Fsdb(3)               User Contributed Perl Documentation              Fsdb(3)



[1mNAME[0m
       Fsdb - a flat-text database for shell scripting

[1mSYNOPSIS[0m
       Fsdb, the flatfile streaming database is package of commands for
       manipulating flat-ASCII databases from shell scripts.  Fsdb is useful
       to process medium amounts of data (with very little data you'd do it by
       hand, with megabytes you might want a real database).  Fsdb was known
       as as Jdb from 1991 to Oct. 2008.

       Fsdb is very good at doing things like:

       +o   extracting measurements from experimental output

       +o   examining data to address different hypotheses

       +o   joining data from different experiments

       +o   eliminating/detecting outliers

       +o   computing statistics on data (mean, confidence intervals,
           correlations, histograms)

       +o   reformatting data for graphing programs

       Fsdb is built around the idea of a flat text file as a database.  Fsdb
       files (by convention, with the extension [4m.fsdb[24m), have a header
       documenting the schema (what the columns mean), and then each line
       represents a database record (or row).

       For example:

               #fsdb experiment duration
               ufs_mab_sys 37.2
               ufs_mab_sys 37.3
               ufs_rcp_real 264.5
               ufs_rcp_real 277.9

       Is a simple file with four experiments (the rows), each with a
       description, size parameter, and run time in the first, second, and
       third columns.

       Rather than hand-code scripts to do each special case, Fsdb provides
       higher-level functions.  Although it's often easy throw together a
       custom script to do any single task, I believe that there are several
       advantages to using this library:

       +o   these programs provide a higher level interface than plain Perl, so

           **  Fewer lines of simpler code:

                   dbrow '_experiment eq "ufs_mab_sys"' | dbcolstats duration

               Picks out just one type of experiment and computes statistics
               on it, rather than:

                   while (<>) { split; $sum+=$F[1]; $ss+=$F[1]**2; $n++; }
                   $mean = $sum / $n; $std_dev = ...

               in dozens of places.

       +o   the library uses names for columns, so

           **  No more $F[1], use "_duration".

           **  New or different order columns?  No changes to your scripts!

           Thus if your experiment gets more complicated with a size
           parameter, so your log changes to:

                   #fsdb experiment size duration
                   ufs_mab_sys 1024 37.2
                   ufs_mab_sys 1024 37.3
                   ufs_rcp_real 1024 264.5
                   ufs_rcp_real 1024 277.9
                   ufs_mab_sys 2048 45.3
                   ufs_mab_sys 2048 44.2

           Then the previous scripts still work, even though duration is now
           the third column, not the second.

       +o   A series of actions are self-documenting (each program records what
           it does).

           **  No more wondering what hacks were used to compute the final
               data, just look at the comments at the end of the output.

           For example, the commands

               dbrow '_experiment eq "ufs_mab_sys"' | dbcolstats duration

           add to the end of the output the lines
               #    | dbrow _experiment eq "ufs_mab_sys"
               #    | dbcolstats duration

       +o   The library is mature, supporting large datasets, corner cases,
           error handling, backed by an automated test suite.

           **  No more puzzling about bad output because your custom script
               skimped on error checking.

           **  No more memory thrashing when you try to sort ten million
               records.

       +o   Fsdb-2.x supports Perl scripting (in addition to shell scripting),
           with libraries to do Fsdb input and output, and easy support for
           pipelines.  The shell script

               dbcol name test1 | dbroweval '_test1 += 5;'

           can be written in perl as:

               dbpipeline(dbcol(qw(name test1)), dbroweval('_test1 += 5;'));

       (The disadvantage is that you need to learn what functions Fsdb
       provides.)

       Fsdb is built on flat-ASCII databases.  By storing data in simple text
       files and processing it with pipelines it is easy to experiment (in the
       shell) and look at the output.  To the best of my knowledge, the
       original implementation of this idea was "/rdb", a commercial product
       described in the book [4mUNIX[24m [4mrelational[24m [4mdatabase[24m [4mmanagement:[24m [4mapplication[0m
       [4mdevelopment[24m [4min[24m [4mthe[24m [4mUNIX[24m [4menvironment[24m by Rod Manis, Evan Schaffer, and
       Robert Jorgensen (and also at the web page <http://www.rdb.com/>).
       Fsdb is an incompatible re-implementation of their idea without any
       accelerated indexing or forms support.  (But it's free, and probably
       has better statistics!).

       Fsdb-2.x supports threading and will exploit multiple processors or
       cores, and provides Perl-level support for input, output, and threaded-
       pipelines.

       Installation instructions follow at the end of this document.  Fsdb-2.x
       requires Perl 5.8 to run.  All commands have manual pages and provide
       usage with the "--help" option.  All commands are backed by an
       automated test suite.

       The most recent version of Fsdb is available on the web at
       <http://www.isi.edu/~johnh/SOFTWARE/FSDB/index.html>.

[1mWHAT'S NEW[0m
   [1m2.37, 2013-02-26 quick bugfix to support parallel sort and merge from[0m
       [1mrecent releases[0m
       BUG FIX Since 2.35, dbmerge delayed removal of input files given by
       "--xargs".  This problem is now fixed.

[1mREADME CONTENTS[0m
       executive summary
       what's new
       README CONTENTS
       installation
       basic data format
       basic data manipulation
       list of commands
       another example
       a gradebook example
       a password example
       history
       related work
       release notes
       copyright
       comments

[1mINSTALLATION[0m
       Fsdb now uses the standard Perl build and installation from
       [4mExtUtil::MakeMaker[24m(3), so the quick answer to installation is to type:

           perl Makefile.PL
           make
           make test
           make install

       Or, if you want to install it somewhere else, change the first line to

           perl Makefile.PL PREFIX=$HOME

       and it will go in your home directory's [4mbin[24m, etc.  (See
       [4mExtUtil::MakeMaker[24m(3) for more details.)

       Fsdb requires perl 5.8 or later and uses ithreads.

       A test-suite is available, run it with

           make test

       A FreeBSD port to Fsdb is available, see
       <http://www.freshports.org/databases/fsdb/>.

       A Fink (MacOS X) port is available, see
       <http://pdb.finkproject.org/pdb/package.php/fsdb>.  (Thanks to Lars
       Eggert for maintaining this port.)

[1mBASIC DATA FORMAT[0m
       These programs are based on the idea storing data in simple ASCII
       files.  A database is a file with one header line and then data or
       comment lines.  For example:

               #fsdb account passwd uid gid fullname homedir shell
               johnh * 2274 134 John_Heidemann /home/johnh /bin/bash
               greg * 2275 134 Greg_Johnson /home/greg /bin/bash
               root * 0 0 Root /root /bin/bash
               # this is a simple database

       The header line must be first and begins with "#h".  There are rows
       (records) and columns (fields), just like in a normal database.
       Comment lines begin with "#".  Column names are any string not
       containing spaces or single quote (although it is prudent to keep them
       alphanumeric with underscore).

       By default, columns are delimited by whitespace.  With this default
       configuration, the contents of a field cannot contain whitespace.
       However, this limitation can be relaxed by changing the field separator
       as described below.

       The big advantage of simple flat-text databases is that it is usually
       easy to massage data into this format, and it's reasonably easy to take
       data out of this format into other (text-based) programs, like gnuplot,
       jgraph, and LaTeX.  Think Unix.  Think pipes.  (Or even output to Excel
       and HTML if you prefer.)

       Since no-whitespace in columns was a problem for some applications,
       there's an option which relaxes this rule.  You can specify the field
       separator in the table header with "-F x" where "x" is a code for the
       new field separator.  A full list of codes is at [4mdbfilealter[24m(1), but
       two common special values are "-F t" which is a separator of a single
       tab character, and "-F S", a separator of two spaces.  Both allowing
       (single) spaces in fields.  An example:

               #fsdb -F S account passwd uid gid fullname homedir shell
               johnh  *  2274  134  John Heidemann  /home/johnh  /bin/bash
               greg  *  2275  134  Greg Johnson  /home/greg  /bin/bash
               root  *  0  0  Root  /root  /bin/bash
               # this is a simple database

       See [4mdbfilealter[24m(1) for more details.  Regardless of what the column
       separator is for the body of the data, it's always whitespace in the
       header.

