This page documents the free, open-source program pgn-extract, which is a command-line program for searching, manipulating and formatting chess games recorded in the Portable Game Notation (PGN) or something close. It is capable of handling files containing millions of games. It also recognises Chess960 encodings.

There are several ways to specify the criteria on which to extract; for instance: textual move sequences, the position reached after a sequence of moves, information in the tag fields, and material balance in the ending. Full ANSI C source and a 32-bit Windows binary for the program are available under the terms of the GNU General Public License. The program includes a semantic analyser which will report errors in game scores and it is also able to detect duplicate games found in one or more of its input files.

The range of input move formats accepted is fairly wide and includes recognition of lower-case piece letters for English and upper-case piece letters for Dutch and German. The default output is in English Standard Algebraic Notation (SAN), although there is some support for output in different notations.

Extracted games may be written out either including or excluding comments, NAGs, and variations. Games may be given ECO classifications derived from the accompanying file eco.pgn, or a customised version provided by the user.

Plus, lots of other useful features that have gradually found their way into what was once a relatively simple program!


Flag/Command-line argument summary

There follows a brief summary of the different flags taken by pgn-extract, such as is produced by the -h flag. However, you are strongly advised to read the remainder of this file before attempting to use pgn-extract in earnest.

  • -7 - output only the seven tag roster for each game. Other tags (apart from FEN and possibly ECO/Opening/Variation) are discarded (See -e).
  • -aoutputfile - the file to which extracted games are to be appended. See -o flag for overwriting an existing file.
  • -Aargsfile - read the program's arguments from argsfile.
  • -b[elu]num - restricted bounds on the number of moves in a game.
    • lnum set a lower bound of 'num' moves,
    • unum set an upper bound of 'num' moves,
    • otherwise num (or enum) means equal-to 'num' moves.
  • -cfile[.pgn] - Use file as a list of check files for duplicates.
  • -C - don't include comments in the output. Ordinarily these are retained.
  • -dduplicatefile - the file to which duplicate extracted games are to be written.
  • -D - don't output duplicate extracted game scores.
  • -eECO_file - perform ECO classification of games. The optional ECO_file should contain a PGN format list of ECO lines Default is to use eco.pgn from the current directory.
  • -E[123 etc.] - split output into separate files according to ECO.
    • E1 : Produce files from ECO letter, A.pgn, B.pgn, ...
    • E2 : Produce files from ECO letter and first digit, A0.pgn, ...
    • E3 : Produce files from full ECO code, A00.pgn, A01.pgn, ...
    • Further digits may be used to produce non-standard further refined division of games.
    All files are opened in append mode.
  • -ffile_list - file_list contains the list of PGN files to be searched - one per line (see -f).
  • -F[text] - output a FEN string comment after the final (or other selected) move.
  • -h - print a list of command-line options.
  • -Hhash -- match games in which the given Zobrist polyglot hash value occurs.
  • -? - same as --help.
  • -llogfile - Create a new logfile for the diagnostics rather than using stderr (see -l).
  • -Llogfile - Append all diagnostics to logfile (see -l).
  • -M - Match only games which end in checkmate.
  • -noutputfile - Write all valid games not otherwise output to outputfile.
  • -N - don't include NAGs in the output. Ordinarily these are retained.
  • -ooutputfile - the file to which extracted games are to be written. Any existing contents of the file are lost (see -a flag).
  • -p[elu]num - restricted bounds on the number of ply in a game.
    • lnum set a lower bound of 'num' ply,
    • unum set an upper bound of 'num' ply,
    • otherwise num (or enum) means equal-to 'num' ply.
  • -P - don't match permutations of the textual variations (-v).
  • -r - report any errors but don't extract (-r).
  • -Rtagorder - Use the tag ordering specified in the file tagorder.
  • -s - silent mode don't report each game as it is extracted. (See --quiet).
  • -S - Use a simple soundex algorithm for tag matches. If used, this option must precede the -t or -T options.
  • -ttagfile - file of player, date, result, or FEN extraction criteria.
  • -Tcriterion - player, date, eco code, hashcode, annotator or result, extraction criteria.
  • -U - don't output games that only occur once. (Use with -d to identify duplicates in multiple files.)
  • -vvariations - the file variations contains the textual lines of interest.
  • -V - don't include variations in the output. Ordinarily these are retained.
  • -wwidth - set width as an approximate line width for output.
  • -W - don't rewrite the moves into Standard Algebraic Notation.
  • -W[cm|epd|halg|lalg|elalg|xlalg|san|uci] - specify the output format to use.
    • Default (i.e., without this flag) is SAN.
    • -W (without anything following) selects the input format. I don't know if the output produced is still valid.
    • -Wepd is EPD format.
    • -Whalg is hyphenated long algebraic.
    • -Wlalg is long algebraic
    • -Welalg[PNBRQK] is enhanced long algebraic. Use the characters PNBRQK for language specific output, e.g: -WelalgBSLTDK for German.
    • -Wxlalg[PNBRQK] is enhanced long algebraic with hyphens for non-capture moves and x's for capture moves. Use the characters PNBRQK for language specific output, e.g: -WxlalgBSLTDK for German.
    • -Wsan[PNBRQK] Use the characters PNBRQK for language specific output, e.g: -WsanBSLTDK for German.
    • -Wuci is output compatible with the UCI protocol.
    • -Wcm is a legacy option that output ChessMaster format.
  • -xvariations - the file variations contains the lines resulting in positions of interest.
  • -zendings - the file endings contains the end positions of interest.
  • -Z - use the file virtual.tmp as an external hash table for duplicates. Use when MallocOrDie messages occur with big datasets.
  • -#num - output num games per file, to files named 1.pgn, 2.pgn, etc.
  • --addhashcode - output a HashCode tag.
  • --append - append matched games to an existing output file (see -a).
  • --checkfile - Use file as a list of check files for duplicates (see -c).
  • --checkmate - only output games that end in checkmate
  • --duplicates - file to write duplicate games to (see -a).
  • --evaluation - include a position evaluation after each move.
  • --fencomments - include a position evaluation after each move.
  • --fifty - only output games that include fifty moves with no capture or pawn move.
  • --fixresulttags - correct Result tags that conflict with the game outcome (checkmate or stalemate).
  • --fuzzydepth plies - positional duplicates match.
  • --help - see -h
  • --keepbroken - retain games with errors.
  • --linelength - see -w
  • --markmatches comment - mark positional and material matches with the given comment. see -t, -x
  • --nochecks - don't output + and # after moves.
  • --nocomments - see -C
  • --noduplicates - see -D
  • --nofauxep - don't output ep squares in FEN when the capture is not possible.
  • --nomovenumbers - don't output move numbers.
  • --nonags - see -N
  • --noresults - don't output results.
  • --nosetuptags - don't match games with a SetUp tag.
  • --notags - don't output any tags.
  • --nounique - see -U
  • --onlysetuptags - only match games with a SetUp tag.
  • --output - write matched games to an output file (see -a).
  • --plylimit N - limit the number of plies output (default no limit).
  • --quiescent N - position quiescence length (default 0)",
  • --quiet - No process status output (see, also, -s).
  • --repetition - only output games that include 3-fold repetition.
  • --selectonly range[,range ...] - only output the selected matched game(s)
  • --seven - see -7
  • --skipmatching range[,range ...] - don't output the selected matched game(s)
  • --stalemate - only output games that end in stalemate.
  • --stopafter N - stop after matching N games (N > 0)
  • --tagsubstr - match in any part of a tag (see -T and -t).
  • --totalplycount - include a tag with the total number of plies in a game.
  • --version - print current version number and exit.

