Fsdb - a flat-text database for shell scripting


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 .fsdb), 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 parlallel when possible.

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

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

    can be written in perl as:

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

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

Fsdb is built on flat-ASCII databases. By storing data in simple text files and processing it with pipelines it is easy to experiment (in the shell) and look at the output. To the best of my knowledge, the original implementation of this idea was /rdb, a commercial product described in the book UNIX relational database management: application development in the UNIX environment by Rod Manis, Evan Schaffer, and Robert Jorgensen (1988 by Prentice Hall, and also at the web page 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


3.4, 2024-01-05 Correct propagation of TMPDIR options into subprograms.


dbcolsdecimate now has examples in its documentation.


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.


executive summary
what's new
basic data format
basic data manipulation
list of commands
another example
a gradebook example
a password example
release notes


Fsdb now uses the standard Perl build and installation from ExtUtil::MakeMaker(3), so the quick answer to installation is to type:

    perl Makefile.PL
    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 bin, etc. (See ExtUtil::MakeMaker(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.


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 dbfilealter(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 dbfilealter(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


        #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 dbfilealter(1) for more details.


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
        # 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.


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


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 dbroweval(1) for more details about writing code.


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.



add columns to a database


set the column headings for a non-Fsdb file



select columns from a table


select rows from a table


sort rows based on a set of columns


compute the natural join of two tables


rename a column


merge two columns into one


split one column into two or more columns


split one column into multiple rows


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


check that db file doesn't have some common errors



compute statistics over a column (mean,etc.,optionally median)


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


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


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


computing moving statistics over a column of data


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


compute the rank or percentile of a column


compute histograms over a column of data


compute the coefficient of correlation over several columns


drop rows selectively, keeping large changes and periodic samples


compute linear regression and correlation for two columns


compute a running sum over a column of data


count the number of rows (a subset of dbstats)


compute differences between a columns in each row of a table


number each row


run arbitrary Perl code on each row


count/eliminate identical rows (like Unix uniq(1))


compare fields on rows of a file (something like Unix diff(1))



pretty-print columns


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


remove comments from a table


generate a script that sends form mail based on each row


(These programs convert data into fsdb. See their web pages for details.)




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




the output of SQL SELECT tables to db


spreadsheet tab-delimited files to db


(see man tcpdump(8) on any reasonable system)


XML input to fsdb, assuming they're very regular

(And out of fsdb:)


Comma-separated-value format from fsdb.


simple conversion of Fsdb to html tables


Many programs have common options:

-? or --help

Show basic usage.

-N on --new-name

When a command creates a new column like dbrowaccumulate's accum, this option lets one override the default name of that new column.

-T TmpDir

where to put tmp files. Also uses environment variable TMPDIR, if -T is not specified. Default is /tmp.

Show basic usage.

-c FRACTION or --confidence FRACTION

Specify confidence interval FRACTION (dbcolstats, dbmultistats, etc.)

-C S or --element-separator S

Specify column separator S (dbcolsplittocols, dbcolmerge).

-d or --debug

Enable debugging (may be repeated for greater effect in some cases).

-a or --include-non-numeric

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)

-S or --pre-sorted

Assume the data is pre-sorted. May be repeated to disable verification (saving a small amount of work).

-e E or --empty E

give value E as the value for empty (null) records

-i I or --input I

Input data from file I.

-o O or --output O

Write data out to file O.

--header H

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.


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 dbrow(1) or dbroweval(1) for examples.)

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


A number of programs do sorting, or depend on defining an ordering of rows. Such programs use these standard sorting options:

-r or --descending

sort in reverse order (high to low)

-R or --ascending

sort in normal order (low to high)

-t or --type-inferred-sorting

sort fields by type (numeric or leicographic), automatically

-n or --numeric

sort numerically

-N or --lexical

sort lexicographically


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.)


