[4mFsdb[24m(3)                User Contributed Perl Documentation               [4mFsdb[24m(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:

     *   extracting measurements from experimental output

     *   examining data to address different hypotheses

     *   joining data from different experiments

     *   eliminating/detecting outliers

     *   computing statistics on data (mean, confidence intervals, correlations,
         histograms)

     *   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 Fsdb:

     *   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.

     *   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.

     *   A series of actions are self-documenting (the provenance of processsing
         done to produce each output is recorded in comments).

         **  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

     *   The library is mature, supporting large datasets (more than 100GB),
         parallelism, 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.

         **  Makes use of multiple cores in your computer when it can, because
             each pipeline component runs in parallel, and because key tools
             (dbsort, dbmapreduce) run in parallel when possible.

     *   Fsdb 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;'));

     *   Fsdb-3.x supports optional typing, allowing languages that require
         casting on import to do so automatically.

     (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[24m [4mdevelopment[24m [4min[0m
     [4mthe[24m [4mUNIX[24m [4menvironment[24m by Rod Manis, Evan Schaffer, and Robert Jorgensen
     (1988 by Prentice Hall, 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 will exploit multiple processors or cores, and provides Perl-level
     support for input, output, and threaded-pipelines.  (As of Fsdb-2.44 it no
     longer uses Perl threading, just processes, since they are faster.)

     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
   [1m3.16, 2026-06-08 Make dbcoldefine default to fill in null fields with the[0m
     [1mFsdb empty value.[0m
     ENHANCEMENT
         dbcoldefine now fills columns that have no value with the Fsdb empty
         value.  Use "--no-fill-empty" to disable.

     ENHANCEMENT
         tcpdump_to_db now handles epoch-style time formats and UDP, and
         optionally splits out ports with "--ports".  (But: tcpdump is not
         really parsable; use tshark instead.)

[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
     [1mExtUtil::MakeMaker[22m(3), so the quick answer to installation is to type:

         perl Makefile.PL
         make
         make test
         sudo make install

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

         perl Makefile.PL PREFIX=$HOME

     then the other commands ("make; make test; make install"; but now without
     the sudo), and it will go in your home directory's [4mbin[24m, etc.  (See
     [1mExtUtil::MakeMaker[22m(3) for more details.)

     Fsdb requires perl 5.8 or later.

     A test-suite is available, run it with

         make test

     In the past, the ports existed for FreeBSD and MacOS.  If someone running
     one of those OSes wants to contribute a new port, please let me know.

[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 "#fsdb".  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).

     Columns can optionally include type anntations by following name with :t
     where t is some type.  (Types are not used in Perl, but are relevant in
     Python and Go Fsdb bindings.)  Types use a subset of perl pack specifiers:
     c, s, l, q are signed 8, 16, 32, and 64-bit integers, f is a float, d is
     double float, a is utf-8 string, and &gt; and &lt; can force big or little
     endianness.

     By default, columns are delimited by any amount of 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 [1mdbfilealter[22m(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 [1mdbfilealter[22m(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 [1mdbfilealter[22m(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 arbitrary 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.

     (Or:

             cat DATA/passwd | dbcolcreate sort | dbroweval -b 'use Fsdb::Support'
                     '_sort = _fullname; _sort =~ s/_/ /g; _sort = fullname_to_sort(_sort);'

[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:

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

     *   In code, _last_columnname gets the value from columname's previous row.

     See [1mdbroweval[22m(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 corresponding 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

     dbcolsdecimate
         drop rows selectively, keeping large changes and periodic samples

     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 [1muniq[22m(1))

     dbfilediff
         compare fields on rows of a file (something like Unix [1mdiff[22m(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>

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

     sqlselect_to_db
         the output of SQL SELECT tables to db

     tabdelim_to_db
         spreadsheet tab-delimited files to db

     tcpdump_to_db
         (see man [1mtcpdump[22m(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.

     [1m-T TmpDir[0m
         where to put tmp files.  Also uses environment variable TMPDIR, if -T
         is not specified.  Default is /tmp.

         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--header [22mH
         Use H as the full Fsdb header, rather than reading a header from then
         input.  This option is particularly useful when using Fsdb under
         Hadoop, where split files don't have heades.

     [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 [1mdbrow[22m(1) or [1mdbroweval[22m(1) for
     examples.)

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

   [1mSTANDARD SORTING OPTIONS[0m
     A number of programs do sorting, or depend on defining an ordering of rows.
     Such programs use these standard sorting options:

     [1m-r [22mor [1m--descending[0m
         sort in reverse order (high to low)

     [1m-R [22mor [1m--ascending[0m
         sort in normal order (low to high)

     [1m-t [22mor [1m--type-inferred-sorting[0m
         sort fields by type (numeric or leicographic), automatically

     [1m-n [22mor [1m--numeric[0m
         sort numerically

     [1m-N [22mor [1m--lexical[0m
         sort lexicographically

[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 assuming
     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 large 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 (although I also
     welcome feedback and suggestions for improvement), normal rounding due to
     computer floating point approximations can result in inaccuracies when data
     spans a large range of precision.  (See for example the [4mdbcolstats_extrema[0m
     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.

     Fsdb does depend on Perl's correctness, and Perl (and Fsdb) have some bugs.
     Fsdb should work on perl from version 5.10 onward.