Usage and flags/command-line arguments

pgn-extract takes an arbitrary number of game scores as input and outputs zero or more of these games, typically in English Standard Algebraic Notation (SAN). Which of the input games are output, and the style of the output, depend upon the particular set of command line flags passed to pgn-extract. The general form for calling pgn-extract is as follows:

pgn-extract [flags] [input-game-files]

In its simplest form, calling pgn-extract with no arguments will cause it to read games from its standard input, check them and reproduce those without errors in SAN notation on its standard output.

Game input format

This program's principle aim is to be able to read PGN files and output games of interest. It follows that the input should look reasonably like PGN to start with. This means that it doesn't cope well with files that contain news article or mail headers, for instance, although it does make an attempt to skip text that is obviously not game related between games. Having said that, it does not require the move text be in Standard Algebraic Notation (SAN). It will accept quite a few common formats including:

  • Algebraic
  • Long Algebraic
  • various commonly-used intervening characters, such as : - x
  • Dutch and German upper case piece letters. (Support for Russian piece letters is in prototype.)
  • lower-case English piece characters (except that it will always prefer 'b' to mean a pawn move rather than a Bishop move).

It does not require that there be any move numbers or PGN headers preceding a game, as long as the move text is terminated by a valid result designation: *, 1-0, 0-1, 1/2-1/2 (1/2 is also accepted). This makes the program reasonably suitable for entering raw game text and having it reformatted in proper SAN with a full set of headers.

File of PGN files (-f)

Normally, the input files from which games are to be extracted are listed on the command line:

pgn-extract file1.pgn [file2.pgn ...]

An alternative to listing the game files on the command line is to list their names, one per line, in a file which is then given after the -f flag:

pgn-extract -ffile_list

In order to save the output in a file rather than standard output, use -o, --output, -a, --append to indicate the output file name, for instance:

pgn-extract -oall.pgn file1.pgn file2.pgn file3.pgn
pgn-extract --output all.pgn file1.pgn file2.pgn file3.pgn

While pgn-extract can be used simply to check and reformat all the input games, it is more usual to use it to select subsets of the input games. Several different criteria are available on which to extract: move variations, information in the tag fields, and material balance in the ending, for instance. All of these criteria are described in detail below.

