Skip to content
Fast and lightweight binary search on text files
C Shell
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.gitignore extended .gitignore Sep 6, 2017
README.txt
compile_darwinc32.sh added compilation scripts for macOS, xstatic and xtiny Sep 6, 2017
compile_darwinc64.sh
compile_mingw.sh added compilation for Windows (compile_mingw.sh) with MinGW gcc Sep 6, 2017
compile_xstatic.sh added compilation scripts for macOS, xstatic and xtiny Sep 6, 2017
compile_xtiny.sh added compilation scripts for macOS, xstatic and xtiny Sep 6, 2017
pts_lbsearch.c added compilation for Windows (compile_mingw.sh) with MinGW gcc Sep 6, 2017
pts_line_bisect.py added link to the article Dec 1, 2013
pts_line_bisect_test.py added alternative implementation mini_bisect_left fo pts_line_bisect_… Dec 1, 2013

README.txt

README for pts_lbsearch
^^^^^^^^^^^^^^^^^^^^^^^
pts_lbsearch is a fast binary search in a line-sorted text file. It finds
and prints lines within a given range or with a given prefix. It works
even if lines have variable length. pts_lbsearch is written in C, it is
optimized for speed and it is using buffering and cache to avoid unnecessary
read(2)s and lseek(2)s. It also has a very small, constant memory footprint:
barely larger than the read buffer (8KB by default).

Send donations to the author of pts-line-bisect:
https://flattr.com/submit/auto?user_id=pts&url=https://github.com/pts/pts-line-bisect

pts_lbsearch gives correct results only if the bytes of the input file are
sorted lexicographically. Use LC_ALL=C to get 8-bit, bytewise,
locale-independent sort(1):

  $ LC_ALL=C sort <file >file.sorted

Prefix search: print all lines starting with foo, sorted:

  $ pts_lbsearch -p file.sorted foo

Prefix search: print all lines starting with foo, sorted, but ignore the
last incomplete line (if any):

  $ pts_lbsearch -pi file.sorted foo

Prefix range search: print all lines starting with bar or foo or between:

  $ pts_lbsearch -p file.sorted bar foo

Closed ended interval search: print all lines at least foo and at most bar:

  $ pts_lbsearch -t file.sorted bar foo

Open ended interval search: print all lines at least foo and smaller than bar:

  $ pts_lbsearch -e file.sorted bar foo

Prefix detection: exit(0) iff a line starting with foo is present:

  $ pts_lbsearch -qp file.sorted foo

Exact detection: exit(0) iff a line equals to foo is present:

  $ pts_lbsearch -qt file.sorted foo

Prepend position: print the smallest offset where foo can be inserted (please
note that this does not report whether foo is present, always exit(0)).

  $ pts_lbsearch -oe file.sorted foo

Append position: print the largest offset where foo can be inserted (please
note that this does not report whether foo is present, always exit(0), also
you may need to append a '\n' before appending foo at EOF).

  $ pts_lbsearch -oae file.sorted foo

Prefix range: print the range (start byte offset and after-the-last-byte end
offset) of lines starting with foo:

  $ pts_lbsearch -op file.sorted foo

Exact range: print the range (start byte offset and after-the-last-byte end
offset) of lines containing only foo:

  $ pts_lbsearch -ot file.sorted foo

See http://pts.github.io/pts-line-bisect/line_bisect_evolution.html
for a detailed article about the design and analysis of the algorithms
pts_lbsearch implements.

pts_lbsearch supports only '\n' as the line separator. '\r' is treated as a
regular character within the line.

pts_lbsearch supports incomplete lines at the end of the input. When
printing them, it never adds a '\n' to the end if there wasn't one. To make
it ignore the incomplete last line (possibly because an other, slow process
has not finished writing it), pass the `-i' flag.

If the input is not sorted, pts_lbsearch may print incorrect lines or
offsets (can be more or less than expected). But it wouldn't crash or fall
to an infinite loop.

pts_lbsearch works for large files (i.e. larger than 2GB) correctly.

Please note that a lookup in a btree or hash index is usually faster than a
binary search, beause btree and hash need much fewer disk seeks because of
the large branching factor. So if you can afford to build an index, there
are faster lookup solutions than binary search. But if indeed you want
binary search on variable-length records, most probably pts_lbsearch is the
fastest.

Python implementation
~~~~~~~~~~~~~~~~~~~~~
There is a Python implementation in the file pts_line_bisect.py.

Differences in the C and Python implementations:

* The C implementation is much faster, because avoids lseek(2) and read(2)
  calls as much as possible, while the Python implementation doesn't,
  because in Python the `file' object discards the read buffer after each
  file.seek.
* The Python implementation is more compact, contains more comments, and it
  is easier to understand, to reuse as a library and to extend.
* The C implementation has a more versatile command-line interface. The Python
  implementation supports flags -etco only.
* The Python implementation has unit tests.
* The C implementation supports prefix search (CM_LP).
* The C equivalent of bisect_interval does a CM_LE search and then a CM_LT
  for the rest. This is faster if the result interval is a short range near
  the end of the file.
* The C implementation has a very small memory footprint: only dozens of
  offsets and flags in addition to a single file read buffer (of 8KB by
  default).
* The C implementation doesn't do any dynamic memory allocation (except
  possibly by the printfs generating messages), it's so lightweight and so
  low-overhead that it can be used in memory-constrained environments such
  as routers.

__EOF__
You can’t perform that action at this time.