       There's also a third format: a "list".  Because it's often hard to see
       what's columns past the first two, in list format each "column" is on a
       separate line.  The programs dblistize and dbcolize convert to and from
       this format, and all programs work with either formats.  The command

           dbfilealter -R C  < DATA/passwd.fsdb

       outputs:

               #fsdb -R C account passwd uid gid fullname homedir shell
               account:  johnh
               passwd:   *
               uid:      2274
               gid:      134
               fullname: John_Heidemann
               homedir:  /home/johnh
               shell:    /bin/bash

               account:  greg
               passwd:   *
               uid:      2275
               gid:      134
               fullname: Greg_Johnson
               homedir:  /home/greg
               shell:    /bin/bash

               account:  root
               passwd:   *
               uid:      0
               gid:      0
               fullname: Root
               homedir:  /root
               shell:    /bin/bash

               # this is a simple database
               #  | dblistize

       See [4mdbfilealter[24m(1) for more details.

[1mBASIC DATA MANIPULATION[0m
       A number of programs exist to manipulate databases.  Complex functions
       can be made by stringing together commands with shell pipelines.  For
       example, to print the home directories of everyone with ``john'' in
       their names, you would do:

               cat DATA/passwd | dbrow '_fullname =~ /John/' | dbcol homedir

       The output might be:

               #fsdb homedir
               /home/johnh
               /home/greg
               # this is a simple database
               #  | dbrow _fullname =~ /John/
               #  | dbcol homedir

       (Notice that comments are appended to the output listing each command,
       providing an automatic audit log.)

       In addition to typical database functions (select, join, etc.) there
       are also a number of statistical functions.

       The real power of Fsdb is that one can apply arbitary code to rows to
       do powerful things.

               cat DATA/passwd | dbroweval '_fullname =~ s/(\w+)_(\w+)/$2,_$1/'

       converts "John_Heidemann" into "Heidemann,_John".  Not too much more
       work could split fullname into firstname and lastname fields.

[1mTALKING ABOUT COLUMNS[0m
       An advantage of Fsdb is that you can talk about columns by name
       (symbolically) rather than simply by their positions.  So in the above
       example, "dbcol homedir" pulled out the home directory column, and
       "dbrow '_fullname =~ /John/'" matched against column fullname.

       In general, you can use the name of the column listed on the "#fsdb"
       line to identify it in most programs, and _name to identify it in code.

       Some alternatives for flexibility:

       +o   Numeric values identify columns positionally, numbering from 0.  So
           0 or _0 is the first column, 1 is the second, etc.

       +o   In code, _last_columnname gets the value from columname's previous
           row.

       See [4mdbroweval[24m(1) for more details about writing code.

[1mLIST OF COMMANDS[0m
       Enough said.  I'll summarize the commands, and then you can experiment.
       For a detailed description of each command, see a summary by running it
       with the argument "--help" (or "-?" if you prefer.)  Full manual pages
       can be found by running the command with the argument "--man", or
       running the Unix command "man dbcol" or whatever program you want.

   [1mTABLE CREATION[0m
       dbcolcreate
           add columns to a database

       dbcoldefine
           set the column headings for a non-Fsdb file

   [1mTABLE MANIPULATION[0m
       dbcol
           select columns from a table

       dbrow
           select rows from a table

       dbsort
           sort rows based on a set of columns

       dbjoin
           compute the natural join of two tables

       dbcolrename
           rename a column

       dbcolmerge
           merge two columns into one

       dbcolsplittocols
           split one column into two or more columns

       dbcolsplittorows
           split one column into multiple rows

       dbfilepivot
           "pivots" a file, converting multiple rows correponding to the same
           entity into a single row with multiple columns.

       dbfilevalidate
           check that db file doesn't have some common errors

   [1mCOMPUTATION AND STATISTICS[0m
       dbcolstats
           compute statistics over a column (mean,etc.,optionally median)

       dbmultistats
           group rows by some key value, then compute stats (mean, etc.) over
           each group (equivalent to dbmapreduce with dbcolstats as the
           reducer)

       dbmapreduce
           group rows (map) and then apply an arbitrary function to each group
           (reduce)

       dbrvstatdiff
           compare two samples distributions (mean/conf interval/T-test)

       dbcolmovingstats
           computing moving statistics over a column of data

       dbcolstatscores
           compute Z-scores and T-scores over one column of data

       dbcolpercentile
           compute the rank or percentile of a column

       dbcolhisto
           compute histograms over a column of data

       dbcolscorrelate
           compute the coefficient of correlation over several columns

       dbcolsregression
           compute linear regression and correlation for two columns

       dbrowaccumulate
           compute a running sum over a column of data

       dbrowcount
           count the number of rows (a subset of dbstats)

       dbrowdiff
           compute differences between a columns in each row of a table

       dbrowenumerate
           number each row

       dbroweval
           run arbitrary Perl code on each row

       dbrowuniq
           count/eliminate identical rows (like Unix [4muniq[24m(1))

       dbfilediff
           compare fields on rows of a file (something like Unix [4mdiff[24m(1))

   [1mOUTPUT CONTROL[0m
       dbcolneaten
           pretty-print columns

       dbfilealter
           convert between column or list format, or change the column
           separator

       dbfilestripcomments
           remove comments from a table

       dbformmail
           generate a script that sends form mail based on each row

   [1mCONVERSIONS[0m
       (These programs convert data into fsdb.  See their web pages for
       details.)

       cgi_to_db
           <http://stein.cshl.org/boulder/>

       combined_log_format_to_db
           <http://httpd.apache.org/docs/2.0/logs.html>

       html_table_to_db
           HTML tables to fsdb (assuming they're reasonably formatted).

       kitrace_to_db
           http://ficus-www.cs.ucla.edu/ficus-members/geoff/kitrace.html
           <http://ficus-www.cs.ucla.edu/ficus-members/geoff/kitrace.html>

       ns_to_db
           http://mash-www.cs.berkeley.edu/ns/ <http://mash-
           www.cs.berkeley.edu/ns/>

       tabdelim_to_db
           spreadsheet tab-delimited files to db

       tcpdump_to_db
           (see man [4mtcpdump[24m(8) on any reasonable system)

       xml_to_db
           XML input to fsdb, assuming they're very regular

       (And out of fsdb:)

       db_to_csv
           Comma-separated-value format from fsdb.

       db_to_html_table
           simple conversion of Fsdb to html tables

   [1mSTANDARD OPTIONS[0m
       Many programs have common options:

       [1m-? [22mor [1m--help[0m
           Show basic usage.

       [1m-N [22mon [1m--new-name[0m
           When a command creates a new column like dbrowaccumulate's "accum",
           this option lets one override the default name of that new column.

           Show basic usage.

       [1m-c FRACTION [22mor [1m--confidence FRACTION[0m
           Specify confidence interval FRACTION (dbcolstats, dbmultistats,
           etc.)

       [1m-C S [22mor "--element-separator S"
           Specify column separator S (dbcolsplittocols, dbcolmerge).

       [1m-d [22mor [1m--debug[0m
           Enable debugging (may be repeated for greater effect in some
           cases).

       [1m-a [22mor [1m--include-non-numeric[0m
           Compute stats over all data (treating non-numbers as zeros).  (By
           default, things that can't be treated as numbers are ignored for
           stats purposes)

       [1m-S [22mor [1m--pre-sorted[0m
           Assume the data is pre-sorted.  May be repeated to disable
           verification (saving a small amount of work).

       [1m-e E [22mor [1m--empty E[0m
           give value E as the value for empty (null) records

       [1m-i I [22mor [1m--input I[0m
           Input data from file I.

       [1m-o O [22mor [1m--output O[0m
           Write data out to file O.

       [1m--nolog[22m.
           Skip logging the program in a trailing comment.

       When giving Perl code (in dbrow and dbroweval) column names can be
       embedded if preceded by underscores.  Look at [4mdbrow[24m(1) or [4mdbroweval[24m(1)
       for examples.)

       Most programs run in constant memory and use temporary files if
       necessary.  Exceptions are dbcolneaten, dbcolpercentile, dbmapreduce,
       dbmultistats, dbrowsplituniq.

[1mANOTHER EXAMPLE[0m
       Take the raw data in "DATA/http_bandwidth", put a header on it
       ("dbcoldefine size bw"), took statistics of each category
       ("dbmultistats -k size bw"), pick out the relevant fields ("dbcol size
       mean stddev pct_rsd"), and you get:

               #fsdb size mean stddev pct_rsd
               1024    1.4962e+06      2.8497e+05      19.047
               10240   5.0286e+06      6.0103e+05      11.952
               102400  4.9216e+06      3.0939e+05      6.2863
               #  | dbcoldefine size bw
               #  | /home/johnh/BIN/DB/dbmultistats -k size bw
               #  | /home/johnh/BIN/DB/dbcol size mean stddev pct_rsd

       (The whole command was:

               cat DATA/http_bandwidth |
               dbcoldefine size |
               dbmultistats -k size bw |
               dbcol size mean stddev pct_rsd

       all on one line.)

