Re: Inverted index to accelerate guix package search

[Arun Isaac <arunisaac@systemreboot.net>]

[-- Attachment #1.1: Type: text/plain, Size: 1749 bytes --]

Pierre Neidhardt <mail@ambrevar.xyz> writes:

> By the way, what about using Xapian in Guix?

I looked up xapian's features at https://xapian.org/features and it is
quite impressive. I was introduced to xapian through notmuch. notmuch
does not utilize xapian to the fullest and I therefore ended up
underestimating its value. Of particular importance might be the
following.

- Relevance feedback - given one or more documents, Xapian can suggest
  the most relevant index terms to expand a query, suggest related
  documents, categorise documents, etc.
- Phrase and proximity searching - users can search for words occurring
  in an exact phrase or within a specified number of words, either in a
  specified order, or in any order.
- Supports stemming of search terms (e.g. a search for "football" would
  match documents which mention "footballs" or "footballer")

I think these features would really help in Pierre's work trying to
improve search and discoverability on Guix. If we are planning to have a
"Software Center" like interface at some point in the future, xapian's
search could come in handy.

Not directly related to Guix, but I also wonder if info manuals would be
a lot more useful if they had good full text search using xapian.

For the time being, since we don't have xapian bindings, I think we
should settle for sqlite's full text search capabilities.

https://www.sqlite.org/fts5.html

I have attached a short proof of concept script for an sqlite based
search. Speedup is around 200x, and populating the database only takes
around 2.5 seconds. Here is a sample run.

Sqlite database populated in 2.5516340732574463 seconds
Brute force search took 0.11850595474243164 seconds
Sqlite search took 5.459785461425781e-4 seconds


[-- Attachment #1.2: sqlite-search.scm --]
[-- Type: text/plain, Size: 2145 bytes --]

(use-modules (gnu packages)
             (guix packages)
             (ice-9 match)
             (sqlite3)
             (srfi srfi-26))

(define db (sqlite-open "/tmp/index.sqlite"))

(define schema
  "CREATE VIRTUAL TABLE packages USING fts5(name, description)")

(define (build-sqlite-database)
  (sqlite-exec db schema)
  (sqlite-exec db "BEGIN")
  (fold-packages
   (lambda (package _)
     (let ((statement (sqlite-prepare db "INSERT INTO packages(name, description) VALUES(:name, :description)")))
       (sqlite-bind-arguments statement
                              #:name (package-name package)
                              #:description (package-description package))
       (sqlite-fold cons '() statement)
       (sqlite-finalize statement)))
   #f)
  (sqlite-exec db "COMMIT;"))

(define (sqlite-retrieve query)
  (let ((statement (sqlite-prepare db "SELECT name FROM packages WHERE description MATCH :query")))
    (sqlite-bind-arguments statement #:query query)
    (let ((result (sqlite-fold (lambda (v result)
                                 (match v
                                   (#(name) (cons name result))))
                               '() statement)))
      (sqlite-finalize statement)
      result)))

(define (brute-force-retrieve query)
  "Return names of all packages whose descriptions contain the string QUERY.
Search brute force by folding over all packages."
  (fold-packages (lambda (package result)
                   (if (string-contains (package-description package) query)
                       (cons package result)
                       result))
                 '()))

(define (time-us thunk)
  (define pair->sec
    (match-lambda
      ((sec . usec)
       (+ sec (/ usec 1e6)))))

  (let ((start (gettimeofday)))
    (thunk)
    (let ((stop (gettimeofday)))
      (- (pair->sec stop) (pair->sec start)))))

(format #t "Sqlite database populated in ~a seconds
Brute force search took ~a seconds
Sqlite search took ~a seconds
"
        (time-us build-sqlite-database)
        (time-us (cut brute-force-retrieve "strategy"))
        (time-us (cut sqlite-retrieve "strategy")))

(sqlite-close db)

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]