Output files (-o, --output, -a, --append)

In order to output all matched games to a single new file, the -o flag is used:

pgn-extract -onew.pgn file1.pgn file2.pgn

This has the effect of creating new.pgn from the contents of file1.pgn and file2.pgn. The games in both source files are checked and rewritten, if necessary, into SAN. Any previous contents of new.pgn will be lost with the -o flag. In order to avoid this and append to an existing file, use the -a flag:

pgn-extract -anew.pgn file1.pgn file2.pgn

Note that there must be no space between either -o or -a and the output file name.

The long-form --output and --append are provided as alternatives to -o and -a, respectively. In these cases, there must be a space between the flag and the output filename. For instance:

pgn-extract --output new.pgn file1.pgn file2.pgn
pgn-extract --append new.pgn file1.pgn file2.pgn

Check for errors (-r)

Check the input files for errors but do not output any matched games. Useful for cleaning up files of games before proper processing.

pgn-extract -r file.pgn

Useful with -s (silent mode) for checking a big file of games without having progress reported and just seeing the errors.

Retaining games with errors

Normally, pgn-extract reports games with errors but does not output them. Games with errors may be output with the --keepbroken argument. The errors are still reported but the moves from the point where the error was detected onwards are placed in a comment rather than being retained as part of the game.

Log files (-l, -L)

Error messages and verbose reporting is done to the standard error output unless the -l or -L flag is used. Both are immediately followed by the name of a file to which a log should be written. The -l flag creates a new log file, while -L appends to an existing log file:

pgn-extract -llog.txt file.pgn
pgn-extract -Llog.txt file.pgn

This option is useful in combination with -r (report) to generate diagnostic information without outputting games while game data is being checked and cleaned.

A log file will contain only error reports if the -s (silent) flag is used.

Variations (-v, -x and -P)

There are two distinct ways to specify variations of interest; positional variations (the the -x flag) and textual variations (the -v flag). The major difference between the two is that positional variations specify a complete move sequence whose end position is the primary point of interest, whereas textual variations allow incomplete and fuzzy move sequence matches on the text of a game to select games. Whilst it is possible to use both flags together, this would be unusual as a game must match with both to be extracted.

  • Positional Variations (-x)

    The variations in which you are interested should be placed in a file whose name is supplied with the -x flag. For instance:

    pgn-extract -xvars

    where each variation is listed on a single line in the file vars (the filename is immaterial). The following set of moves:

    e4 c5 Nf3 d6 d4 cxd4 Nxd4 Nf6 Nc3 a6

    indicates that you wish to pick up all games reaching the Najdorf variation position of the Sicilian Defence. Games reaching the end position of this sequence are selected regardless of the route that was taken to reach it. This allows various transpositional sequences to be specified by quoting just one line to reach the required point. Therefore, games employing the following move order will be picked up by quoting the line above.

    e4 c5 Nc3 d6 Nge2 Nf6 d4 cxd4 Nxd4 a6

    A position is considered to match a required variation if it generates the same board hash value. In the interests of reasonable efficiency, no attempt is made to actually examine the state of the board. There is, therefore, the potential for false hits but in my usage of pgn-extract I have not found this to be a problem.

    With this option, games are only searched to a depth approximately equal to the length of the longest positional variation, in order to make processing of large data sets faster than with a search of the whole game.

    A comment line may be placed in a variation file by using a '%' as the first character of the line. Move numbers are optional within the list of moves.

    Positional matches are also available using a FEN description of the desired position. See the description of the -t flag for how to specify a FEN position, and the -F flag for a simple way to generate a FEN description from a game score.

  • Textual Variations (-v)

    With this option, the matching is purely textual in nature, in contrast to the -x flag. The -v flag works by string matching on the input text of moves, so there is no facility for picking up transpositions automatically. The variations in which you are interested should be placed in a file whose name is supplied with the -v flag. For instance:

    pgn-extract -vvars

    Each variation should be listed on a single line in the file vars (the filename is immaterial). The move sequence:

    e4 c5 Nf3 d6 d4 cxd4 Nxd4 Nf6 Nc3 a6

    indicates that you wish to pick up all games following the normal move order of the Najdorf variation of the Sicilian Defence, and

    d4 Nf6 c4 e6 Nc3 Bb4

    that you are interested in Nimzo-Indian games. The order in which the moves are played by either White or Black is immaterial. All combinations are tried, so the ordering:

    c4 e6 Nc3 Bb4 d4 Nf6

    will produce the same set of matches as the previous ordering of the Nimzo-Indian moves (see the -P flag for how to prevent this).

    A comment line may be placed in a variation file by using a '%' as the first character of the line. Move numbers are optional within the list of moves.

    As transpositions are not picked up automatically with this flag, if you also wanted to recognise the following as a Najdorf, you would have to add this line to the variations file in addition to that given above:

    e4 c5 Nc3 d6 Nge2 Nf6 d4 cxd4 Nxd4 a6

    However, because of the way in which the matching is done, it is possible to specify slight alternatives on the way in which individual moves are written. Notational alternatives for a single move are just written separated from each other with a non-move character. This variation specifies both the shorter and longer ways of writing the captures in a Najdorf:

    e4 c5 Nf3 d6 d4 cxd4|cd Nxd4|Nd4 Nf6 Nc3 a6

    However, given the variety of possible ways of writing various moves in non-SAN format, e.g.


    variation lists can get quite messy and I believe that this approach is best avoided by ensuring that the input is proper SAN and only using SAN notation in the variations file. In this way, the alternative-separator can then be used purely for indicating genuine alternative moves at that point, e.g.

    e4 c5 Nf3 d6 d4|d3

    An important point when listing moves is that check and mate indicators should be included where appropriate, otherwise moves incorporating these characters in games to be searched will fail to match.