       Then post-process them to get rid of the exponential notation by adding
       this to the end of the pipeline:

           dbroweval '_mean = sprintf("%8.0f", _mean); _stddev = sprintf("%8.0f", _stddev);'

       (Actually, this step is no longer required since dbcolstats now uses a
       different default format.)

       giving:

               #fsdb      size    mean    stddev  pct_rsd
               1024     1496200          284970        19.047
               10240    5028600          601030        11.952
               102400   4921600          309390        6.2863
               #  | dbcoldefine size bw
               #  | dbmultistats -k size bw
               #  | dbcol size mean stddev pct_rsd
               #  | dbroweval   { _mean = sprintf("%8.0f", _mean); _stddev = sprintf("%8.0f", _stddev); }

       In a few lines, raw data is transformed to processed output.

       Suppose you expect there is an odd distribution of results of one
       datapoint.  Fsdb can easily produce a CDF (cumulative distribution
       function) of the data, suitable for graphing:

           cat DB/DATA/http_bandwidth | \
               dbcoldefine size bw | \
               dbrow '_size == 102400' | \
               dbcol bw | \
               dbsort -n bw | \
               dbrowenumerate | \
               dbcolpercentile count | \
               dbcol bw percentile | \
               xgraph

       The steps, roughly: 1. get the raw input data and turn it into fsdb
       format, 2. pick out just the relevant column (for efficiency) and sort
       it, 3. for each data point, assign a CDF percentage to it, 4. pick out
       the two columns to graph and show them

[1mA GRADEBOOK EXAMPLE[0m
       The first commercial program I wrote was a gradebook, so here's how to
       do it with Fsdb.

       Format your data like DATA/grades.

               #fsdb name email id test1
               a a@ucla.example.edu 1 80
               b b@usc.example.edu 2 70
               c c@isi.example.edu 3 65
               d d@lmu.example.edu 4 90
               e e@caltech.example.edu 5 70
               f f@oxy.example.edu 6 90

       Or if your students have spaces in their names, use "-F S" and two
       spaces to separate each column:

               #fsdb -F S name email id test1
               alfred aho  a@ucla.example.edu  1  80
               butler lampson  b@usc.example.edu  2  70
               david clark  c@isi.example.edu  3  65
               constantine drovolis  d@lmu.example.edu  4  90
               debrorah estrin  e@caltech.example.edu  5  70
               sally floyd  f@oxy.example.edu  6  90

       To compute statistics on an exam, do

               cat DATA/grades | dbstats test1 |dblistize

       giving

               #fsdb -R C  ...
               mean:        77.5
               stddev:      10.84
               pct_rsd:     13.987
               conf_range:  11.377
               conf_low:    66.123
               conf_high:   88.877
               conf_pct:    0.95
               sum:         465
               sum_squared: 36625
               min:         65
               max:         90
               n:           6
               ...

       To do a histogram:

               cat DATA/grades | dbcolhisto -n 5 -g test1

       giving

               #fsdb low histogram
               65      *
               70      **
               75
               80      *
               85
               90      **
               #  | /home/johnh/BIN/DB/dbhistogram -n 5 -g test1

       Now you want to send out grades to the students by e-mail.  Create a
       form-letter (in the file [4mtest1.txt[24m):

               To: _email (_name)
               From: J. Random Professor <jrp@usc.example.edu>
               Subject: test1 scores

               _name, your score on test1 was _test1.
               86+   A
               75-85 B
               70-74 C
               0-69  F

       Generate the shell script that will send the mail out:

               cat DATA/grades | dbformmail test1.txt > test1.sh

       And run it:

               sh <test1.sh

       The last two steps can be combined:

               cat DATA/grades | dbformmail test1.txt | sh

       but I like to keep a copy of exactly what I send.

       At the end of the semester you'll want to compute grade totals and
       assign letter grades.  Both fall out of dbroweval.  For example, to
       compute weighted total grades with a 40% midterm/60% final where the
       midterm is 84 possible points and the final 100:

               dbcol -rv total |
               dbcolcreate total - |
               dbroweval '
                       _total = .40 * _midterm/84.0 + .60 * _final/100.0;
                       _total = sprintf("%4.2f", _total);
                       if (_final eq "-" || ( _name =~ /^_/)) { _total = "-"; };' |
               dbcolneaten

       If you got the data originally from a spreadsheet, save it in "tab-
       delimited" format and convert it with tabdelim_to_db (run
       tabdelim_to_db -? for examples).

[1mA PASSWORD EXAMPLE[0m
       To convert the Unix password file to db:

               cat /etc/passwd | sed 's/:/  /g'| \
                       dbcoldefine -F S login password uid gid gecos home shell \
                       >passwd.fsdb

       To convert the group file

               cat /etc/group | sed 's/:/  /g' | \
                       dbcoldefine -F S group password gid members \
                       >group.fsdb

       To show the names of the groups that div7-members are in (assuming DIV7
       is in the gecos field):

               cat passwd.fsdb | dbrow '_gecos =~ /DIV7/' | dbcol login gid | \
                       dbjoin -i - -i group.fsdb gid | dbcol login group

[1mSHORT EXAMPLES[0m
       Which Fsdb programs are the most complicated (based on number of test
       cases)?

               ls TEST/*.cmd | \
                       dbcoldefine test | \
                       dbroweval '_test =~ s@^TEST/([^_]+).*$@$1@' | \
                       dbrowuniq -c | \
                       dbsort -nr count | \
                       dbcolneaten

       (Answer: dbmapreduce, then dbcolstats, dbfilealter and dbjoin.)

       Stats on an exam (in $FILE, where $COLUMN is the name of the exam)?

               cat $FILE | dbcolstats -q 4 $COLUMN <$FILE | dblistize | dbstripcomments

               cat $FILE | dbcolhisto -g -n 20 $COLUMN | dbcolneaten | dbstripcomments

       Merging a the hw1 column from file hw1.fsdb into grades.fsdb assuing
       there's a common student id in column "id":

               dbcol id hw1 <hw1.fsdb >t.fsdb

               dbjoin -a -e - grades.fsdb t.fsdb id | \
                   dbsort  name | \
                   dbcolneaten >new_grades.fsdb

       Merging two fsdb files with the same rows:

               cat file1.fsdb file2.fsdb >output.fsdb

       or if you want to clean things up a bit

               cat file1.fsdb file2.fsdb | dbstripextraheaders >output.fsdb

       or if you want to know where the data came from

               for i in 1 2
               do
                       dbcolcreate source $i < file$i.fsdb
               done >output.fsdb

       (assumes you're using a Bourne-shell compatible shell, not csh).

[1mWARNINGS[0m
       As with any tool, one should (which means [4mmust[24m) understand the limits
       of the tool.

       All Fsdb tools should run in [4mconstant[24m [4mmemory[24m.  In some cases (such as
       [4mdbcolstats[24m with quartiles, where the whole input must be re-read),
       programs will spool data to disk if necessary.

       Most tools buffer one or a few lines of data, so memory will scale with
       the size of each line.  (So lines with many columns, or when columns
       have lots data, may cause larege memory consumption.)

       All Fsdb tools should run in constant or at worst "n log n" time.

       All Fsdb tools use normal Perl math routines for computation.  Although
       I make every attempt to choose numerically stable algorithms, normal
       rounding due to computer floating point approximations can result in
       inaccuracies when data spans a large range of precisions.  (See for
       example the [4mdbcolstats_extrema[24m test cases.)

       Any requirements and limitations of each Fsdb tool is documented on its
       manual page.

       If any Fsdb program violates these assumptions, that is a bug that
       should be documented on the tool's manual page or ideally fixed.

[1mHISTORY[0m
       There have been three versions of Fsdb; fsdb 1.0 is a complete re-write
       of the pre-1995 versions, and was distributed from 1995 to 2007.  Fsdb
       2.0 is a significant re-write of the 1.x versions for reasons described
       below.

       Fsdb (in its various forms) has been used extensively by its author
       since 1991.  Since 1995 it's been used by two other researchers at UCLA
       and several at ISI.  In February 1998 it was announced to the Internet.
       Since then it has found a few users, some outside where I work.

