Agree with pretty much everything inside this. The big difference I note however is that my experiences with running searchcode.com suggest that for public search engines that closer to 60% of searches appear to be "How to use a function". By contrast 60% of API calls appear to be looking for AWS keys, passwords and exploits.<p>This is probably totally different to search working over internal codebases however.<p>I must confess... I was originally looking from a vanity point of view and have mixed feelings to see that searchcode was mentioned in the references but not linked.
Kind of funny that this is coming from Google considering how bad Google has become for code search. I wish there was a way to turn off the "I'm ignoring what you searched for and returning what I think you meant." feature.
I imagine typing a code comment in my text-editor before I start writing the code (e.g., calculate fibonacci sequence) and my editor (on a machine connected to the internet) populates with sample open-source code in my language (inferred from the file extension, or shebang, etc), that I can use as a boilerplate and make changes. The populated code is one search result, with the option to look at second most relevant result, third-most relevant and so on. And it shows the source url where I can go for details.<p>I think something like this with rosetta-code snippets is very doable (a weekend project, assuming you're good with your editor's programmability).
25% of the total is documentation related .. Thats really interesting ... A 'smartly designed' documentation website can cater to 25% of total development queries ?