    There is little point in using the -v flag in preference to the -x flag if you are only interested in finding games that reach a particular position. The real use for -v is when you wish to pick up games in a more general way. For instance, the character '*' may be used in place of any move to indicate that you don't care what was played at that point. So the following:

    * b6

    means that you are interested in all games in which Black replied 1 ... b6 regardless of White's first move. The sequence:

    d4 * c4 * Nc3 *

    will pick up Nimzo-Indian, Grunfeld, King's Indian, etc. defences. This notation is not possible with positional variations.

    In addition, the character '!' may be used in front of any move to indicate that you wish to disallow particular moves from matching at this point. For instance, if you want to find Sicilian games where White did not reply with Nf3 at move 2 you would specify:

    e4 c5 !Nf3

    If you wished to disallow 2.Ne2 as well then

    e4 c5 !Nf3|Ne2

    does the job. (Adding parentheses makes no difference as the '!' is applied to all of the following move string.)

    Care should be taken combining '!', '*' and variation permutations (see the -P flag). Disallowed moves take precedence over '*' moves. If a single disallowed move is found in a game within the length of the variation, that game is excluded. This was the most sensible interpretation that I could find to place on this usage.

  • Textual Variation Permutations (-P)

    Normally, all permutations of a textual variation (see the -v flag) are tried against the moves of a game. This cuts down on the number of separate transpositional orderings that it is necessary to list, at the cost of slower matching of each game. If the following were used to look for Nimzo-Indian games:

    d4 Nf6 c4 e6 Nf6 Nc3 Bb4

    a side-effect would be that it will also pick up games which start as:

    1. c4 Nf6 2. Nc3 e6 3. d4 Bb4

    for instance. The -P flag requests that textual variations are matched against the moves of the game strictly in the order in which they are listed, without trying different orders. So, if you want to find only those games that follow a particular move order, use this flag to suppress permutations.

Duplicate games (-d, --duplicates and -D or --noduplicates, plus -Z)

If either the -d, --duplicates or -D flag is used, pgn-extract attempts to recognise duplicate extracted games. Using the -d or --duplicates flag indicates that you wish copies of the duplicate games to be written to the indicated file:

pgn-extract -ddupes.pgn -ounique.pgn file.pgn
pgn-extract --duplicates dupes.pgn --output unique.pgn file.pgn

will both extract from file.pgn the unique set of games into unique.pgn and the duplicates (i.e., the second and subsequent copies of a game) to dupes.pgn. A comment identifying in which file a duplicate was found precedes the first duplicate found in that file and each duplicate game has a prefix comment indicating the file in which the first version was found. Note that there must be no space between -d and the filename but a space between if --duplicates is used.

With the -D flag duplicate games are suppressed from the output. These two flags are mutually exclusive, therefore.

Duplicates are identified by comparing a hash value for the board of the end positions of extracted games and an additional cumulative hash value generated from the move sequence. If these both values match then games are considered to be duplicates. This is not guaranteed to be exact but it gives a good approximation. If the position is important but the move sequence is not then use --fuzzydepth.

You should note that games are only considered to be duplicates on the basis of the moves played. It may be that a game considered to be a duplicate contains annotations and variations not present in the one found earlier, so it might be necessary to do some swapping around to obtain those you really wish to retain. You should, therefore, use the -D flag with caution if you are trying to reorganise your master collection rather than selecting out specific games for examination. (See also the -U flag.)

Detecting duplicates requires memory for the storage of a hash table containing information on each game. Large databases can result in a MallocOrDie error. If this is the case, try using the -Z flag which forces pgn-extract to store its hash table externally, in a file called virtual.tmp. Each game requires 16 bytes of file space. Clearly, if a very large database is being processed, there is a risk of filling up the available file space if there is insufficient available.

Positional duplicates match

This flag allows a match on the basis of board position at the indicated number of plies or the end of the game. The flag is followed by the ply depth at which matches are to be considered. The value 0 is used to request matching at the end of games. It should always be used in combination with at least one of: -d/--duplicates, -D/--noduplicates, -U.

In contrast to the --duplicates matching, the match does not consider the move sequence used to reach the match position.

For example:

pgn-extract --fuzzydepth 40 -D games.pgn

would suppress from the output multiple copies of games reaching identical positions after 40 ply.