[1mHISTORY[0m
     There have been four major versions of Fsdb: fsdb-0.x was begun in 1991 for
     my personal use.  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 to systematically use a library and threads (although
     threads were replaced with full processes in 2.44).  Fsdb 3.0 in 2022 adds
     type specifiers to the schema, mostly to support use in languages with
     stronger typing (like Python, Go, and C).

     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.

     Major changes:

     0.1 1991: begun for my personal use, to replace awk.
     1.0 1997-07-22: first public release.
     2.0 2008-01-25: rewrite to use a common library, and starting to use
     threads.
     2.12 2008-10-16: completion of the rewrite, and first RPM package.
     2.44 2013-10-02: replacing threads with processes for improved performance
     3.0 2022-04-04: adding type specifiers to the schema

   [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.  Benchmarking in 2013 showed
     that threading performed much worse than just using pipes, because Perl's
     requirements for data that is shared between multiple threads is quite
     heavyweight.  Fsdb-2.44 therefore uses threading "style", but implemented
     with processes (via my "Freds" library).

   [1mFsdb And Muliple Processors[0m
     Fsdb's use of Unix pipelines means Fsdb automatically benefits for
     multiprocessor computers---each pipeline stage can run on a separate core.
     In addition, compute-intensive Fsdb modules like dbsort and dbmapreduce are
     explicitly multi-process and will use as many cores as they can, up to the
     number of cores on the local computer.

     Although Fsdb takes advanatage of as much parallelism as it can, a five
     stage pipeline won't necessarily saturate five cores.  Pipeline stages
     almost always have different amounts of work to do, and some stages are
     often data limited.  (Dbsort is attempts as much parallelism as it can, and
     can run 10-way parallel or more over a large enough input dataset.  But it
     cannot sustain high parallelism because of the requirement that it produce
     one global output.)

   [1mFsdb 3.0 Rationale[0m
     There are two motiviations for adding optional typing to Fsdb.  First,
     languages such as Python and Go would really like type information.  As of
     2022 there are now users of those languages, so the basic system should
     support them.

     Second, while pure text is flexible, it's very inefficient---converting
     numbers to and from decimal is thousands of instructions, and binary
     encodings are often much smaller than text.  In the future, I would love to
     have a flag that enables a binary encoding.

     Typing is optional---omitting types is never wrong.

     One somewhat odd thing about typing is that we reuse the Perl pack
     definitions of types, so q (for "quadword") stands for 64-bit integer.
     These are perhaps not the most mnemonic choices in 2022, but I would rather
     pick someone's existing set than try to define my own.

   [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, Michael Schwendt, 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, Christopher
     Meng, Calvin Ardi, H. Merijn Brand, Lan Wei, Hang Guo, Wes Hardaker, Erica
     Stutz.

     Fsdb includes datasets contributed from NIST ([4mDATA/nist_zarr13.fsdb[24m), 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[24m [4mdatabase[0m
     [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 Perl start-up costs in 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
     *   adds median and quartile options to dbstats

     *   adds dmalloc_to_db converter

     *   fixes some warnings

     *   dbjoin now can run on unsorted input

     *   fixes a dbjoin bug

     *   some more tests in the test suite

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

     *   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 quartiles 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 [1mdblib_text2html() [22mthat 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[0m
     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 "jdb" 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 precede 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 selection 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 aggressive 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 processes, 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
         [1mdbroweval[22m(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[0m
         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 Fedora[0m
     [1m12.)[0m
     BUG FIX
         Logging 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 separators.

     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 separator (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 [4mdbcolcopylast[24m and several bug fixes for Perl[0m
     [1m5.10.[0m
     BUG FIX
         [4mdbmerge[24m now correctly handles n-way merges.  Bug reported by Yuri
         Pradkin.

     INCOMPARABLE 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 requested 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 converting 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 [4mdbcolstats[24m for[0m
     [1mlarge datasets[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 different[0m
     [1mfield separators[0m
     IMPROVEMENT
         The [4mdbfilealter[24m command had a "--correct" option to work-around from
         incompatible field-separators, 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, [4mxml_to_db[24m and [4mdbfilepivot[24m, 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 corresponding 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 [1mFsdb::IO::Reader[22m(3) manual page.

     BUG FIX
         Fixed problems where dbcolstats was truncating floating point numbers
         when sorting.  This strange behavior happens 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 correctly 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 improvements 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[24m 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 [1mdiff[22m(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 compilers than [4mexactly[0m
         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.

   [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.

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

   [1m2.38, 2013-04-29 minor bug fixes[0m
     CLARIFICATION
         Configure now rejects Windows since tests seem to hang on some versions
         of Windows.  (I would love help from a Windows developer to get this
         problem fixed, but I cannot do it.)  See
         [4mhttps://rt.cpan.org/Ticket/Display.html?id=84201[24m.

     IMPROVEMENT
         All programs that use temporary files (dbcolpercentile,
         dbcolscorrelate, dbcolstats, dbcolstatscores) now take the "-T" option
         and set the temporary directory consistently.

         In addition, error messages are better when the temporary directory has
         problems.  Problem reported by Liang Zhu.

     BUG FIX
         dbmapreduce was failing with external, map-reduce aware reducers (when
         invoked with -M and an external program).  (Sigh, did this case ever
         work?)  This case should now work.  Thanks to Yuri Pradkin for
         reporting this bug (in 2011).

