works with any programming language and L^ATEX [#!latex!#]. I wanted to use L^ATEX because it supports a multi-level sectioning scheme and has facilities for drawing figures. I wanted to be able to work with arbitrary programming languages because my friends and I write programs in many languages (and sometimes combinations of several languages), e.g., C, Fortran, C++, yacc, lex, Scheme, assembly, Postscript, and so forth. The need to support arbitrary programming languages has many consequences:

No pretty printing: Both WEB and CWEB are able to pretty print the code sections of their documents because they understand the language well enough to parse it. Since we want to use any language, we've got to abandon this feature.
No index of identifiers: Because WEB knows about Pascal, it is able to construct an index of all the identifiers occurring in the code sections (filtering out keywords and the standard type identifiers). Unfortunately, this isn't as easy in our case. We don't know what an identifiers looks like in each language and we certainly don't know all the keywords. (On the other hand, see the end of Section 1.3)

Of course, we've got to have some compensation for our losses or the whole idea would be a waste. Here are the advantages I can see:

Simplicity: The majority of the commands in WEB are concerned with control of the automatic pretty printing. Since we don't pretty print, many commands are eliminated. A further set of commands is subsumed by L^ATEX and may also be eliminated. As a result, our set of commands is reduced to only four members (explained in the next section). This simplicity is also reflected in the size of this tool, which is quite a bit smaller than the tools used with other approaches.
No pretty printing: Everyone disagrees about how their code should look, so automatic formatting annoys many people. One approach is to provide ways to control the formatting. Our approach is simpler – we perform no automatic formatting and therefore allow the programmer complete control of code layout.
Control: We also offer the programmer complete control of the layout of his output files (the files generated during tangling). Of course, this is essential for languages that are sensitive to layout; but it is also important in many practical situations, e.g., debugging.
Speed: Since nuweb doesn't do to much, the nuweb tool runs quickly. I combine the functions of tangle and weave into a single program that performs both functions at once.
Page numbers: Inspired by the example of noweb, nuweb refers to all scraps by page number to simplify navigation. If there are multiple scraps on a page (say page 17), they are distinguished by lower-case letters (e.g., 17a, 17b, and so forth).
Multiple file output: The programmer may specify more than one output file in a single nuweb file. This is required when constructing programs in a combination of languages (say, Fortran and C)@. It's also an advantage when constructing very large programs that would require a lot of compile time.

This last point is very important. By allowing the creation of multiple output files, we avoid the need for monolithic programs. Thus we support the creation of very large programs by groups of people.

A further reduction in compilation time is achieved by first writing each output file to a temporary location, then comparing the temporary file with the old version of the file. If there is no difference, the temporary file can be deleted. If the files differ, the old version is deleted and the temporary file renamed. This approach works well in combination with make (or similar tools), since make will avoid recompiling untouched output files.