   [1mFsdb 2.0 Rationale[0m
       I've thought about fsdb-2.0 for many years, but it was started in
       earnest in 2007.  Fsdb-2.0 has the following goals:

       in-one-process processing
           While fsdb is great on the Unix command line as a pipeline between
           programs, it should [4malso[24m be possible to set it up to run in a
           single process.  And if it does so, it should be able to avoid
           serializing and deserializing (converting to and from text) data
           between each module.  (Accomplished in fsdb-2.0: see dbpipeline,
           although still needs tuning.)

       clean IO API
           Fsdb's roots go back to perl4 and 1991, so the fsdb-1.x library is
           very, very crufty.  More than just being ugly (but it was that
           too), this made things reading from one format file and writing to
           another the application's job, when it should be the library's.
           (Accomplished in fsdb-1.15 and improved in 2.0: see Fsdb::IO.)

       normalized module APIs
           Because fsdb modules were added as needed over 10 years, sometimes
           the module APIs became inconsistent.  (For example, the 1.x
           "dbcolcreate" required an empty value following the name of the new
           column, but other programs specify empty values with the "-e"
           argument.)  We should smooth over these inconsistencies.
           (Accomplished as each module was ported in 2.0 through 2.7.)

       everyone handles all input formats
           Given a clean IO API, the distinction between "colized" and
           "listized" fsdb files should go away.  Any program should be able
           to read and write files in any format.  (Accomplished in fsdb-2.1.)

       Fsdb-2.0 preserves backwards compatibility where possible, but breaks
       it where necessary to accomplish the above goals.  In August 2008,
       fsdb-2.7 was declared preferred over the 1.x versions.

   [1mContributors[0m
       Fsdb includes code ported from Geoff Kuenning
       ("Fsdb::Support::TDistribution").

       Fsdb contributors: Ashvin Goel [4mgoel@cse.oge.edu[24m, Geoff Kuenning
       [4mgeoff@fmg.cs.ucla.edu[24m, Vikram Visweswariah [4mvisweswa@isi.edu[24m, Kannan
       Varadahan [4mkannan@isi.edu[24m, Lars Eggert [4mlarse@isi.edu[24m, Arkadi Gelfond
       [4markadig@dyna.com[24m, David Graff [4mgraff@ldc.upenn.edu[24m, Haobo Yu
       [4mhaoboy@packetdesign.com[24m, Pavlin Radoslavov [4mpavlin@catarina.usc.edu[24m,
       Graham Phillips, Yuri Pradkin, Alefiya Hussain, Ya Xu, Fabio Silva
       [4mfabio@isi.edu[24m, Jerry Zhao [4mzhaoy@isi.edu[24m, Ning Xu [4mnxu@aludra.usc.edu[24m,
       Martin Lukac [4mmlukac@lecs.cs.ucla.edu[24m, Xue Cai, Michael McQuaid.

       Fsdb includes datasets contributed from NIST (DATA/nist_zarr13.fsdb),
       from
       <http://www.itl.nist.gov/div898/handbook/eda/section4/eda4281.htm>, the
       NIST/SEMATECH e-Handbook of Statistical Methods, section 1.4.2.8.1.
       Background and Data.  The source is public domain, and reproduced with
       permission.

[1mRELATED WORK[0m
       As stated in the introduction, Fsdb is an incompatible reimplementation
       of the ideas found in "/rdb".  By storing data in simple text files and
       processing it with pipelines it is easy to experiment (in the shell)
       and look at the output.  The original implementation of this idea was
       /rdb, a commercial product described in the book [4mUNIX[24m [4mrelational[0m
       [4mdatabase[24m [4mmanagement:[24m [4mapplication[24m [4mdevelopment[24m [4min[24m [4mthe[24m [4mUNIX[24m [4menvironment[24m by
       Rod Manis, Evan Schaffer, and Robert Jorgensen (and also at the web
       page <http://www.rdb.com/>).

       While Fsdb is inspired by Rdb, it includes no code from it, and Fsdb
       makes several different design choices.  In particular: rdb attempts to
       be closer to a "real" database, with provision for locking, file
       indexing.  Fsdb focuses on single user use and so eschews these
       choices.  Rdb also has some support for interactive editing.  Fsdb
       leaves editing to text editors like emacs or vi.

       In August, 2002 I found out Carlo Strozzi extended RDB with his package
       NoSQL <http://www.linux.it/~carlos/nosql/>.  According to Mr. Strozzi,
       he implemented NoSQL in awk to avoid the Perl start-up of RDB.
       Although I haven't found Perl startup overhead to be a big problem on
       my platforms (from old Sparcstation IPCs to 2GHz Pentium-4s), you may
       want to evaluate his system.  The Linux Journal has a description of
       NoSQL at <http://www.linuxjournal.com/article/3294>.  It seems quite
       similar to Fsdb.  Like /rdb, NoSQL supports indexing (not present in
       Fsdb).  Fsdb appears to have richer support for statistics, and, as of
       Fsdb-2.x, its support for Perl threading may support faster performance
       (one-process, less serialization and deserialization).

[1mRELEASE NOTES[0m
       Versions prior to 1.0 were released informally on my web page but were
       not announced.

   [1m0.0 1991[0m
       started for my own research use

   [1m0.1 26-May-94[0m
       first check-in to RCS

   [1m0.2 15-Mar-95[0m
       parts now require perl5

   [1m1.0, 22-Jul-97[0m
       adds autoconf support and a test script.

   [1m1.1, 20-Jan-98[0m
       support for double space field separators, better tests

   [1m1.2, 11-Feb-98[0m
       minor changes and release on comp.lang.perl.announce

   [1m1.3, 17-Mar-98[0m
       +o   adds median and quartile options to dbstats

       +o   adds dmalloc_to_db converter

       +o   fixes some warnings

       +o   dbjoin now can run on unsorted input

       +o   fixes a dbjoin bug

       +o   some more tests in the test suite

   [1m1.4, 27-Mar-98[0m
       +o   improves error messages (all should now report the program that
           makes the error)

       +o   fixed a bug in dbstats output when the mean is zero

   [1m1.5, 25-Jun-98[0m
       BUG FIX dbcolhisto, dbcolpercentile now handles non-numeric values like
       dbstats
       NEW dbcolstats computes zscores and tscores over a column
       NEW dbcolscorrelate computes correlation coefficients between two
       columns
       INTERNAL ficus_getopt.pl has been replaced by DbGetopt.pm
       BUG FIX all tests are now ``portable'' (previously some tests ran only
       on my system)
       BUG FIX you no longer need to have the db programs in your path (fix
       arose from a discussion with Arkadi Gelfond)
       BUG FIX installation no longer uses cp -f (to work on SunOS 4)

   [1m1.6, 24-May-99[0m
       NEW dbsort, dbstats, dbmultistats now run in constant memory (using tmp
       files if necessary)
       NEW dbcolmovingstats does moving means over a series of data
       NEW dbcol has a -v option to get all columns except those listed
       NEW dbmultistats does quartitles and medians
       NEW dbstripextraheaders now also cleans up bogus comments before the
       fist header
       BUG FIX dbcolneaten works better with double-space-separated data

   [1m1.7,  5-Jan-00[0m
       NEW dbcolize now detects and rejects lines that contain embedded copies
       of the field separator
       NEW configure tries harder to prevent people from improperly
       configuring/installing fsdb
       NEW tcpdump_to_db converter (incomplete)
       NEW tabdelim_to_db converter:  from spreadsheet tab-delimited files to
       db
       NEW mailing lists for fsdb are     "fsdb-announce@heidemann.la.ca.us"
       and  "fsdb-talk@heidemann.la.ca.us"
           To subscribe to either, send mail
           to    "fsdb-announce-request@heidemann.la.ca.us"   or
           "fsdb-talk-request@heidemann.la.ca.us"     with "subscribe" in the
           BODY of the message.

       BUG FIX dbjoin used to produce incorrect output if there were extra,
       unmatched values in the 2nd table. Thanks to Graham Phillips for
       providing a test case.
       BUG FIX the sample commands in the usage strings now all should
       explicitly include the source of data (typically from "cat foo.fsdb
       |").  Thanks to Ya Xu for pointing out this documentation deficiency.
       BUG FIX (DOCUMENTATION) dbcolmovingstats had incorrect sample output.

   [1m1.8, 28-Jun-00[0m
       BUG FIX header options are now preserved when writing with dblistize
       NEW dbrowuniq now optionally checks for uniqueness only on certain
       fields
       NEW dbrowsplituniq makes one pass through a file and splits it into
       separate files based on the given fields
       NEW converter for "crl" format network traces
       NEW anywhere you use arbitrary code (like dbroweval), _last_foo now
       maps to the last row's value for field _foo.
       OPTIMIZATION comment processing slightly changed so that dbmultistats
       now is much faster on files with lots of comments (for example, ~100k
       lines of comments and 700 lines of data!) (Thanks to Graham Phillips
       for pointing out this performance problem.)
       BUG FIX dbstats with median/quartiles now correctly handles singleton
       data points.

