[Beaver-devel] Search algorithms

[Michael Terry]

Hello again, Leslie.  I was thinking about how I wanted to implement the 
new search stuff, and I thought I'd run my ideas past you, to get some 
feedback.

I'd like to allow searching in this file, searching in all open buffers, 
and searching on a set of files (matched by a regexp, like 
/home/foo/bar*.txt).  Now, ideally, we would support all sorts of 
options, like case sensitive or not, match whole word, regular 
expressions, etc.

I figure this is a lot of new code to add, especially since there is a 
tool that we know we can depend on existing that does all this for us -- 
grep.  So, I was thinking we could just call grep and handle the output, 
mapping it to whatever is open.

What I have in mind for output is similar to how UltraEdit for Windows 
handles things, if you've used it.

So, for example, i want to search in all files matching 
/home/foo/bar*.txt for the string 'dood'.  We could call "grep dood 
/home/foo/bar*.txt" with appropriate output modifying flags and display 
this to the user.

The user is presented (in a view that appears in the main window (by 
default under the notebook, I guess)) with a tree list of lines that 
matched the query.  So, if bar1.txt exists and bar2.txt exists, there 
would be a little tree view on the bottom that looks like:

V bar1.txt
-- this line matches the dood token
-- doodle on the paper
V bar2.txt
-- cradle dood

Where you could collapse or expand each file's matches (i.e., the user 
could make it look like:

> bar1.txt
V bar2.txt
-- cradle dood

).  Double clicking (or single-clicking?) on the matching line would 
take you to the line in the file.  To do so, we would scan through the 
open buffers, and if one of them is the same file and also has not been 
modified, then we use it.  Otherwise, if there is no unmodified versions 
of the file open, we open a new file for it.  And, we would go to the line.


Ok.  That's for searching on files that match a pattern.  If we search 
on an open buffer or buffers, we would pass the text to grep via 
standard input and parse the output like normal.

In either case, we would use GtkTextMark's to mark where each line 
starts so that if a user changes the number of lines, our places 
wouldn't be lost.  If the search was on an unopened file, we would wait 
to place the marks until it was open.


In the case of replacing, we could probably use sed in a similar fashion.


I think the idea of outsourcing our searching algorithms to grep is a 
reasonable choice.  It would give us a lot of power at little 
programming cost, would avoid duplication of work, and avoid bugs.  We 
can rely on it being present on any platform we need (if we were going 
to autotool Beaver, they promise that any platform that autotools 
support will have certain programs; among them are grep and sed).


What do you think?

-mt

pgpuiinBsXBy5.pgp
Description: PGP signature