The following example would suppress the unique games and store the games considered to be duplicate at their final position in dupes.pgn:

pgn-extract --fuzzydepth 0 -U -ddupes.pgn games.pgn

Suppression of unique games (-U or --nounique)

The -U flag suppresses output of the first occurrence of a particular game. This is useful when combined with the -d flag as a means of identifying just those games that are duplicated in a list of multiple files. As the duplicate games are commented with the file in which they were located, it then becomes possible to prune a set of files containing common games. For instance, suppose oldfile.pgn contains a set of games without duplicates, and you wish to know which games in newfile.pgn already occur in oldfile.pgn:

pgn-extract -U -ddupes.pgn oldfile.pgn newfile.pgn

will write to dupes.pgn the duplicate games so that you can go through newfile.pgn and remove them. Of course, if you simply want to hold the combined set of unique games in a single file you would use something like:

pgn-extract -D -onewset.pgn oldfile.pgn newfile.pgn

See Duplicate Games for dealing with MallocOrDie errors.

Check files for duplicates (-c, --checkfile)

Check files contain games that are to be used in duplicate detection, but not to form part of the output. If the filename appended to the argument has a .pgn/.PGN suffix it is assumed to be a single file of games. If it does not have this suffix then it is assumed to be a file containing a list of the names of PGN game files, one per line, to be used as check files.

A typical use for this is to select new games of interest from a file that probably contains games that exist elsewhere. In the following example, we wish to select Nimzo-Indian games from newfile.pgn that don't already occur in the master file nimzo.pgn:

pgn-extract -cnimzo.pgn -vnimzo.var -D -onewnimzo.pgn newfile.pgn

The games in nimzo.pgn act as the source for duplicate detection so duplicates of these will be suppressed (the -D flag). Only those games from newfile.pgn that are not in nimzo.pgn will be output to newnimzo.pgn. Contrast this behaviour with the following, which would create a new master file of games from the combination of nimzo.pgn and newfile.pgn:

pgn-extract -vnimzo.var -D -onewnimzo.pgn nimzo.pgn newfile.pgn

--checkfile is available as an alternative to -c and must be followed by a space before the filename, e.g.:

pgn-extract --checkfile nimzo.pgn -vnimzo.var -D -onewnimzo.pgn newfile.pgn

Matching on tag criteria (-t)

There are two ways to specify that you wish to use information in the tag fields as extraction criteria: the -t flag and the -T flag. The -t flag takes a file name argument and is the preferred method because of its ease of use and greater flexibility:

pgn-extract -ttags games.pgn

where tags is an arbitrary file name. In the file are listed tag name and value pairs corresponding to the extraction criteria you wish to use. Each line of this file should be of the form:

PGN-Tag-name Tag-string
for instance:
White "Tal"

(note the need to include double quotes around the tag value). This requests that only those games where Tal had the White pieces are to be considered for extraction. If you wish to limit the year in which those games were played you might list:

White "Tal"
Date "1962"

Multiple pairs with the same tag name are or-ed together so:

% Find games in the period 1960-1962.
Date "1960"
Date "1961"
Date "1962"

will select all games from the three listed years. Note that comments may be included in the tag file.

In general, tags names that differ are and-ed together, so:

White "Tal"
Black "Fischer"
Date "1962"
Result "1-0"

selects only those games that Tal won with the White pieces against Fischer in 1962.

It is important to note that:

White "Tal"
Black "Tal"

does not find all games played by Tal, but only those that he played against himself. In order to overcome this, I have introduced a non-PGN tag that should only be used in the extraction criteria file:

Player "Tal"
Date "1962"

finds all games from 1962 in which Tal had either the White pieces or the Black. In effect, the White and Black player lists are or-ed together rather than and-ed using this pseudo-tag.

By default, prefix matching on tag values is done so that a criterion should be a prefix of the complete Tag string. Thus,

Player "Karpov"

would match:

[White "Karpov"]
[White "Karpov, A"]
[White "Karpov, An"]
[White "Karpov, Alexander"]

but not

[White "Anatoli Karpov"]

Matches anywhere within a tag, rather than just the prefix, are performed if the --tagsubstr flag is used and the example immediately above would be matched in this case.

See the -S flag for a soundex facility with tag matching.

All tag criteria except ECO classification are checked before the moves of the game in the interests of efficiency (tag checking is relatively fast whereas positional checking of the game is not). Only once the game has been processed is it checked to see whether an ECO tag match has been requested. The consequence of this is that using the -e flag in combination with ECO tag criteria you can search for games in particular ECO lines without an ECO tag having been present in the input form.

  • FEN positional matches with -t

    Use of a FEN tag with the -t flag has a special meaning. Rather than using this to match FEN tags in the header of a game, a FEN description is used to indicate a search for a positional match (similar to use of the -x flag). If a FEN description is provided with the -t flag, the indicated position is searched for in each game processed, and only those games that reach the indicated position are output. A FEN tag-pair for the starting position would be described by:

    FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"

    The position after the two moves e4 c5 would be:

    FEN "rnbqkbnr/pp1ppppp/8/2p5/4P3/8/PPPP1PPP/RNBQKBNR w KQkq c6 0 2"

    However, note that it is only the board position that is relevant for the match. Castling rights are not considered in the determination of equality.