     BUG FIX
         Fixed perl-5.10 problem with dbmerge.  Thanks to Yuri Pradkin for
         reporting this bug (in 2013).

   [1m2.39, 2013-05-31 quick release for the dbrowuniq extension[0m
     BUG FIX
         Actually in 2.38, the Fedora [4m.spec[24m got cleaner dependencies.
         Suggestion from Christopher Meng via
         <https://bugzilla.redhat.com/show_bug.cgi?id=877096>.

     ENHANCEMENT
         Fsdb files are now explicitly set into UTF-8 encoding, unless one
         specifies "-encoding" to "Fsdb::IO".

     ENHANCEMENT
         dbrowuniq now supports "-I" for incremental counting.

   [1m2.40, 2013-07-13 small bug fixes[0m
     BUG FIX
         dbsort now has more respect for a user-given temporary directory; it no
         longer is ignored for merging.

     IMPROVEMENT
         dbrowuniq now has options to output the first, last, and both first and
         last rows of a run ("-F", "-L", and "-B").

     BUG FIX
         dbrowuniq now correctly handles "-N".  Sigh, it didn't work before.

   [1m2.41, 2013-07-29 small bug and packaging fixes[0m
     ENHANCEMENT
         Documentation to dbrvstatdiff improved (inspired by questions from Qian
         Kun).

     BUG FIX
         dbrowuniq no longer duplicates singleton unique lines when outputting
         both (with "-B").

     BUG FIX
         Add missing "XML::Simple" dependency to [4mMakefile.PL[24m.

     ENHANCEMENT
         Tests now show the diff of the failing output if run with "make test
         TEST_VERBOSE=1".

     ENHANCEMENT
         dbroweval now includes documentation for how to output extra rows.
         Suggestion from Yuri Pradkin.

     BUG FIX
         Several improvements to the Fedora package from Michael Schwendt via
         <https://bugzilla.redhat.com/show_bug.cgi?id=877096>, and from the
         harsh master that is [4mrpmlint[24m.  (I am stymied at teaching it that
         "outliers" is spelled correctly.  Maybe I should send it Schneier's
         book.  And an unresolvable invalid-spec-name lurks in the SRPM.)

   [1m2.42, 2013-07-31 A bug fix and packaging release.[0m
     ENHANCEMENT
         Documentation to dbjoin improved to better memory usage.  (Based on
         problem report by Lin Quan.)

     BUG FIX
         The [4m.spec[24m is now [4mperl-Fsdb.spec[24m to satisfy [4mrpmlint[24m.  Thanks to
         Christopher Meng for a specific bug report.

     BUG FIX
         Test [4mdbroweval_last.cmd[24m no longer has a column that caused failures
         because of numerical instability.

     BUG FIX
         Some tests now better handle bugs in old versions of perl (5.10, 5.12).
         Thanks to Calvin Ardi for help debugging this on a Mac with perl-5.12,
         but the fix should affect other platforms.

   [1m2.43, 2013-08-27 Adds in-file compression.[0m
     BUG FIX
         Changed the sort on [4mTEST/dbsort_merge.cmd[24m to strings (from numerics) so
         we're less susceptible to false test-failures due to floating point IO
         differences.

     EXPERIMENTAL ENHANCEMENT
         Yet more parallelism in dbmerge: new "endgame-mode" builds a merge tree
         of processes at the end of large merge tasks to get maximally
         parallelism.  Currently this feature is off by default because it can
         hang for some inputs.  Enable this experimental feature with
         "--endgame".

     ENHANCEMENT
         "Fsdb::IO" now handles being given "IO::Pipe" objects (as exercised by
         dbmerge).

     BUG FIX
         Handling of NamedTmpfiles now supports concurrency.  This fix will
         hopefully fix occasional "Use of uninitialized value $_ in string ne at
         ...NamedTmpfile.pm line 93." errors.

     BUG FIX
         Fsdb now requires perl 5.10.  This is a bug fix because some test cases
         used to require it, but this fact was not properly documented.
         (Back-porting to 5.008 would require removing all "//" operators.)

     ENHANCEMENT
         Fsdb now handles automatic compression of file contents.  Enable
         compression with "dbfilealter -Z xz" (or "gz" or "bz2").  All programs
         should operate on compressed files and leave the output with the same
         level of compression.  "xz" is recommended as fastest and most
         efficient.  "gz" is produces unrepeatable output (and so has no output
         test), it seems to insist on adding a timestamp.

   [1m2.44, 2013-10-02 A major change--all threads are gone.[0m
     ENHANCEMENT
         Fsdb is now thread free and only uses processes for parallelism.  This
         change is a big change--the entire motivation for Fsdb-2 was to exploit
         parallelism via threading.  Parallelism--good, but perl threading--bad
         for performance.  Horribly bad for performance.  About 20x worse than
         pipes on my box.  (See perl bug #119445 for the discussion.)

     NEW
         "Fsdb::Support::Freds" provides a thread-like abstraction over forking,
         with some nice support for callbacks in the parent upon child
         termination.

     ENHANCEMENT
         Details about removing threads: "dbpipeline" is thread free, and new
         tests to verify each of its parts.  The easy cases are
         "dbcolpercentile", "dbcolstats", "dbfilepivot", "dbjoin", and
         "dbcolstatscores", each of which use it in simple ways (2013-09-09).
         "dbmerge" is now thread free (2013-09-13), but was a significant
         rewrite, which brought "dbsort" along.  "dbmapreduce" is partly thread
         free (2013-09-21), again as a rewrite, and it brings "dbmultistats"
         along.  Full "dbmapreduce" support took much longer (2013-10-02).