        #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 | \

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


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 1 80
        b 2 70
        c 3 65
        d 4 90
        e 5 70
        f 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  1  80
        butler lampson  2  70
        david clark  3  65
        constantine drovolis  4  90
        debrorah estrin  5  70
        sally floyd  6  90

To compute statistics on an exam, do

        cat DATA/grades | dbstats test1 |dblistize


        #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


        #fsdb low histogram
        65      *
        70      **
        80      *
        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 test1.txt):

        To: _email (_name)
        From: J. Random Professor <>
        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 >

And run it:

        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 = "-"; };' | 

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).


To convert the Unix password file to db:

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

To convert the group file

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

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


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 | \

(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
                dbcolcreate source $i < file$i.fsdb
        done >output.fsdb

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


As with any tool, one should (which means must) understand the limits of the tool.

All Fsdb tools should run in constant memory. In some cases (such as dbcolstats 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 dbcolstats_extrema 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.


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

Fsdb 2.0 Rationale

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 also 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).

Fsdb And Muliple Processors

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.)

Fsdb 3.0 Rationale

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.


Fsdb includes code ported from Geoff Kuenning (Fsdb::Support::TDistribution).

Fsdb contributors: Ashvin Goel, Geoff Kuenning, Vikram Visweswariah, Kannan Varadahan, Lars Eggert, Arkadi Gelfond, David Graff, Haobo Yu, Pavlin Radoslavov, Graham Phillips, Yuri Pradkin, Alefiya Hussain, Ya Xu, Michael Schwendt, Fabio Silva, Jerry Zhao, Ning Xu, Martin Lukac, Xue Cai, Michael McQuaid, Christopher Meng, Calvin Ardi, H. Merijn Brand, Lan Wei, Hang Guo, Wes Hardaker.

Fsdb includes datasets contributed from NIST (DATA/nist_zarr13.fsdb), from, the NIST/SEMATECH e-Handbook of Statistical Methods, section Background and Data. The source is public domain, and reproduced with permission.


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 UNIX relational database management: application development in the UNIX environment by Rod Manis, Evan Schaffer, and Robert Jorgensen (and also at the web page

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 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 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).


Versions prior to 1.0 were released informally on my web page but were not announced.

0.0 1991

started for my own research use

0.1 26-May-94

first check-in to RCS

0.2 15-Mar-95

parts now require perl5

1.0, 22-Jul-97

adds autoconf support and a test script.

1.1, 20-Jan-98

support for double space field separators, better tests

1.2, 11-Feb-98

minor changes and release on comp.lang.perl.announce

1.3, 17-Mar-98

  • 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

1.4, 27-Mar-98

  • improves error messages (all should now report the program that makes the error)

  • fixed a bug in dbstats output when the mean is zero

1.5, 25-Jun-98

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 has been replaced by
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)

1.6, 24-May-99

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

1.7, 5-Jan-00

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 and

To subscribe to either, send mail to or 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.

1.8, 28-Jun-00

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.

1.9, 6-Nov-00

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

1.10, 10-Apr-01

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

1.11, 2-Nov-01

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

1.12, 30-Oct-02

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/ 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 dblib_text2html() that will do simple conversion of iso-8859-1 to HTML

1.13, 4-Feb-04

NEW fsdb added to the freebsd ports tree Maintainer:
BUG FIX properly handle trailing spaces when data must be numeric (ex. dbstats with -FS, see test dbstats_trailing_spaces). Fix from Ning Xu
NEW dbcolize error message improved (bug report from Terrence Brannon), and list format documented in the README.
NEW cgi_to_db converts 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

1.14, 24-Aug-06

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)

1.15, 12-Nov-07

NEW fsdb-1.14 added to the MacOS Fink system (Thanks to Lars Eggert for maintaining this port.)
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 is now moved to
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)

2.0, 25-Jan-08

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);
        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.

2.1, 6-Apr-08

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 automatically read either format. This capability was one of the milestone goals for 2.0, so yea!

2.2, 23-May-08

Release 2.2 is another 2.x alpha release. Now most of the commands are ported, but a few remain, and I plan one last incompatible change (to the file header) before 2.x final.


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.


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


combined_log_format_to_db to convert Apache logfiles


Options to dbrowdiff are now -B and -I, not -a and -i.


dbstripcomments is now dbfilestripcomments.