   [1m1.9,  6-Nov-00[0m
       NEW dbfilesplit, split a single input file into multiple output files
       (based on code contributed by Pavlin Radoslavov).
       BUG FIX dbsort now works with perl-5.6

   [1m1.10, 10-Apr-01[0m
       BUG FIX dbstats now handles the case where there are more n-tiles than
       data
       NEW dbstats now includes a -S option to optimize work on pre-sorted
       data (inspired by code contributed by Haobo Yu)
       BUG FIX dbsort now has a better estimate of memory usage when run on
       data with very short records (problem detected by Haobo Yu)
       BUG FIX cleanup of temporary files is slightly better

   [1m1.11,  2-Nov-01[0m
       BUG FIX dbcolneaten now runs in constant memory
       NEW dbcolneaten now supports "field specifiers" that allow some control
       over how wide columns should be
       OPTIMIZATION dbsort now tries hard to be filesystem cache-friendly
       (inspired by "Information and Control in Gray-box Systems" by the
       Arpaci-Dusseau's at SOSP 2001)
       INTERNAL t_distr now ported to perl5 module DbTDistr

   [1m1.12,  30-Oct-02[0m
       BUG FIX dbmultistats documentation typo fixed
       NEW dbcolmultiscale
       NEW dbcol has -r option for "relaxed error checking"
       NEW dbcolneaten has new -e option to strip end-of-line spaces
       NEW dbrow finally has a -v option to negate the test
       BUG FIX math bug in dbcoldiff fixed by Ashvin Goel (need to check
       Scheaffer test cases)
       BUG FIX some patches to run with Perl 5.8. Note: some programs
       (dbcolmultiscale, dbmultistats, dbrowsplituniq) generate warnings like:
       "Use of uninitialized value in concatenation (.)" or "string at
       /usr/lib/perl5/5.8.0/FileCache.pm line 98, <STDIN> line 2". Please
       ignore this until I figure out how to suppress it. (Thanks to Jerry
       Zhao for noticing perl-5.8 problems.)
       BUG FIX fixed an autoconf problem where configure would fail to find a
       reasonable prefix (thanks to Fabio Silva for reporting the problem)
       NEW db_to_html_table: simple conversion to html tables (NO fancy stuff)
       NEW dblib now has a function [4mdblib_text2html()[24m that will do simple
       conversion of iso-8859-1 to HTML

   [1m1.13,  4-Feb-04[0m
       NEW fsdb added to the freebsd ports tree
       <http://www.freshports.org/databases/fsdb/>.  Maintainer:
       "larse@isi.edu"
       BUG FIX properly handle trailing spaces when data must be numeric (ex.
       dbstats with -FS, see test dbstats_trailing_spaces). Fix from Ning Xu
       "nxu@aludra.usc.edu".
       NEW dbcolize error message improved (bug report from Terrence Brannon),
       and list format documented in the README.
       NEW cgi_to_db converts CGI.pm-format storage to fsdb list format
       BUG FIX handle numeric synonyms for column names in dbcol properly
       ENHANCEMENT "talking about columns" section added to README. Lack of
       documentation pointed out by Lars Eggert.
       CHANGE dbformmail now defaults to using Mail ("Berkeley Mail") to send
       mail, rather than sendmail (sendmail is still an option, but mail
       doesn't require running as root)
       NEW on platforms that support it (i.e., with perl 5.8), fsdb works fine
       with unicode
       NEW dbfilevalidate: check a db file for some common errors

   [1m1.14,  24-Aug-06[0m
       ENHANCEMENT README cleanup
       INCOMPATIBLE CHANGE dbcolsplit renamed dbcolsplittocols
       NEW dbcolsplittorows  split one column into multiple rows
       NEW dbcolsregression compute linear regression and correlation for two
       columns
       ENHANCEMENT cvs_to_db: better error handling, normalize field names,
       skip blank lines
       ENHANCEMENT dbjoin now detects (and fails) if non-joined files have
       duplicate names
       BUG FIX minor bug fixed in calculation of Student t-distributions
       (doesn't change any test output, but may have caused small errors)

   [1m1.15, 12-Nov-07[0m
       NEW fsdb-1.14 added to the MacOS Fink system
       <http://pdb.finkproject.org/pdb/package.php/fsdb>. (Thanks to Lars
       Eggert for maintaining this port.)
       NEW Fsdb::IO::Reader and Fsdb::IO::Writer now provide reasonably clean
       OO I/O interfaces to Fsdb files.  Highly recommended if you use fsdb
       directly from perl.  In the fullness of time I expect to reimplement
       the entire thing using these APIs to replace the current dblib.pl which
       is still hobbled by its roots in perl4.
       NEW dbmapreduce now implements a Google-style map/reduce abstraction,
       generalizing dbmultistats.
       ENHANCEMENT fsdb now uses the Perl build system (Makefile.PL, etc.),
       instead of autoconf.  This change paves the way to better perl-5-style
       modularization, proper manual pages, input of both listize and colize
       format for every program, and world peace.
       ENHANCEMENT dblib.pl is now moved to Fsdb::Old.pm.
       BUG FIX dbmultistats now propagates its format argument (-f). Bug and
       fix from Martin Lukac (thanks!).
       ENHANCEMENT dbformmail documentation now is clearer that it doesn't
       send the mail, you have to run the shell script it writes.  (Problem
       observed by Unkyu Park.)
       ENHANCEMENT adapted to autoconf-2.61 (and then these changes were
       discarded in favor of The Perl Way.
       BUG FIX dbmultistats memory usage corrected (O(# tags), not O(1))
       ENHANCEMENT dbmultistats can now optionally run with pre-grouped input
       in O(1) memory
       ENHANCEMENT dbroweval -N was finally implemented (eat comments)

   [1m2.0, 25-Jan-08[0m
       2.0, 25-Jan-08 --- a quiet 2.0 release (gearing up towards complete)

       ENHANCEMENT: shifting old programs to Perl modules, with the front-end
       program as just a wrapper. In the short-term, this change just means
       programs have real man pages. In the long-run, it will mean that one
       can run a pipeline in a single Perl program. So far: dbcol, dbroweval,
       the new dbrowcount. dbsort the new dbmerge, the old "dbstats" (renamed
       dbcolstats), dbcolrename, dbcolcreate,
       NEW: Fsdb::Filter::dbpipeline is an internal-only module that lets one
       use fsdb commands from within perl (via threads).
           It also provides perl function aliases for the internal modules, so
           a string of fsdb commands in perl are nearly as terse as in the
           shell:

               use Fsdb::Filter::dbpipeline qw(:all);
               dbpipeline(
                   dbrow(qw(name test1)),
                   dbroweval('_test1 += 5;')
               );

       INCOMPATIBLE CHANGE: The old dbcolstats has been renamed
       dbcolstatscores. The new dbcolstats does the same thing as the old
       dbstats. This incompatibility is unfortunate but normalizes program
       names.
       CHANGE: The new dbcolstats program always outputs "-" (the default
       empty value) for statistics it cannot compute (for example, standard
       deviation if there is only one row), instead of the old mix of "-" and
       "na".
       INCOMPATIBLE CHANGE: The old dbcolstats program, now called
       dbcolstatscores, also has different arguments.  The "-t mean,stddev"
       option is now "--tmean mean --tstddev stddev".  See dbcolstatscores for
       details.
       INCOMPATIBLE CHANGE: dbcolcreate now assumes all new columns get the
       default value rather than requiring each column to have an initial
       constant value. To change the initial value, sue the new "-e" option.
       NEW: dbrowcount counts rows, an almost-subset of dbcolstats's "n"
       output (except without differentiating numeric/non-numeric input), or
       the equivalent of "dbstripcomments | wc -l".
       NEW: dbmerge merges two sorted files. This functionality was previously
       embedded in dbsort.
       INCOMPATIBLE CHANGE: dbjoin's "-i" option to include non-matches is now
       renamed "-a", so as to not conflict with the new standard option "-i"
       for input file.

   [1m2.1,  6-Apr-08[0m
       2.1,  6-Apr-08 --- another alpha 2.0, but now all converted programs
       understand both listize and colize format

       ENHANCEMENT: shifting more old programs to Perl modules. New in 2.1:
       dbcolneaten, dbcoldefine, dbcolhisto, dblistize, dbcolize, dbrecolize
       ENHANCEMENT dbmerge now handles an arbitrary number of input files, not
       just exactly two.
       NEW dbmerge2 is an internal routine that handles merging exactly two
       files.
       INCOMPATIBLE CHANGE dbjoin now specifies inputs like dbmerge2, rather
       than assuming the first two arguments were tables (as in fsdb-1).
           The old dbjoin argument "-i" is now "-a" or <--type=outer>.