     BUG FIX
         When running with user-only output ("-n"), dbroweval now resets the
         output vector $ofref after it has been output.

     NEW
         dbcolcreate will create all columns at the head of each row with the
         "--first" option.

     NEW
         dbfilecat will concatenate two files, verifying that they have the same
         schema.

     ENHANCEMENT
         dbmapreduce now passes comments through, rather than eating them as
         before.

         Also, dbmapreduce now supports a "--" option to prevent misinterpreting
         sub-program parameters as for dbmapreduce.

     INCOMPATIBLE CHANGE
         dbmapreduce no longer figures out if it needs to add the key to the
         output.  For multi-key-aware reducers, it never does (and cannot).  For
         non-multi-key-aware reducers, it defaults to add the key and will now
         fail if the reducer adds the key (with error "dbcolcreate: attempt to
         create pre-existing column...").  In such cases, one must disable
         adding the key with the new option "--no-prepend-key".

     INCOMPATIBLE CHANGE
         dbmapreduce no longer copies the input field separator by default.  For
         multi-key-aware reducers, it never does (and cannot).  For
         non-multi-key-aware reducers, it defaults to [4mnot[24m copying the field
         separator, but it will copy it (the old default) with the "--copy-fs"
         option

   [1m2.45, 2013-10-07 cleanup from de-thread-ification[0m
     BUG FIX
         Corrected a fast busy-wait in dbmerge.

     ENHANCEMENT
         Endgame mode enabled in dbmerge; it (and also large cases of dbsort)
         should now exploit greater parallelism.

     BUG FIX
         Test case with "Fsdb::BoundedQueue" (gone since 2.44) now removed.

   [1m2.46, 2013-10-08 continuing cleanup of our no-threads version[0m
     BUG FIX
         Fixed some packaging details.  (Really, threads are no longer required,
         missing tests in the MANIFEST.)

     IMPROVEMENT
         dbsort now better communicates with the merge process to avoid bursty
         parallelism.

         Fsdb::IO::Writer now can take "-autoflush =&lt; 1" for line-buffered
         IO.

   [1m2.47, 2013-10-12 test suite cleanup for non-threaded perls[0m
     BUG FIX
         Removed some stray "use threads" in some test cases.  We didn't need
         them, and these were breaking non-threaded perls.

     BUG FIX
         Better handling of Fred cleanup; should fix intermittent dbmapreduce
         failures on BSD.

     ENHANCEMENT
         Improved test framework to show output when tests fail.  (This time,
         for real.)

   [1m2.48, 2014-01-03 small bugfixes and improved release engineering[0m
     ENHANCEMENT
         Test suites now skip tests for libraries that are missing.  (Patch for
         missing "IO::Compresss:Xz" contributed by Calvin Ardi.)

     ENHANCEMENT
         Removed references to Jdb in the package specification.  Since the name
         was changed in 2008, there's no longer a huge need for backwards
         compatibility.  (Suggestion from Petr A abata.)

     ENHANCEMENT
         Test suites now invoke the perl using the path from $Config{perlpath}.
         Hopefully this helps testing in environments where there are multiple
         installed perls and the default perl is not the same as the
         perl-under-test (as happens in cpantesters.org).

     BUG FIX
         Added specific encoding to this manpage to account for Unicode.
         Required to build correctly against perl-5.18.

   [1m2.49, 2014-01-04 bugfix to unicode handling in Fsdb IO (plus minor packaging[0m
     [1mfixes)[0m
     BUG FIX
         Restored a line in the [4m.spec[24m to chmod g-s.

     BUG FIX
         Unicode decoding is now handled correctly for programs that read from
         standard input.  (Also: New test scripts cover unicode input and
         output.)

     BUG FIX
         Fix to Fsdb documentation encoding line.  Addresses test failure in
         perl-5.16 and earlier.  (Who knew "encoding" had to be followed by a
         blank line.)

   [1m2.50, 2014-05-27 a quick release for spec tweaks[0m
     ENHANCEMENT
         In dbroweval, the "-N" (no output, even comments) option now implies
         "-n", and it now suppresses the header and trailer.

     BUG FIX
         A few more tweaks to the [4mperl-Fsdb.spec[24m from Petr A abata.

     BUG FIX
         Fixed 3 uses of "use v5.10" in test suites that were causing test
         failures (due to warnings, not real failures) on some platforms.

   [1m2.51, 2014-09-05 Feature enhancements to dbcolmovingstats, dbcolcreate,[0m
     [1mdbmapreduce, and new sqlselect_to_db[0m
     ENHANCEMENT
         dbcolcreate now has a "--no-recreate-fatal" that causes it to ignore
         creation of existing columns (instead of failing).

     ENHANCEMENT
         dbmapreduce once again is robust to reducers that output the key;
         "--no-prepend-key" is no longer mandatory.

     ENHANCEMENT
         dbcolsplittorows can now enumerate the output rows with "-E".

     BUG FIX
         dbcolmovingstats is more mathematically robust.  Previously for some
         inputs and some platforms, floating point rounding could sometimes
         cause squareroots of negative numbers.