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


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).


dbmultistats with quantile option doesn't work currently.


dbcoldiff is renamed dbrvstatdiff.


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

2.3, 27-May-08 (alpha)

Another alpha release, this one just to fix the critical dbjoin bug listed below (that happens to have blocked my MP3 jukebox :-).


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


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.


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.

2.4, 18-Jun-08

Another alpha release, mostly to fix minor usability problems in dbmapreduce and client functions.


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


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


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.


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

2.5, 24-Jun-08

Another alpha release, fixing more minor bugs in dbmapreduce and lossage in Fsdb::IO.


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.


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

2.6, 11-Jul-08

Another alpha release, needed to fix DaGronk. One new port, small bug fixes, and important fix to dbmapreduce.


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.

Sigh, the sense of the --warnings option in dbrow was inverted. No longer.


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


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".


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.


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.


dbfilevalidate now has a -c option to correct errors.


html_table_to_db provides the inverse of db_to_html_table.

2.8, 5-Aug-08

Change header format, preserving forwards compatibility.


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


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.


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

2.9, 6-Aug-08

Still in beta; just a quick bug-fix for dbmapreduce.


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

2.10, 23-Sep-08

Still in beta, but picking up some bug fixes.


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


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


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


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

2.11, 14-Oct-08

Still in beta, but picking up some bug fixes.


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.


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


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

2.12, 16-Oct-08

Finally, a full (non-beta) 2.x release!


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.

2.13, 30-Oct-08


dbrowaccumulate now treats non-numeric data as zero by default.


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

2.14, 26-Nov-08


Improved documentation for dbmapreduce's -f option.


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

2.15, 13-Apr-09


Fix a make install bug reported by Shalindra Fernando.

2.16, 14-Apr-09


Another minor release bug: on some systems programize_module looses executable permissions. Again reported by Shalindra Fernando.

2.17, 25-Jun-09


Typo in the dbroweval manual fixed.


There is no longer a comment line to label columns in dbcolneaten, 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.


It turns out dbcolneaten was not correctly handling trailing spaces when given the -E option to suppress them. This regression is now fixed.


dbroweval(1) can now handle direct references to the last row via $lfref, a dubious but now documented feature.


Separators set with -C in dbcolmerge and dbcolsplittocols were not properly setting the heading, and null fields were not recognized. The first bug was reported by Martin Lukac.

2.18, 1-Jul-09 A minor release


Documentation for Fsdb::IO::Reader has been improved.


The package should now be PGP-signed.

2.19, 10-Jul-09


Internal improvements to debugging output and robustness of dbmapreduce and dbpipeline. TEST/dbpipeline_first_fails.cmd re-enabled.

2.20, 30-Nov-09 (A collection of minor bugfixes, plus a build against Fedora 12.)


Loging for dbmapreduce with code refs is now stable (it no longer includes a hex pointer to the code reference).


Better handling of mixed blank lines in Fsdb::IO::Reader (see test case dbcolize_blank_lines.cmd).


html_table_to_db now handles multi-line input better, and handles tables with COLSPAN.


dbpipeline 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 dbpipeline_first_fails).


dbmapreduce now detects and correctly fails when the input and reducer have incompatible field separators.


dbcolstats, dbcolhisto, dbcolscorrelate, dbcolsregression, and dbrowcount now all take an -F option to let one specify the output field separator (so they work better with dbmapreduce).


An omitted -k from the manual page of dbmultistats is now there. Bug reported by Unkyu Park.

2.21, 17-Apr-10 bug fix release


Fsdb::IO::Writer now no longer fails with -outputheader => never (an obscure bug).


Fsdb (in the warnings section) and dbcolstats 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.


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


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

2.22, 2010-10-31 One new tool dbcolcopylast and several bug fixes for Perl 5.10.


dbmerge now correctly handles n-way merges. Bug reported by Yuri Pradkin.


dbcolneaten now defaults to not padding the last column.


dbrowenumerate now takes -N NewColumn to give the new column a name other than "count". Feature requested by Mike Rouch in January 2005.


New program dbcolcopylast 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.