           A minor change: comments in the source files for dbjoin are now
           intermixed with output rather than being delayed until the end.

       ENHANCEMENT dbsort now no longer produces warnings when null values are
       passed to numeric comparisons.
       BUG FIX dbroweval now once again works with code that lacks a trailing
       semicolon. (This bug fixes a regression from 1.15.)
       INCOMPATIBLE CHANGE dbcolneaten's old "-e" option (to avoid end-of-line
       spaces) is now "-E" to avoid conflicts with the standard empty field
       argument.
       INCOMPATIBLE CHANGE dbcolhisto's old "-e" option is now "-E" to avoid
       conflicts. And its "-n", "-s", and "-w" are now "-N", "-S", and "-W" to
       correspond.
       NEW dbfilealter replaces dbrecolize, dblistize, and dbcolize, but with
       different options.
       ENHANCEMENT The library routines "Fsdb::IO" now understand both list-
       format and column-format data, so all converted programs can now
       [4mautomatically[24m read either format.  This capability was one of the
       milestone goals for 2.0, so yea!

   [1m2.2, 23-May-08[0m
       Release 2.2 is another 2.x alpha release.  Now [4mmost[24m of the commands are
       ported, but a few remain, and I plan one last incompatible change (to
       the file header) before 2.x final.

       ENHANCEMENT
           shifting more old programs to Perl modules.  New in 2.2:
           dbrowaccumulate, dbformmail.  dbcolmovingstats.  dbrowuniq.
           dbrowdiff.  dbcolmerge.  dbcolsplittocols.  dbcolsplittorows.
           dbmapreduce.  dbmultistats.  dbrvstatdiff.  Also dbrowenumerate
           exists only as a front-end (command-line) program.

       INCOMPATIBLE CHANGE
           The following programs have been dropped from fsdb-2.x:
           dbcoltighten, dbfilesplit, dbstripextraheaders,
           dbstripleadingspace.

       NEW combined_log_format_to_db to convert Apache logfiles

       INCOMPATIBLE CHANGE
           Options to dbrowdiff are now [1m-B [22mand [1m-I[22m, not [1m-a [22mand [1m-i[22m.

       INCOMPATIBLE CHANGE
           dbstripcomments is now dbfilestripcomments.

       BUG FIXES
           dbcolneaten better handles empty columns; dbcolhisto warning
           suppressed (actually a bug in high-bucket handling).

       INCOMPATIBLE CHANGE
           dbmultistats now requires a "-k" option in front of the key (tag)
           field, or if none is given, it will group by the first field (both
           like dbmapreduce).

       KNOWN BUG
           dbmultistats with quantile option doesn't work currently.

       INCOMPATIBLE CHANGE
           dbcoldiff is renamed dbrvstatdiff.

       BUG FIXES
           dbformmail was leaving its log message as a  command, not a
           comment.  Oops.  No longer.

   [1m2.3, 27-May-08 (alpha)[0m
       Another alpha release, this one just to fix the critical dbjoin bug
       listed below (that happens to have blocked my MP3 jukebox :-).

       BUG FIX
           Dbsort no longer hangs if given an input file with no rows.

       BUG FIX
           Dbjoin now works with unsorted input coming from a pipeline (like
           stdin).  Perl-5.8.8 has a bug (?) that was making this case
           fail---opening stdin in one thread, reading some, then reading more
           in a different thread caused an lseek which works on files, but
           fails on pipes like stdin.  Go figure.

       BUG FIX / KNOWN BUG
           The dbjoin fix also fixed dbmultistats -q (it now gives the right
           answer).  Although a new bug appeared, messages like:
               Attempt to free unreferenced scalar: SV 0xa9dd0c4, Perl
           interpreter: 0xa8350b8 during global destruction.  So the
           dbmultistats_quartile test is still disabled.

   [1m2.4, 18-Jun-08[0m
       Another alpha release, mostly to fix minor usability problems in
       dbmapreduce and client functions.

       ENHANCEMENT
           dbrow now defaults to running user supplied code without warnings
           (as with fsdb-1.x).  Use "--warnings" or "-w" to turn them back on.

       ENHANCEMENT
           dbroweval can now write different format output than the input,
           using the "-m" option.

       KNOWN BUG
           dbmapreduce emits warnings on perl 5.10.0 about "Unbalanced string
           table refcount" and "Scalars leaked" when run with an external
           program as a reducer.

           dbmultistats emits the warning "Attempt to free unreferenced
           scalar" when run with quartiles.

           In each case the output is correct.  I believe these can be
           ignored.

       CHANGE
           dbmapreduce no longer logs a line for each reducer that is invoked.

   [1m2.5, 24-Jun-08[0m
       Another alpha release, fixing more minor bugs in "dbmapreduce" and
       lossage in "Fsdb::IO".

       ENHANCEMENT
           dbmapreduce can now tolerate non-map-aware reducers that pass back
           the key column in put.  It also passes the current key as the last
           argument to external reducers.

       BUG FIX
           Fsdb::IO::Reader, correctly handle "-header" option again.  (Broken
           since fsdb-2.3.)

   [1m2.6, 11-Jul-08[0m
       Another alpha release, needed to fix DaGronk.  One new port, small bug
       fixes, and important fix to dbmapreduce.

       ENHANCEMENT
           shifting more old programs to Perl modules.  New in 2.2:
           dbcolpercentile.

       INCOMPATIBLE CHANGE and ENHANCEMENTS dbcolpercentile arguments changed,
       use "--rank" to require ranking instead of "-r". Also, "--ascending"
       and "--descending" can now be specified separately, both for
       "--percentile" and "--rank".
       BUG FIX
           Sigh, the sense of the --warnings option in dbrow was inverted.  No
           longer.

       BUG FIX
           I found and fixed the string leaks (errors like "Unbalanced string
           table refcount" and "Scalars leaked") in dbmapreduce and
           dbmultistats.  (All "IO::Handle"s in threads must be manually
           destroyed.)

       BUG FIX
           The "-C" option to specify the column separator in dbcolsplittorows
           now works again (broken since it was ported).

       2.7, 30-Jul-08 beta

       The beta release of fsdb-2.x.  Finally, all programs are ported.  As
       statistics, the number of lines of non-library code doubled from 7.5k
       to 15.5k.  The libraries are much more complete, going from 866 to 5164
       lines.  The overall number of programs is about the same, although 19
       were dropped and 11 were added.  The number of test cases has grown
       from 116 to 175.  All programs are now in perl-5, no more shell scripts
       or perl-4.  All programs now have manual pages.

       Although this is a major step forward, I still expect to rename "fsdb"
       to "fsdb".

       ENHANCEMENT
           shifting more old programs to Perl modules.  New in 2.7:
           dbcolscorellate.  dbcolsregression.  cgi_to_db.  dbfilevalidate.
           db_to_csv.  csv_to_db, db_to_html_table, kitrace_to_db,
           tcpdump_to_db, tabdelim_to_db, ns_to_db.

       INCOMPATIBLE CHANGE
           The following programs have been dropped from fsdb-2.x: db2dcliff,
           dbcolmultiscale, crl_to_db.  ipchain_logs_to_db.  They may come
           back, but seemed overly specialized.  The following program
           dbrowsplituniq was dropped because it is superseded by dbmapreduce.
           dmalloc_to_db was dropped pending a test cases and examples.

       ENHANCEMENT
           dbfilevalidate now has a "-c" option to correct errors.

       NEW html_table_to_db provides the inverse of db_to_html_table.

   [1m2.8,  5-Aug-08[0m
       Change header format, preserving forwards compatibility.

       BUG FIX
           Complete editing pass over the manual, making sure it aligns with
           fsdb-2.x.

       SEMI-COMPATIBLE CHANGE
           The header of fsdb files has changed, it is now #fsdb, not #h (or
           #L) and parsing of -F and -R are also different.  See dbfilealter
           for the new specification.  The v1 file format will be read,
           compatibly, but not written.

       BUG FIX
           dbmapreduce now tolerates comments that preceed the first key,
           instead of failing with an error message.

   [1m2.9, 6-Aug-08[0m
       Still in beta; just a quick bug-fix for dbmapreduce.

       ENHANCEMENT
           dbmapreduce now generates plausible output when given no rows of
           input.

   [1m2.10, 23-Sep-08[0m
       Still in beta, but picking up some bug fixes.

       ENHANCEMENT
           dbmapreduce now generates plausible output when given no rows of
           input.

       ENHANCEMENT
           dbroweval the warnings option was backwards; now corrected.  As a
           result, warnings in user code now default off (like in fsdb-1.x).