     NEW
         sqlselect_to_db converts the output of the MySQL or MarinaDB select
         comment into fsdb format.

     INCOMPATIBLE CHANGE
         dbfilediff now outputs the [4msecond[24m row when doing sloppy numeric
         comparisons, to better support test suites.

   [1m2.52, 2014-11-03 Fixing the test suite for line number changes.[0m
     ENHANCEMENT
         Test suites changes to be robust to exact line numbers of failures,
         since different Perl releases fail on different lines.
         <https://bugzilla.redhat.com/show_bug.cgi?id=1158380>

   [1m2.53, 2014-11-26 bug fixes and stability improvements to dbmapreduce[0m
     ENHANCEMENT
         The dbfilediff how supports a "--quiet" option.

     ENHANCEMENT
         Better documention of dbpipeline_filter.

     BUGFIX
         Added groff-base and perl-podlators to the Fedora package spec.  Fixes
         <https://bugzilla.redhat.com/show_bug.cgi?id=1163149>.  (Also in
         package 2.52-2.)

     BUGFIX
         An important stability improvement to dbmapreduce.  It, plus
         dbmultistats, and dbcolstats now support controlled parallelism with
         the "--pararallelism=N" option.  They default to run with the number of
         available CPUs.  dbmapreduce also moderates its level of parallelism.
         Previously it would create reducers as needed, causing CPU thrashing if
         reducers ran much slower than data production.

     BUGFIX
         The combination of dbmapreduce with dbrowenumerate now works as it
         should.  (The obscure bug was an interaction with dbcolcreate with
         non-multi-key reducers that output their own key.  dbmapreduce has too
         many useful corner cases.)

   [1m2.54, 2014-11-28 fix for the test suite to correct failing tests on[0m
     [1mnot-my-platform[0m
     BUGFIX
         Sigh, the test suite now has a test suite.  Because, yes, I broke it,
         causing many incorrect failures at cpantesters.  Now fixed.

   [1m2.55, 2015-01-05 many spelling fixes and dbcolmovingstats tests are more[0m
     [1mrobust to different numeric precision[0m
     ENHANCEMENT
         dbfilediff now can be extra quiet, as I continue to try to track down a
         numeric difference on FreeBSD AMD boxes.

     ENHANCEMENT
         dbcolmovingstats gave different test output (just reflecting rounding
         error) when stddev approaches zero.  We now detect hand handle this
         case.  See <https://rt.cpan.org/Public/Bug/Display.html?id=101220> and
         thanks to H. Merijn Brand for the bug report.

     BUG FIX
         Many, many spelling bugs found by H. Merijn Brand; thanks for the bug
         report.

     INCOMPATBLE CHANGE
         A number of programs had misspelled "separator" in "--fieldseparator"
         and "--columnseparator" options as "seperator".  These are now
         correctly spelled.

   [1m2.56, 2015-02-03 fix against Getopt::Long-2.43's stricter error checking[0m
     BUG FIX
         Internal argument parsing uses Getopt::Long, but mixed pass-through and
         <>.  Bug reported by Petr Pisar at
         <https://bugzilla.redhat.com/show_bug.cgi?id=1188538>.a

     BUG FIX
         Added missing BuildRequires for "XML::Simple".

   [1m2.57, 2015-04-29 Minor changes, with better performance from dbmulitstats.[0m
     BUG FIX
         dbfilecat now honors "--remove-inputs" (previously it didn't).  This
         omission meant that dbmapreduce (and dbmultistats) would accumulate
         files in [4m/tmp[24m when running.  Bad news for inputs with 4M keys.

     ENHANCMENT
         dbmultistats should be faster with lots of small keys.  dbcolstats now
         supports "-k" to get some of the functionality of dbmultistats (if data
         is pre-sorted and median/quartiles are not required).

         dbfilecat now honors "--remove-inputs" (previously it didn't).  This
         omission meant that dbmapreduce (and dbmultistats) would accumulate
         files in [4m/tmp[24m when running.  Bad news for inputs with 4M keys.

   [1m2.58, 2015-04-30 Bugfix in dbmerge[0m
     BUG FIX
         Fixed a case where dbmerge suffered mojobake in endgame mode.  This bug
         surfaced when dbsort was applied to large files (big enough to require
         merging) with unicode in them; the symptom was something like:
           Wide character in print at /usr/lib64/perl5/IO/Handle.pm line 420,
         <GEN12> line 111.

   [1m2.59, 2016-09-01 Collect a few small bug fixes and documentation[0m
     [1mimprovements.[0m
     BUG FIX
         More IO is explicitly marked UTF-8 to avoid Perl's tendency to mojibake
         on otherwise valid unicode input.  This change helps html_table_to_db.

     ENHANCEMENT
         dbcolscorrelate now crossreferences dbcolsregression.

     ENHANCEMENT
         Documentation for dbrowdiff now clarifies that the default is baseline
         mode.

     BUG FIX
         dbjoin now propagates "-T" into the sorting process (if it is
         required).  Thanks to Lan Wei for reporting this bug.

   [1m2.60, 2016-09-04 Adds support for hash joins.[0m
     ENHANCEMENT
         dbjoin now supports hash joins with "-t lefthash" and "-t righthash".
         Hash joins cache a table in memory, but do not require that the other
         table be sorted.  They are ideal when joining a large table against a
         small one.