    See details of the -F flag for a simple way to generate a FEN description from a game score.

    There is a variation on use of a FEN with -t flag. The pseudo tag FENPattern takes a FEN-like description of a board containing meta characters that allow a fuzzy board match. In addition to the standard FEN characters having their usual meaning (1, 2, ... 8, R, N, B, etc.), the following meta characters are used:

    • ? - match any square. The square may be occupied or unoccupied.
    • ! - match any occupied square. The square may be occupied by a piece of any type and colour.
    • A - match a single White piece.
    • a - match a single Black piece.
    • * - match zero or more squares, occupied or unoccupied.
    • [xyz] - match any of xyz, where xyz represents any of the English piece-letter names (KQRNBPkqrnbp) and is case-sensitive. In addition, 'A' and 'a' (as defined above) are available. For instance: [Qq] matches either a White or Black queen; [BbNn] matches any White or Black bishop or knight; [Ar] matches any White piece or a Black rook.
    • [^xyz] If the first character inside the square brackets is '^' then the match is inverted; i.e., match any piece that is not listed. For instance, [^BbNn] matches any piece that is not a White or Black bishop or knight.

    Ranks within the pattern are separated with a '/' character, as usual, but there should be no characters other than the board position.

    For instance:

    FENPattern "rnbqkbnr/*/*/*/*/*/*/RNBQKBNR"

    would match any board in which at least every non-pawn piece is on its starting square.

    FENPattern "?????rk?/?????aaa/*/*/*/8/P[BP]P*/??KR????"

    would match a board in which Black has a Kingside castled position behind three Black pieces (not necessarily pawns), White has a Queenside castled position with either a White pawn or bishop on b2 and the third rank is empty.

  • Date and Elo Matches with -t

    From a -t tag file, more complex matching of dates and Elo values may be performed by placing an operator between the tag name and the tag string to be matched:

    Date < "1962"

    would only match games played before 1962. Only the year value participates in the matching process, as this is done using integer values rather than strings.

    WhiteElo >= "2500"

    only matches games where White is a strong player. Probably of more general use is another pseudo-tag that I have introduced purely for this purpose: Elo.

    Elo >= "2500"

    matches games in which either player has an Elo tag matching that relationship. The operators allowed are >, >=, <, <=, =, and <> (not equal to).

Tag criteria on the command line (-T)

An alternative to the -t flag is the -T flag, for use where command line arguments are more convenient - perhaps where pgn-extract is being invoked from another program. The tag coverage is not as extensive as with a tag file, and the syntax is rather cumbersome. It is used as follows: after the -T comes a single letter from the limited set [abdeprw] to select string prefixes of the tag fields of a game. For instance:

  • -TaAnnotator - Extract games Annotated by Annotator.
  • -TbPlayer - Extract games where Player has the Black pieces.
  • -TdDate - Extract games played on Date.
  • -TeEco - Extract games with ECO designation Eco.
  • -ThHashCode - Extract games with HashCode designation HashCode.
  • -TpPlayer - Extract games where Player has either colour.
  • -TrResult - Extract games with result Result.
  • -TwPlayer - Extract games where Player has the White pieces.

For example,

pgn-extract -TwTal -TbFischer file.pgn

would extract games from file.pgn in which Tal had the White pieces and Fischer the Black.

Criteria of the same tag type are or-ed together, so

pgn-extract -Tr1-0 -Tr0-1 file.pgn

extracts only decisive games.

Criteria of different tag types are and-ed together so

pgn-extract -TwTal -Td1962 -Tr1-0 file.pgn

would extract only those games in which Tal played with the White pieces in 1962 and won.

The ECO classification (see the -e flag) is performed before attempting to match an ECO tag, so:

pgn-extract -TeA01 -e file.pgn

will perform ECO classification on the input file and extract games with ECO classification A01 (Nimzo-Larsen attack), for instance.

  • Date Matches with -T

    A simple form of relational date matching is available. A date year may be prefixed with either 'b' or 'a' in order to match games played either before or after the specified date. This assumes that the date is stored in the game's date tag string in the normal form: YYYY.MM.DD


    pgn-extract -Tdb1962 file.pgn

    will look for games played before 1962. A much fuller capability is available in tag files with the -t flag.

Argument descriptions in a file (-A)

It can be inconvenient to repeatedly type long argument lists on the command line. The -A flag makes it possible to list arguments in a file, rather than on a command line. Each argument line within the file must be immediately preceded by a ':' (colon) character. Consider selecting games by Tal from a file caro.pgn and writing them to talgames.pgn. Using command line arguments, this would have the following form:

pgn-extract -TpTal -otalgames.pgn caro.pgn

We can do the same job placing the argument list in the file args:

% Select games by Tal.
% Where to output the matched games.

and the same selection made with:

pgn-extract -Aargs caro.pgn

Note that comments may be included using a '%' character.