Several tools (particularly dbmapreduce and dbmultistats) 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.

2.23, 2011-03-10 Several small portability bugfixes; improved dbcolstats for large datasets


Documentation to dbrvstatdiff was changed to use "sd" to refer to standard deviation, not "ss" (which might be confused with sum-of-squares).


This documentation about dbmultistats was missing the -k option in some cases.


dbmapreduce was failing on MacOS-10.6.3 for some tests with the error

    dbmapreduce: cannot run external dbmapreduce reduce program (perl TEST/

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.


The csv_to_db command requires an external Perl library (Text::CSV_XS). 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.


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


dbcolstats 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.

2.24, 2011-04-15 Improvements to fix an old bug in dbmapreduce with different field separators


The dbfilealter 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.


The dbmultistats command previously failed with an error message when invoked on input with a non-default field separator. The root cause was the underlying dbmapreduce 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.

2.25, 2011-08-07 Two new tools, xml_to_db and dbfilepivot, and a bugfix for two people.


kitrace_to_db now supports a --utc 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 :-)


xml_to_db can convert simple, very regular XML files into Fsdb.


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

2.26, 2011-12-12 Bug fixes, particularly for perl-5.14.2.


Bugs fixed in Fsdb::IO::Reader(3) manual page.


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

2.27, 2012-11-15 Accumulated bug fixes.


csv_to_db now reports errors in CVS input with real diagnostics.


dbcolmovingstats can now compute median, when given the -m option.


dbcolmovingstats non-numeric handling (the -a option) now works properly.


The internal t/test_command.t test framework is now documented.


dbrowuniq 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.

2.28, 2012-11-15 A quick release to fix most rpmlint errors.


Fixed a number of minor release problems (wrong permissions, old FSF address, etc.) found by rpmlint.

2.29, 2012-11-20 a quick release for CPAN testing


Tweaked the RPM spec.


Modified Makefile.PL to fail gracefully on Perl installations that lack threads. (Without this fix, I get massive failures in the non-ithreads test system.)

2.30, 2012-11-25 improvements to perl portability


Removed unicode character in documention of dbcolscorrelated so pod tests will pass. (Sigh, that should work :-( )


Fixed test suite failures on 5 tests (dbcolcreate_double_creation 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.


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

2.31, 2012-11-28 A release with actual improvements to dbfilepivot and dbrowuniq.


Documentation fixes: typos in dbcolscorrelated, bugs in dbfilepivot, clarification for comment handling in Fsdb::IO::Reader.


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.


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


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

2.32, 2012-12-21 Test suites should now be more numerically robust.


New dbfilediff does fsdb-aware file differencing. It does not do smart intuition of add/removes like Unix diff(1), but it does know about columns, and with -E, it does numeric-aware differences.


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 exactly what I use.

2.33, 2012-12-23 Minor fixes to some test cases.


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.)


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


Fixed a numerical stability problem in the dbroweval_last test case.


2.34, 2013-02-10 Parallelism in dbmerge.


Documention for dbjoin now includes resource requirements.


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


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.)

2.35, 2013-02-23 Improvements to dbmerge parallelism


Fsdb temporary files are now created more securely (with File::Temp).


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.


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.

2.36, 2013-02-25 dbsort pipelines with dbmerge

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.

2.37, 2013-02-26 quick bugfix to support parallel sort and merge from recent releases

BUG FIX Since 2.35, dbmerge delayed removal of input files given by --xargs. This problem is now fixed.

2.38, 2013-04-29 minor bug fixes


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


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.


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).


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

2.39, date 2013-05-31 quick release for the dbrowuniq extension


Actually in 2.38, the Fedora .spec got cleaner dependencies. Suggestion from Christopher Meng via


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


dbrowuniq now supports -I for incremental counting.

2.40, 2013-07-13 small bug fixes


dbsort now has more respect for a user-given temporary directory; it no longer is ignored for merging.


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


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

2.41, 2013-07-29 small bug and packaging fixes


Documentation to dbrvstatdiff improved (inspired by questions from Qian Kun).


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


Add missing XML::Simple dependency to Makefile.PL.