   [1m2.61, 2016-09-05 Support left and right outer joins.[0m
     ENHANCEMENT
         dbjoin now handles left and right outer joins with "-t left" and "-t
         right".

     ENHANCEMENT
         dbjoin hash joins are now selected with "-m lefthash" and "-m
         righthash" (not the shortlived "-t righthash" option).  (Technically
         this change is incompatible with Fsdb-2.60, but no one but me ever used
         that version.)

   [1m2.62, 2016-11-29 A new yaml_to_db and other minor improvements.[0m
     ENHANCEMENT
         Documentation for xml_to_db now includes sample output.

     NEW
         yaml_to_db converts a specific form of YAML to fsdb.

     BUG FIX
         The test suite now uses "diff -c -b" rather than "diff -cb" to make
         OpenBSD-5.9 happier, I hope.

     ENHANCEMENT
         Comments that log operations at the end of each file now do simple
         quoting of spaces.  (It is not guaranteed to be fully shell-compliant.)

     ENHANCEMENT
         There is a new standard option, "--header", allowing one to specify an
         Fsdb header for inputs that lack it.  Currently it is supported by
         dbcoldefine, dbrowuniq, dbmapreduce, dbmultistats, dbsort, dbpipeline.

     ENHANCEMENT
         dbfilepivot now allows the [1m--possible-pivots [22moption, and if it is
         provided processes the data in one pass.

     ENHANCEMENT
         dbroweval logs are now quoted.

   [1m2.63, 2017-02-03 Re-add some features supposedly in 2.62 but not, and add[0m
     [1mmore --header options.[0m
     ENHANCEMENT
         The option [1m-j [22mis now a synonym for [1m--parallelism[22m.  (And several
         documention bugs about this option are fixed.)

     ENHANCEMENT
         Additional support for "--header" in dbcolmerge, dbcol, dbrow, and
         dbroweval.

     BUG FIX
         Version 2.62 was supposed to have this improvement, but did not (and
         now does): dbfilepivot now allows the [1m--possible-pivots [22moption, and if
         it is provided processes the data in one pass.

     BUG FIX
         Version 2.62 was supposed to have this improvement, but did not (and
         now does): dbroweval logs are now quoted.

   [1m2.64, 2017-11-20 several small bugfixes and enhancements[0m
     BUG FIX
         In dbroweval, the "next row" option previously did not correctly set up
         "_last_fieldname".  It now does.

     ENHANCEMENT
         The csv_to_db converter now has an optional "-F x" option to set the
         field separator.

     ENHANCEMENT
         Finally dbcolsplittocols has a "--header" option, and a new "-N" option
         to give the list of resulting output columns.

     INCOMPATIBLE CHANGE
         Now dbcolstats and dbmultistats produce no output (but a schema) when
         given no input but a schema.  Previously they gave a null row of
         output.  The "--output-on-no-input" and "--no-output-on-no-input"
         options can control this behavior.

   [1m2.65, 2018-02-16 Minor release, bug fix and -F option.[0m
     ENHANCEMENT
         dbmultistats and dbmapreduce now both take a "-F x" option to set the
         field separator.

     BUG FIX
         Fixed missing "use Carp" in dbcolstats.  Also went back and cleaned up
         all uses of croak().  Thanks to Zefram for the bug report.

   [1m2.66, 2018-12-20 Critical bug fix in dbjoin.[0m
     BUG FIX
         Removed old tests from MANIFEST.  (Thanks to Hang Guo for reporting
         this bug.)

     IMPROVEMENT
         Errors for non-existing input files now include the bad filename
         (before: "cannot setup filehandle", now: "cannot open input: cannot
         open TEST/bad_filename").

     BUG FIX
         Hash joins with three identical rows were failing with the assertion
         failure "internal error: confused about overflow" due to a now-fixed
         bug.

   [1m2.67, 2019-07-10 add support for reading and writing hdfs[0m
     IMPROVEMENT
         dbformmail now has an "mh" mechanism that writes messages to individual
         files (an mh-style mailbox).

     BUG FIX
         dbrow failed to include the Carp library, leading to fails on croak.

     BUG FIX
         Fixed dbjoin error message for an unsorted right stream was incorrect
         (it said left).

     IMPROVEMENT
         All Fsdb programs can now read from and write to HDFS, when files that
         start with "hdfs:" are given to -i and -o options.

   [1m2.68, 2019-09-19 All programs now support automatic decompression based on[0m
     [1mfile extension.[0m
     IMPROVEMENT
         The omitted-possible-error test case for dbfilepivot now has an
         altnerative output that I saw on some BSD-running systems (thanks to
         CPAN).

     IMPROVEMENT
         dbmerge and dbmerge2 now support "--header".  dbmerge2 now gives better
         error messages when presented the wrong number of inputs.

     BUG FIX
         dbsort now works with "--header" even when the file is big (due to
         fixes to dbmerge).

     IMPROVEMENT
         cvs_to_db now processes data with the "binary" option, allowing it to
         handle newlines embedded in quoted fields.

     IMPROVEMENT
         All programs now will transparently decompress input files, if they are
         listed as a filename as an input argument that extends with a standard
         extension (.gz, .bz2, and .xz).

   [1m2.69, 2019-11-22 a small bugfix in dbcolstats[0m
     BUG FIX
         Filled in the the test case for autodecompress, which was missing for
         the 2.68 release.