Each argument should be listed on its own line, and all the arguments are available in this way. The PGN source files may also be listed in the argument file. They must be listed one per line, with a preceding colon character. So an alternative for the above would be:

% Select games by Tal.
% Where to output the matched games.
% The game files to be read.

and the command invoked as simply:

pgn-extract -Aargs

The -t, -v, -x, -z, and -R flags have slightly special treatment in an argument file. Where the tags, variations, positions, endings and/or roster ordering are to be read from files of those names, say, then the format of these arguments in the argument file might be as you would expect:


However, within an argument file, the file names are optional and, where omitted, the data that would have been stored in a file for these flags is listed on lines immediately following. For instance, an alternative to:


we could say:

Player "Tal"

Notice that no colon should be present on the lines following the flag line. In the following example, we select games won by Tal as White reaching a particular position in the Caro Kann:

White "Tal"
Result "1-0"
e4 c6 d4 d5 exd5 cxd5
% Which game files to process.

The arguments file may, itself, also contain -A arguments. This should make it possible to build up hierarchies of game selection criteria if desired. However, beware that there is no check for circularities in the dependencies.

Matching only games with or without SetUp tags

Games with non-standard starting positions are indicated with a pair of tags: FEN and SetUp. Such games may be exclusively selected via --onlysetuptags or rejected via --nosetuptags.

Outputting games not matched (-n)

The -n flag will cause all valid games not output via other criteria to be saved in a given file. The purpose of this is to make it easier to reorganise files in different ways. For instance, if you wish to remove all of the games played by Tal from one file, you might do:

pgn-extract -TpTal -otalgames.pgn -nothers.pgn file.pgn

After which, the file others.pgn will contain all of the valid games from the original file, with the exception of Tal's.

Outputting selected matched games (--selectonly)

The --selectonly flag takes a comma-separated list of one or more numerical arguments representing ranges. For instance:

requests that only the first 10, the 15th and the 89th-94th matched games are output. For instance, if only the first game played against Fischer is required from a file of Tal games, the following would be used:
pgn-extract -TpFischer --selectonly 1 talgames.pgn

For the first three and the tenth, the arguments would be:

pgn-extract -TpFischer --selectonly 1:3,10 talgames.pgn

The numbers in the list must be in strictly ascending order with no overlaps. Note that it is the number of matches that is used to select against and not the number of games in the input.

Note that, once the required games have been output, the program will terminate and not continue processing the rest of the input files.

Suppressing the output of selected matched games (--skipmatching)

The --skipmatching flag takes a comma-separated list of one or more numerical arguments representing ranges, for instance:

requests that the first 10, the 15th and the 89th-94th matched games are not output. For instance, if the first game played against Fischer is not required from a file of Tal games, the following would be used:
pgn-extract -TpFischer --skipmatching 1 talgames.pgn

To suppress the first, second, third and fifth, the arguments would be:

pgn-extract -TpFischer --skipmatching 1:3,5 talgames.pgn

The numbers in the list must be in strictly ascending order with no overlaps. Note that it is the number of matches that is used to skip against and not the number of games in the input.

stop after matching N games (--stopafter)

The --stopafter flag takes a single numerical argument N (N > 0) to request that only the first N matched games are output. Note that it is the number of matches that determine the stopping point and not the number of games processed.

This is useful when processing large files but only a small sample of games are required. For instance, the following example would stop after it finds the first game by Petrosian in megafile.pgn:

pgn-extract -TpPetrosian --stopafter 1 megafile.pgn

Don't output tags (--notags)

The tags for a game will not be output.

Suppress annotations in the output (-C -N -V)

If comments (-C or --nocomments), NAGs (-N or --nonags) and/or variations (-V or --novars) are not required in the output then these can be suppressed by using one or more of these flags.

Suppressing move numbers (--nomovenumbers)

Move numbers can be suppressed from the output with --nomovenumbers. Used in combination with --notags, --noresults, -C, -N, and -V this can be used to output just the moves of a game:

pgn-extract --nomovenumbers --noresults --notags -C -N -V file.pgn

If it is desired to have all the moves on a single line, use the -w flag as well.

See also the --plylimit flag.

Suppressing results (--noresults)

Results at the ends of games and variations can be suppressed from the output with --noresults. See suppressing move numbers for a possible use.

Limiting the number of plies (>= 0) output (--plylimit)

The number of moves (actually plies) output for a game can be limited by using --plylimit. This must be followed by the maximum number of plies to be output for a game. For instance,

pgn-extract --plylimit 10 --nomovenumbers --notags file.pgn

will output games up to a maximum of 10 plies (including variation lines), without game tags and no line numbers.

See also --quiescent.

Note: If the game has not ended before the ply limit is reached then * will be used as the terminating result to indicate an incomplete game (see --noresults for how to suppress this.)

Position quiescence length (--quiescent)

Used only in conjunction with --plylimit, this argument is used to defer termination of the output until the position has been quiescent for the given number of ply. Stability is defined as the absence of captures, checks and promotion. For instance:

pgn-extract --plylimit 20 --quiescent 3 file.pgn

will output games up to 20 ply but only if there have been no captures, checks or promotion moves for the past 3 ply. If this condition is not met then moves will continue to be output until that condition is satisfied, or the end of the game is reached.