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


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


Several improvements to the Fedora package from Michael Schwendt via, and from the harsh master that is rpmlint. (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.)

2.42, 2013-07-31 A bug fix and packaging release.


Documentation to dbjoin improved to better memory usage. (Based on problem report by Lin Quan.)


The .spec is now perl-Fsdb.spec to satisfy rpmlint. Thanks to Christopher Meng for a specific bug report.


Test dbroweval_last.cmd no longer has a column that caused failures because of numerical instability.


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.

2.43, 2013-08-27 Adds in-file compression.


Changed the sort on TEST/dbsort_merge.cmd to strings (from numerics) so we're less susceptible to false test-failures due to floating point IO differences.


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.


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


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


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.)


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.

2.44, 2013-10-02 A major change--all threads are gone.


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.)


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


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).


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


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


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


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.


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.


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 not copying the field separator, but it will copy it (the old default) with the --copy-fs option

2.45, 2013-10-07 cleanup from de-thread-ification


Corrected a fast busy-wait in dbmerge.


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


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

2.46, 2013-10-08 continuing cleanup of our no-threads version


Fixed some packaging details. (Really, threads are no longer required, missing tests in the MANIFEST.)


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

Fsdb::IO::Writer now can take -autoflush = 1> for line-buffered IO.

2.47, 2013-10-12 test suite cleanup for non-threaded perls


Removed some stray "use threads" in some test cases. We didn't need them, and these were breaking non-threaded perls.


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


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

2.48, 2014-01-03 small bugfixes and improved release engineering


Test suites now skip tests for libraries that are missing. (Patch for missing IO::Compresss:Xz contributed by Calvin Ardi.)


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 form Petr Šabata.)


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


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

2.49, 2014-01-04 bugfix to unicode handling in Fsdb IO (plus minor packaging fixes)


Restored a line in the .spec to chmod g-s.


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


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.)


2.50, 2014-05-27 a quick release for spec tweaks


In dbroweval, the -N (no output, even comments) option now implies -n, and it now suppresses the header and trailer.


A few more tweaks to the perl-Fsdb.spec from Petr Šabata.


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

2.51, 2014-09-05 Feature enhancements to dbcolmovingstats, dbcolcreate, dbmapreduce, and new sqlselect_to_db


dbcolcreate now has a --no-recreate-fatal that causes it to ignore creation of existing columns (instead of failing).


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


dbcolsplittorows can now enumerate the output rows with -E.


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


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


dbfilediff now outputs the second row when doing sloppy numeric comparisons, to better support test suites.

2.52, 2014-11-03 Fixing the test suite for line number changes.


Test suites changes to be robust to exact line numbers of failures, since different Perl releases fail on different lines.

2.53, 2014-11-26 bug fixes and stability improvements to dbmapreduce


The dbfilediff how supports a --quiet option.


Better documention of dbpipeline_filter.


Added groff-base and perl-podlators to the Fedora package spec. Fixes (Also in package 2.52-2.)


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.


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.)

2.54, 2014-11-28 fix for the test suite to correct failing tests on not-my-platform


Sigh, the test suite now has a test suite. Because, yes, I broke it, causing many incorrect failures at cpantesters. Now fixed.

2.55, 2015-01-05 many spelling fixes and dbcolmovingstats tests are more robust to different numeric precision


dbfilediff now can be extra quiet, as I continue to try to track down a numeric difference on FreeBSD AMD boxes.


dbcolmovingstats gave different test output (just reflecting rounding error) when stddev approaches zero. We now detect hand handle this case. See <> and thanks to H. Merijn Brand for the bug report.


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


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

2.56, 2015-02-03 fix against Getopt::Long-2.43's stricter error checkign


Internal argument parsing uses Getopt::Long, but mixed pass-through and <>. Bug reported by Petr Pisar at


Added missing BuildRequires for XML::Simple.

2.57, 2015-04-29 Minor changes, with better performance from dbmulitstats.


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


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 /tmp when running. Bad news for inputs with 4M keys.

