Re: A package in a league of its own: Helm

[solidius4747]

> 99.9% of the time I work on projects that I worked on previously and
> 
> hence I know its structure. Knowing the *precise* structure is not
> 
> really required as you can type just the part that is characteristic
> 
> enough (the ordering matters, though, as you mentioned.) Having a method
> 
> that minimizes the amount of text that I have to write is very
> 
> convenient.

Since I started using Helm, I never bother to remember the structure of my 
working directory. And that's the point: I don't have to remember such trivial 
details and I can focus my mind on other more important things.
 
> 
> For your query about the existence of a main.c file for the x86 arch, I
> 
> would type "x86main.c". The answer should be evident by just looking at
> 
> the first match. (Acually, "xmac" should be enough if you are willing to
> 
> look beyond the first match.)

And what if it doesn't exist? You will have to move point at the beginning and 
delete whole thing. With ido, you are constrained by its ordering: you cannot 
immediately enter what you directly one, such as an arbitrary file, and refine 
later. If there are a large pool of candidates, you have to move point at the 
beginning and start typing additional information. Worse, you maybe to move 
point to type something in between. Suddenly it's not so convenient to find 
something anymore. And ido is not particularly suit for narrowing large set of 
candidates with complex query.

Here is another example that make sure of Helm interface nicely: 
helm-semantic-or-imenu. Demo: 
http://tuhdo.github.io/static/part3/helm-semantic-or-imenu.gif

You can think of it as an outline tree in regular IDE, but Helm makes it 
interactive, and fast. I can ask question like "is there something that is 
function and has int in it that the function name contains memory?".

Here is another demo on Helm's homepage: 
https://github.com/emacs-helm/helm#advanced-usage

Finally, the strings in Helm are not just strings; they are regexp and I think 
they are more powerful then normal fuzzy matcher.

> 
> I get your point about incrementally refining the query and I realize
> 
> that it can be more convenient than flx for certain uses (such as when
> 
> exploring a large set of previously unknown candidates.) It's only that
> 
> the flx way is more convenient on my case, where I always work on a
> 
> semi-known set of candidates.

Sure, it could be more productive for things you work regularly. But I like a 
general interface that can deal with both known and unknown environment.