     ENHANCEMENT
         The groff program is required for build, and the "Makefile.PL" fails if
         groff is missing at build time.  Thanks to Chris Williams for
         suggesting this check, and the CPAN auto-building system for trying
         many platforms.

     BUG FIX
         The dbcolstats program had numerical instability that sometimes results
         in failing with a square-root of a negative number when many values
         varied right at the edge of floating-point precision.  We now detect
         and report that case as 0 stddev.  Thanks to Hang Guo for providing a
         test case.

   [1m2.70, 2020-11-12 Some small quality-of-life enhancements and corner-case[0m
     [1mbugfixes.[0m
     ENHANCEMENT
         dbcol can now take an option "-a" to include all columns, allowing
         reordering of certain columns while passing the rest through.

     ENHANCEMENT
         dbrowuniq and dbmerge now buffer comments in a way that the last row of
         data output is no longer in the last block of comments.  (The data is
         identical, but for humans looking at output, this change makes it less
         likely to lose the last row.)

     BUG FIX
         dbmultistats and dbpipeline documentation now indicates that they
         support "--header" (something they did since version 2.62 in
         2016-11-29, but now documented.

     ENHANCEMENT
         dbcolcreate now supports "--header".

     BUG FIX
         Fixed several spelling errors in deprecated programs and removed
         information about the no-longer existing FreeBSD and MacOS ports.
         Thanks to Calvin Ardi for the patch.

     BUG FIX
         dbmerge now handles --xargs when only one file is provided (and passes
         the file through unchanged).  It also throws a clean error with --xargs
         if zero files are provided.  (To support dbmerge, dbcol now has an
         internal "--saveoutput" option.)  Thanks to Yuri Pradkin for reporting
         the unhandled corner-case.

   [1m2.71, 2020-11-16 Fix a race condition breaking test suites.[0m
     BUG FIX
         Suppress a race condition in dbcolmerge was sometimes throwing the
         error "Fsdb::Support::Freds: ending, but running process:
         dbmerge:xargs" in the dbmerge_0_xargs test case, on exit.

   [1m2.72, 2020-12-01 A small bug and a packaging improvement.[0m
     BUG FIX
         dbcolhisto now handles the degenerate case where everything has the
         same value (previously it would throw "illegal division by zero").

     ENHANCEMENT
         The spec for Fedora now includes "make" as BuildRequires, something
         required for Fedora 34.

   [1m2.73, 2021-05-18 Updates dbcolpercentile with "--weighted", and with more[0m
     [1mipv6.[0m
     ENHANCEMENT
         dbcolpercentile now has a "--weighted" option.

     ENHANCEMENT
         The new Fsdb::Support::IPv6 package includes ipv6_normalize,
         ipv6_zeroize to rewrite ipv6 print addresses in IPv6 normal form, with
         a 0 in each 4-nybble field.

   [1m2.74, 2021-06-23 More ipv6.[0m
     ENHANCEMENT
         Fsdb::Support::IPv6 package includes ipv6_fullhex to rewrite ipv6 print
         addresses as full, 128-bit hex values.

   [1m2.75, 2022-04-02 New type specifications in the schema to better support type[0m
     [1mconversions in python.[0m
     ENHANCEMENT
         Add optional type specifications to the schema.  Types are not used in
         Perl, but are relevant in Python and Go Fsdb bindings.  Types use a
         subset of perl pack specifiers: c, s, l, q are signed 8, 16, 32, and
         64-bit integers, f is a float, d is double float, a is utf-8 string,
         and &gt; and &lt; can force big or little endianness.  The default type
         for everything is "a", that is, utf-8 strings.  Thanks to Wes Hardaker
         for pushing to get this long-desired feature out the door; his Python
         bindings need types.

     ENHANCEMENT
         dbcol, dbcolcreate, dbcolcopylast, and dbcolrename now understand and
         propagate schema types.  dbsort, dbjoin, dbmerge, dbmerge2 and
         dbfilepivot all take a new option "-t" to sort by type-inferred
         comparision, if a type is given.

     ENHANCEMENT
         dbcolstat, dbmultistats, and dbcolmovingstats now include type
         information in their output schema.  (They assumes input variables are
         floats, not integers.)

     ENHANCEMENT
         Even more IPv6: the functions in Fsdb::Support::IPv6 package now
         support strings of hex digits as an alternate encoding for IP address
         (and they are already the output of ipv6_fullhex), and
         "ip_fullhex_to_normal" converts full hex-encoded IPv4 or IPv6 addresses
         to their "normal" form (dotted-quad or IPv6 printable format).

   [1m3.0, 2022-04-04 Complete type support and accordingly bump major version.[0m
     NEW
         The major version number is now 3.0 to correspond to the addition of
         types (although they were actually added in 2.75).  Old fsdb files are
         supported (Fsdb-3.0 is backwards compatible with databases), but older
         versions will confuse types in new files (new Fsdb files are not
         forward compatible with old versions).

     ENHANCEMENT
         Type specifications in a few more programs: dbcolhisto,
         dbcolscorrelate, dbcolsregression, dbcolstatscores, dbrowaccumulate,
         dbrowcount, dbrowdiff, dbrvstatdiff.

     ENHANCEMENT
         dbcolhisto now puts an empty value on any empty rows.