2.58, 2015-04-30 Bugfix in dbmerge


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 soemthing like: Wide character in print at /usr/lib64/perl5/IO/ line 420, <GEN12> line 111.

2.59, 2016-09-01 Collect a few small bug fixes and documentation improvements.


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.


dbcolscorrelate now crossreferences dbcolsregression.


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


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

2.60, 2016-09-04 Adds support for hash joins.


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.

2.61, 2016-09-05 Support left and right outer joins.


dbjoin now handles left and right outer joins with -t left and -t right.


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

2.62, 2016-11-29 A new yaml_to_db and other minor improvements.


Documentation for xml_to_db now includes sample output.


yaml_to_db converts a specific form of YAML to fsdb.


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


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.)


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.


dbfilepivot now allows the --possible-pivots option, and if it is provided processes the data in one pass.


dbroweval logs are now quoted.

2.63, 2017-02-03 Re-add some features supposedly in 2.62 but not, and add more --header options.


The option -j is now a synonym for --parallelism. (And several documention bugs about this option are fixed.)


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


Version 2.62 was supposed to have this improvement, but did not (and now does): dbfilepivot now allows the --possible-pivots option, and if it is provided processes the data in one pass.


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

2.64, 2017-11-20 several small bugfixes and enhancements


In dbroweval, the next row option previously did not correctly set up _last_fieldname. It now does.


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


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


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.

2.65, 2018-02-16 Minor release, bug fix and -F option.


dbmultistats and dbmapreduce now both take a -F x option to set the field separator.


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

2.66, 2018-12-20 Critical bug fix in dbjoin.


Removed old tests from MANIFEST. (Thanks to Hang Guo for reporting this bug.)


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


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

2.67, 2019-07-10 add support for reading and writing hdfs


dbformmail now has an "mh" mechanism that writes messages to individual files (an mh-style mailbox).


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


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


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

2.68, 2019-09-19 All programs now support automatic decompression based on file extension.


The omitted-possible-error test case for dbfilepivot now has an altnerative output that I saw on some BSD-running systems (thanks to CPAN).


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


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


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


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).

2.69, 2019-11-22 a small bugfix in dbcolstats


Filled in the the test case for autodecompress, which was missing for the 2.68 release.


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.


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.

2.70, 2020-11-12 Some small quality-of-life enhancements and corner-case bugfixes.


dbcol can now take an option -a to include all columns, allowing reordering of certain columns while passing the rest through.


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.)


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


dbcolcreate now supports --header.


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.


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.

2.71, 2020-11-16 Fix a race condition breaking test suites.


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.

2.72, 2020-12-01 A small bug and a packaging improvement.


dbcolhisto now handles the degenerate case where everything has the same value (previously it would throw "illegal division by zero").


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

2.73, 2021-05-18 Updates dbcolpercentile with --weighted, and with more ipv6.


dbcolpercentile now has a --weighted option.


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.

2.74, 2021-06-23 More ipv6.


Fsdb::Support::IPv6 package includes ipv6_fullhex to rewrite ipv6 print addresses as full, 128-bit hex values.

2.75, 2022-04-02 New type specifications in the schema to better support type conversions in python.


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.


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.


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


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).

3.0, 2022-04-04 Complete type support and accordingly bump major version.


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).


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


dbcolhisto now puts an empty value on any empty rows.


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

3.1, 2022-11-22 A post-3.0 cleanup release with minor fixes.


Type specifications in a few more programs that I missed: dbrowuniq, dbcolpercentile.


Minor documentation improvements.

3.2, 2023-10-11 Add new module dbcolsdecimate


dbcolsdecimate reduces density in timeseries data to make graphs with overly dense points visually similar but smaller.


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


Clearer installation instructions.

3.3, 2023-10-13 Quickly making dbcolsdecimate more flexible.


dbcolsdecimate now takes either relative (-p) or absolute (-P) precision, and precision now affects only subsequent columns. Also, if absolute precisions are given for all columns, data is not buffered.


John Heidemann,

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


Fsdb is Copyright (C) 1991-2024 by John Heidemann <>.

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''.


Any comments about these programs should be sent to John Heidemann