       BUG FIX
           dbcolpercentile now defaults to assuming the target column is
           numeric.  The new option "-N" allows selectin of a non-numeric
           target.

       BUG FIX
           dbcolscorrelate now includes "--sample" and "--nosample" options to
           compute the sample or full population correlation coefficients.
           Thanks to Xue Cai for finding this bug.

   [1m2.11, 14-Oct-08[0m
       Still in beta, but picking up some bug fixes.

       ENHANCEMENT
           html_table_to_db is now more agressive about filling in empty cells
           with the official empty value, rather than leaving them blank or as
           whitespace.

       ENHANCEMENT
           dbpipeline now catches failures during pipeline element setup and
           exits reasonably gracefully.

       BUG FIX
           dbsubprocess now reaps child prcoesses, thus avoiding running out
           of processes when used a lot.

   [1m2.12, 16-Oct-08[0m
       Finally, a full (non-beta) 2.x release!

       INCOMPATIBLE CHANGE
           Jdb has been renamed Fsdb, the flatfile-streaming database.  This
           change affects all internal Perl APIs, but no shell command-level
           APIs.  While Jdb served well for more than ten years, it is easily
           confused with the Java debugger (even though Jdb was there first!).
           It also is too generic to work well in web search engines.
           Finally, Jdb stands for ``John's database'', and we're a bit beyond
           that.  (However, some call me the ``file-system guy'', so one could
           argue it retains that meeting.)

           If you just used the shell commands, this change should not affect
           you.  If you used the Perl-level libraries directly in your code,
           you should be able to rename "Jdb" to "Fsdb" to move to 2.12.

           The jdb-announce list not yet been renamed, but it will be shortly.

           With this release I've accomplished everything I wanted to in
           fsdb-2.x.  I therefore expect to return to boring, bugfix releases.

   [1m2.13, 30-Oct-08[0m
       BUG FIX
           dbrowaccumulate now treats non-numeric data as zero by default.

       BUG FIX
           Fixed a perl-5.10ism in dbmapreduce that breaks that program under
           5.8.  Thanks to Martin Lukac for reporting the bug.

   [1m2.14, 26-Nov-08[0m
       BUG FIX
           Improved documentation for dbmapreduce's "-f" option.

       ENHANCEMENT
           dbcolmovingstats how computes a moving standard deviation in
           addition to a moving mean.

   [1m2.15, 13-Apr-09[0m
       BUG FIX
           Fix a [4mmake[24m [4minstall[24m bug reported by Shalindra Fernando.

   [1m2.16, 14-Apr-09[0m
       BUG FIX
           Another minor release bug: on some systems [4mprogramize_module[24m looses
           executable permissions.  Again reported by Shalindra Fernando.

   [1m2.17, 25-Jun-09[0m
       TYPO FIXES
           Typo in the [4mdbroweval[24m manual fixed.

       IMPROVEMENT
           There is no longer a comment line to label columns in [4mdbcolneaten[24m,
           instead the header line is tweaked to line up.  This change
           restores the Jdb-1.x behavior, and means that repeated runs of
           dbcolneaten no longer add comment lines each time.

       BUG FIX
           It turns out  [4mdbcolneaten[24m was not correctly handling trailing
           spaces when given the "-E" option to suppress them.  This
           regression is now fixed.

       EXTENSION
           [4mdbroweval[24m(1) can now handle direct references to the last row via
           [4m$lfref[24m, a dubious but now documented feature.

       BUG FIXES
           Separators set with "-C" in [4mdbcolmerge[24m and [4mdbcolsplittocols[24m were
           not properly setting the heading, and null fields were not
           recognized.  The first bug was reported by Martin Lukac.

   [1m2.18,  1-Jul-09  A minor release[0m
       IMPROVEMENT
           Documentation for [4mFsdb::IO::Reader[24m has been improved.

       IMPROVEMENT
           The package should now be PGP-signed.

   [1m2.19,  10-Jul-09[0m
       BUG FIX
           Internal improvements to debugging output and robustness of
           [4mdbmapreduce[24m and [4mdbpipeline[24m.  [4mTEST/dbpipeline_first_fails.cmd[24m re-
           enabled.

   [1m2.20, 30-Nov-09 (A collection of minor bugfixes, plus a build against[0m
       [1mFedora 12.)[0m
       BUG FIX
           Loging for [4mdbmapreduce[24m with code refs is now stable (it no longer
           includes a hex pointer to the code reference).

       BUG FIX
           Better handling of mixed blank lines in [4mFsdb::IO::Reader[24m (see test
           case [4mdbcolize_blank_lines.cmd[24m).

       BUG FIX
           [4mhtml_table_to_db[24m now handles multi-line input better, and handles
           tables with COLSPAN.

       BUG FIX
           [4mdbpipeline[24m now cleans up threads in an "eval" to prevent "cannot
           detach a joined thread" errors that popped up in perl-5.10.
           Hopefully this prevents a race condition that causes the test
           suites to hang about 20% of the time (in [4mdbpipeline_first_fails[24m).

       IMPROVEMENT
           [4mdbmapreduce[24m now detects and correctly fails when the input and
           reducer have incompatible field seperators.

       IMPROVEMENT
           [4mdbcolstats[24m, [4mdbcolhisto[24m, [4mdbcolscorrelate[24m, [4mdbcolsregression[24m, and
           [4mdbrowcount[24m now all take an "-F" option to let one specify the
           output field seperator (so they work better with [4mdbmapreduce[24m).

       BUG FIX
           An omitted "-k" from the manual page of [4mdbmultistats[24m is now there.
           Bug reported by Unkyu Park.

   [1m2.21, 17-Apr-10 bug fix release[0m
       BUG FIX
           [4mFsdb::IO::Writer[24m now no longer fails with -outputheader => never
           (an obscure bug).

       IMPROVEMENT
           [4mFsdb[24m (in the warnings section) and [4mdbcolstats[24m now more carefully
           document how they handle (and do not handle) numerical precision
           problems, and other general limits.  Thanks to Yuri Pradkin for
           prompting this documentation.

       IMPROVEMENT
           "Fsdb::Support::fullname_to_sortkey" is now restored from "Jdb".

       IMPROVEMENT
           Documention for multiple styles of input approaches (including
           performance description) added to Fsdb::IO.

   [1m2.22, 2010-10-31 One new tool [4m[22mdbcolcopylast[24m [1mand several bug fixes for Perl[0m
       [1m5.10.[0m
       BUG FIX
           [4mdbmerge[24m now correctly handles n-way merges.  Bug reported by Yuri
           Pradkin.

       INCOMPATABLE CHANGE
           [4mdbcolneaten[24m now defaults to [4mnot[24m padding the last column.

       ADDITION
           [4mdbrowenumerate[24m now takes [1m-N NewColumn [22mto give the new column a name
           other than "count".  Feature reuested by Mike Rouch in January
           2005.

       ADDITION
           New program [4mdbcolcopylast[24m copies the last value of a column into a
           new column copylast_column of the next row.  New program requested
           by Fabio Silva; useful for convereting dbmultistats output into
           dbrvstatdiff input.

       BUG FIX
           Several tools (particularly [4mdbmapreduce[24m and [4mdbmultistats[24m) would
           report errors like "Unbalanced string table refcount: (1) for
           "STDOUT" during global destruction" on exit, at least on certain
           versions of Perl (for me on 5.10.1), but similar errors have been
           off-and-on for several Perl releases.  Although I think my code
           looked OK, I worked around this problem with a different way of
           handling standard IO redirection.

   [1m2.23, 2011-03-10 Several small portability bugfixes; improved [4m[22mdbcolstats[0m
       [1mfor large datsets[0m
       IMPROVEMENT
           Documentation to [4mdbrvstatdiff[24m was changed to use "sd" to refer to
           standard deviation, not "ss" (which might be confused with sum-of-
           squares).

       BUG FIX
           This documentation about [4mdbmultistats[24m was missing the [4m-k[24m option in
           some cases.

       BUG FIX
           [4mdbmapreduce[24m was failing on MacOS-10.6.3 for some tests with the
           error

               dbmapreduce: cannot run external dbmapreduce reduce program (perl TEST/dbmapreduce_external_with_key.pl)

           The problem seemed to be only in the error, not in operation.  On
           MacOS, the error is now suppressed.  Thanks to Alefiya Hussain for
           providing access to a Mac system that allowed debugging of this
           problem.

       IMPROVEMENT
           The [4mcsv_to_db[24m command requires an external Perl library
           ([4mText::CSV_XS[24m).  On computers that lack this optional library,
           previously Fsdb would configure with a warning and then test cases
           would fail.  Now those test cases are skipped with an additional
           warning.