     NEW
         dbcoltype redefines column types, or clears them with the "-v" option.

   [1m3.1, 2022-11-22 A post-3.0 cleanup release with minor fixes.[0m
     ENHANCEMENT
         Type specifications in a few more programs that I missed: dbrowuniq,
         dbcolpercentile.

     ENHANCEMENT
         Minor documentation improvements.

   [1m3.2, 2023-10-11 Add new module dbcolsdecimate[0m
     NEW
         dbcolsdecimate reduces density in timeseries data to make graphs with
         overly dense points visually similar but smaller.

     ENHANCEMENT
         yaml_to_db now flattens one level of arrays into comma-separated lists.

     ENHANCEMENT
         Clearer installation instructions.

   [1m3.3, 2023-10-13 Quickly making dbcolsdecimate more flexible.[0m
     INCOMPATBILE ENHANCEMENT
         dbcolsdecimate now takes either relative ([1m-p[22m) or absolute ([1m-P[22m)
         precision, and precision now affects only subsequent columns.  Also, if
         absolute precisions are given for all columns, data is not buffered.

   [1m3.4, 2024-01-05 Correct propagation of TMPDIR options into subprograms.[0m
     ENHANCEMENT
         dbcolsdecimate now has examples in its documentation.

     BUG FIX
         dbcolsstats, dbmapreduce, dbcolpercentile, dbfilepivot, and
         dbmultistats now correctly propagate the temporary directory into the
         sort route, if required.  All of these programs sometimes require sort
         internally, and previously may have failed to use the correct tmpdir
         when it was set on the command line as an option.  Thanks to Erica
         Stutz for noticing this bug.

   [1m3.5, 2024-06-07 Weighted stats added to dbcolstats and dbcolscorrelate[0m
     ENHANCEMENT
         dbrowdiff now has a "--future" option that compares incrementally with
         the next row rather than the previous.

     ENHANCEMENT
         dbcolstats and dbcolscorrelate now optionally do weighted stats with
         the "--weight" option.

   [1m3.6, 2024-06-08 Bugfix in weighted stats for dbcolscorrelate; column options[0m
     [1min dbrowdiff.[0m
     BUG FIX
         dbcolscorrelate was not properly applying weighting.

     ENHANCEMENT
         dbrowdiff now has "-A" and "-P" options to set output column names.

   [1m3.7, 2024-06-19 Adding weighted stats for dbmultistats.[0m
     ENHANCEMENT
         dbmultistats now supports weighted stats.

   [1m3.8, 2024-07-01 Bugfix in dbmerge.[0m
     BUG FIX
         dbmerge now correctly handles the case when invoked with "--xargs" with
         exactly two input files.  Thanks to Erica Stutz for reporting this
         error.

   [1m3.9, 2024-08-02 Fixing dbjoin type propagation.[0m
     ENHANCEMENT
         dbrowsdiff now has a -e option to specify the value to use in
         future-mode for the last row.

     BUG FIX
         dbjoin now propagates types, rather than eating them.

   [1m3.10, 2025-03-01 Bug fix for some IPv6 addresses, snzip support, and auto[0m
     [1mcompression.[0m
     ENHANCEMENT
         The license in the .spec is now SPDX-compliant.  Thanks to Miroslav
         Such~A1/2 for the patch.

     BUG FIX
         Fsdb::Support::IPv6 now handles IPv4-embedded-in-IPv6 print-form
         addresses.

     ENHANCEMENT
         We now automatically decompress ".sz" files (assuming "snzip") is
         installed).

     ENHANCEMENT
         We now automatically compress files written with a standard compression
         extension.

   [1m3.11, 2025-03-06 internal performance enhancements[0m
     ENHANCEMENT
         Internet temporary files for sort, if required, are now compressed,
         using snzip if possible, or gzip otherwise.  (In many cases CPU is
         cheaper than I/O, particularly when the data compresses well.)

     ENHANCEMENT
         When dbsort overflows to disk, it now removes temporary files as it
         goes, when possible.  It now requires less secondary storage for very
         large inputs.  (Secondary storage during merge was actually O(n log n)
         before, and now is back down to O(n), where n is input size.)

   [1m3.12, 2025-03-06 internal performance enhancements[0m
     ENHANCEMENT
         dbmerge intermediate files are now compressed, reducing intermediate
         storage.

   [1m3.13, 2025-04-16 bugfix for intermediate compression[0m
     BUG FIX
         Temporary file autocompression was locked to gzip due to a bug.  We
         correctly now use snzip when it is available.

     BUG FIX
         Improved test suite handling of compressed files, and improved
         packaging for RHEL-8.

   [1m3.14, 2025-06-09 New filter dbrowuniqcount computes counts of all unique[0m
     [1mrows.[0m
     ENHANCEMENT
         New filter dbrowuniqcount computes counts of all unique rows,
         regardless of adjacency, in O(u) memory where u is the number of unique
         rows.

   [1m3.15, 2025-12-08 Bug fixes and cleanup for a messy 3.14 release.[0m
     ENHANCEMENT
         tcpdump_to_db now handles some forms of modern tcpdump output for TCP.

     ENHANCEMENT
         dbformmail now opens the template in unicode, and handles
         comma-separated CCs in Mail mode.

[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) 1991-2024 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 v5.44.0                       2026-07-22                            [4mFsdb[24m(3)
