Hi,

I'm accustomed to being able to search for documentation quickly and am 
not satisfied with apropos, whatis, and man -K.

I propose a new output mode, -Tindex.  This will crawl through the NAME 
and SYNOPSIS section of a manual, indexing function names, variable 
types, header files (searching for all manuals that mention a particular 
header file = golden), utility names, etc.  It writes these to a 
Berkeley database file.  Each record consists of the keyword, then a 
bit-field (the type of record), then the corresponding file.

Other utilities can then mine this data...

Enclosed is a quick sketch.  It dumps manual names (Nm in NAME), utility 
names (Nm in SYNOPSIS), and function names (Fo, FN in SYNOPSIS) into the 
index.  It only does -mdoc, but -man can heuristically grab at least the 
name by grabbing ^[[:alpha:]]+ from the NAME section.  It requires some 
modifications to mdoc.h to associate a file-name with a parse.

This is just a sketch; a "real" version would need to be [at least] much 
more careful about stripping non-character escapes and so on.

I'm still not sure whether it's a good idea to have this /in/ mandoc. 
The BSD db.h not standard across Unices.  I've started cleaning up 
main.c to push the main file-reading routines into a utility class, 
which would allow different users of the entire backend library.  Lots 
of things to think about.

Thoughts?

Kristaps