       BUG FIX
           The test suite now supports alternative valid output, as a hack to
           account for last-digit floating point differences.  (Not very
           satisfying :-(

       BUG FIX
           [4mdbcolstats[24m output for confidence intervals on very large datasets
           has changed.  Previously it failed for more than 2^31-1 records,
           and handling of T-Distributions with thousands of rows was a bit
           dubious.  Now datasets with more than 10000 are considered
           infinitely large and hopefully correctly handled.

   [1m2.24, 2011-04-15 Improvements to fix an old bug in dbmapreduce with[0m
       [1mdifferent field separators[0m
       IMPROVEMENT
           The [4mdbfilealter[24m command had a "--correct" option to work-around
           from incompatible field-seperators, but it did nothing.  Now it
           does the correct but sad, data-loosing thing.

       IMPROVEMENT
           The [4mdbmultistats[24m command previously failed with an error message
           when invoked on input with a non-default field separator.  The root
           cause was the underlying [4mdbmapreduce[24m that did not handle the case
           of reducers that generated output with a different field separator
           than the input.  We now detect and repair incompatible field
           separators.  This change corrects a problem originally documented
           and detected in Fsdb-2.20.  Bug re-reported by Unkyu Park.

   [1m2.25, 2011-08-07 Two new tools, [4m[22mxml_to_db[24m [1mand [4m[22mdbfilepivot[24m[1m, and a bugfix for[0m
       [1mtwo people.[0m
       IMPROVEMENT
           [4mkitrace_to_db[24m now supports a [4m--utc[24m option, which also fixes this
           test case for users outside of the Pacific time zone.  Bug reported
           by David Graff, and also by Peter Desnoyers (within a week of each
           other :-)

       NEW [4mxml_to_db[24m can convert simple, very regular XML files into Fsdb.

       NEW [4mdbfilepivot[24m "pivots" a file, converting multiple rows correponding
           to the same entity into a single row with multiple columns.

   [1m2.26, 2011-12-12 Bug fixes, particularly for perl-5.14.2.[0m
       BUG FIX
           Bugs fixed in [4mFsdb::IO::Reader[24m(3) manual page.

       BUG FIX
           Fixed problems where dbcolstats was truncating floating point
           numbers when sorting.  This strange behavior happenes as of
           perl-5.14.2 and it [4mseems[24m like a Perl bug.  I've worked around it
           for the test suites, but I'm a bit nervous.

   [1m2.27, 2012-11-15 Accumulated bug fixes.[0m
       IMPROVEMENT
           [4mcsv_to_db[24m now reports errors in CVS input with real diagnostics.

       IMPROVEMENT
           [4mdbcolmovingstats[24m can now compute median, when given the "-m"
           option.

       BUG FIX
           [4mdbcolmovingstats[24m non-numeric handling (the "-a" option) now works
           properly.

       DOCUMENTATION
           The internal [4mt/test_command.t[24m test framework is now documented.

       BUG FIX
           [4mdbrowuniq[24m now corretly handles the case where there is no input
           (previously it output a blank line, which is a malformed fsdb
           file).  Thanks to Yuri Pradkin for reporting this bug.

   [1m2.28, 2012-11-15 A quick release to fix most rpmlint errors.[0m
       BUG FIX
           Fixed a number of minor release problems (wrong permissions, old
           FSF address, etc.) found by rpmlint.

   [1m2.29, 2012-11-20 a quick release for CPAN testing[0m
       IMPROVEMENT
           Tweaked the RPM spec.

       IMPROVEMENT
           Modified [4mMakefile.PL[24m to fail gracefully on Perl installations that
           lack threads.  (Without this fix, I get massive failures in the
           non-ithreads test system.)

   [1m2.30, 2012-11-25 imporovements to perl portability[0m
       BUG FIX
           Removed unicode character in documention of [4mdbcolscorrelated[24m so pod
           tests will pass.  (Sigh, that should work :-( )

       BUG FIX
           Fixed test suite failures on 5 tests ([4mdbcolcreate_double_creation[0m
           was the first) due to Carp's addition of a period.  This problem
           was breaking Fsdb on perl-5.17.  Thanks to Michael McQuaid for
           helping diagnose this problem.

       IMPROVEMENT
           The test suite now prints out the names of tests it tries.

   [1m2.31, 2012-11-28 A release with actual improvements to dbfilepivot and[0m
       [1mdbrowuniq.[0m
       BUG FIX
           Documentation fixes: typos in dbcolscorrelated, bugs in
           dbfilepivot, clarification for comment handling in
           Fsdb::IO::Reader.

       IMPROVEMENT
           Previously dbfilepivot assumed the input was grouped by keys and
           didn't very that pre-condition.  Now there is no pre-condition (it
           will sort the input by default), and it checks if the invariant is
           violated.

       BUG FIX
           Previously dbfilepivot failed if the input had comments (oops :-);
           no longer.

       IMPROVEMENT
           Now dbrowuniq has the "-L" option to preserve the last unique row
           (instead of the first), a common idiom.

   [1m2.32, 2012-12-21 Test suites should now be more numerically robust.[0m
       NEW New dbfilediff does fsdb-aware file differencing.  It does not do
           smart intuition of add/removes like Unix [4mdiff[24m(1), but it does know
           about columns, and with "-E", it does numeric-aware differences.

       IMPROVEMENT
           Test suites that are numeric now use dbfilediff to do numeric-aware
           comparisons, so the test suite should now be robust to slightly
           different computers and operating systems and complilers than
           [4mexactly[24m what I use.

   [1m2.33, 2012-12-23 Minor fixes to some test cases.[0m
       IMPROVEMENT
           dbfilediff and dbrowuniq now supports the "-N" option to give the
           new column a different name.  (And a test cases where this
           duplication mattered have been fixed.)

       IMPROVEMENT
           dbrvstatdiff now show the t-test breakpoint with a reasonable
           number of floating point digits.

       BUG FIX
           Fixed a numerical stability problem in the [4mdbroweval_last[24m test
           case.

[1mWHAT'S NEW[0m
   [1m2.34, 2013-02-10 Parallelism in dbmerge.[0m
       IMPROVEMENT
           Documention for dbjoin now includes resource requirements.

       IMPROVEMENT
           Default memory usage for dbsort is now about 256MB.  (The world
           keeps moving forward.)

       IMPROVEMENT
           dbmerge now does merging in parallel.  As a side-effect, dbsort
           should be faster when input overflows memory.  The level of
           parallelism can be limited with the "--parallelism" option.  (There
           is more work to do here, but we're off to a start.)

   [1m2.35, 2013-02-23 Improvements to dbmerge parallelism[0m
       BUG FIX
           Fsdb temporary files are now created more securely (with
           File::Temp).

       IMPROVEMENT
           Programs that sort or merge on fields (dbmerge2, dbmerge, dbsort,
           dbjoin) now report an error if no fields on which to join or merge
           are given.

       IMPROVEMENT
           Parallelism in dbmerge is should now be more consistent, with less
           starting and stopping.

       IMPROVEMENT In dbmerge, the "--xargs" option lets one give input
       filenames on standard input, rather than the command line. This feature
       paves the way for faster dbsort for large inputs (by pipelining sorting
       and merging), expected in the next release.

   [1m2.36, 2013-02-25 dbsort pipelines with dbmerge[0m
       IMPROVEMENT For large inputs, dbsort now pipelines sorting and merging,
       allowing earlier processing.
       BUG FIX Since 2.35, dbmerge delayed cleanup of intermediate files,
       thereby requiring extra disk space.

[1mAUTHOR[0m
       John Heidemann, "johnh@isi.edu"

       See "Contributors" for the many people who have contributed bug reports
       and fixes.

[1mCOPYRIGHT[0m
       Fsdb is Copyright (C) 1991-2013 by John Heidemann <johnh@isi.edu>.

       This program is free software; you can redistribute it and/or modify it
       under the terms of version 2 of the GNU General Public License as
       published by the Free Software Foundation.

       This program is distributed in the hope that it will be useful, but
       WITHOUT ANY WARRANTY; without even the implied warranty of
       MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
       General Public License for more details.

       You should have received a copy of the GNU General Public License along
       with this program; if not, write to the Free Software Foundation, Inc.,
       675 Mass Ave, Cambridge, MA 02139, USA.

       A copy of the GNU General Public License can be found in the file
       ``COPYING''.

[1mCOMMENTS and BUG REPORTS[0m
       Any comments about these programs should be sent to John Heidemann
       "johnh@isi.edu".



perl v5.16.2                      2013-02-26                           Fsdb(3)