Setting bounds on the number of moves or ply in a game (-b and -p)

The -b flag allows you to select games which have a number of moves within the bounds you set. Alternatively, -p allows finer control over the number of ply. In all of the following examples of usage, if -b is replaced with -p then the numerical values apply to 'ply' rather than 'moves'.

You can set a lower bound on the number of moves by using -bl ('l' = lower bound), or an upper limit by using -bu ('u' = upper bound). Both are followed by the number of moves so

pgn-extract -bu20 file.pgn

will find brevities of 20 moves or less, whilst

pgn-extract -bl60 file.pgn

will find games of 60 moves or move. Bounds may be combined so

pgn-extract -bl30 -bu40 file.pgn

will find games in the range [30..40] moves. If neither 'l' nor 'u' is used, but just a number following the -b, this means that the number of moves must exactly match that number. Alternatively, 'e' can be used to stand for 'equal to'. The following are equivalent and find all games of exactly 35 moves.

pgn-extract -b35 file.pgn
pgn-extract -be35 file.pgn

Matching only games that end in checkmate (-M or --checkmate)

The -M flag requests that only games that end in checkmate are matched:

pgn-extract -M file.pgn

Matching only games that include fifty moves with no capture or pawn move (--fifty)

The --fifty flag requests that only games that contain at least fifty moves with no capture or pawn move are matched:

pgn-extract --fifty file.pgn
See --markmatches for a way to mark the position at which this occurs.

Matching only games that include three-fold repetition (--repetition)

The --repetition flag requests that only games that include three-fold repetition matched:

pgn-extract --repetition file.pgn

Games will be matched regardless of whether the game was ended by that occurrence. See --markmatches for a way to mark the position at which the repetition occurs.

Matching only games that end in stalemate (--stalemate)

The --stalemate flag requests that only games that end in stalemate are matched:

pgn-extract --stalemate file.pgn

Matching only games that contain an underpromotion (--underpromotion)

The --underpromotion flag requests that only games that contain an underpromotion are matched:

pgn-extract --underpromotion file.pgn

ECO Classification (-e)

A PGN file of ECO classifications is distributed with this version. I believe that this was put together by Ewart Shaw, Franz Hemmer and others, to whom appropriate thanks is due. The -e flag requests pgn-extract to add/replace ECO classifications in the games it outputs. This is done by firstly reading a file of ECO lines in PGN format (eco.pgn in the current directory, by default) and building a table of resulting positions. As the games are then read they are looked up in the table to find a classification. The deepest match is found. A match is allowed within six half moves of the length of the ECO line. The supplied file has ECO, Opening, and Variation tag strings for many lines. If present, pgn-extract will add/replace these as well as SubVariation tags if available.

An alternative file to the default eco.pgn may be supplied in two ways:

  • Appending a file name to the -e flag


    Note that there must not be a space between the -e and the name of the file, otherwise the default ECO file will be assumed.

  • By setting the environment variable ECO_FILE to the full path name of the file. Under Windows this can be done with

    set ECO_FILE=full-eco-file-path

    at the Cmd window prompt, or more permanently via the System/Environment/Advanced area. Under UNIX csh this can be done with

    setenv ECO_FILE full-eco-file-path

    in the .cshrc, for instance.

Having the ECO data read as plain text on program startup has the obvious disadvantage that there is a high initial time overhead. On the other hand, it has the advantage that users may add their own classifications to the file very easily. It is fairly demanding of memory, so you advised not to combine this with duplicate detection (-U, -D and -d), which can also consume a lot of memory with big databases.

Because an ECO tag match with either the -t flag or the -T flag is delayed until after ECO classification, this makes it relatively easy to select games with particular ECO codes even if they weren't present in the source form.

Usage of -e with the Seven Tag Roster flag (-7) results in the ECO tags (ECO, Opening, Variation, SubVariation) being included in the output games.

Separate output files (-#, -E)

The -# and -E flags permit the output to be split into multiple files. However, be warned that where the input involves a lot of games, these flags might result in the creation of a large number of output files.

The -# flag takes an unsigned integer argument specifying the maximum number of games to output to a single file. Successive output files are numbered 1.pgn, 2.pgn, etc. Any existing contents of these files are always overwritten on each run of pgn-extract.

pgn-extract -#250 file.pgn

will check and split file.pgn into separate files of, at most, 250 games each.

pgn-extract -#1 file.pgn

will split file.pgn into separate files containing only a single game each.

The -E flag normally takes a numeric argument of value 1, 2, or 3. This is used to indicate the level of subdivision required based upon the ECO tag found in a game.

pgn-extract -E3 file.pgn

will fully subdivide file.pgn into separate files based on the full ECO code of each game, with names such as B03.pgn, A01.pgn, D45.pgn, etc. If a game does not contain an ECO tag, or the tag appears to be malformed, it will be written to a file called, noeco.pgn. All of these files are written to in append mode, so that existing contents are not lost. However, beware of using an input file whose name is the same as one that will be written to by this operation. This could lead to infinite operation.

Level 1 classification uses just the initial letter of the ECO classification to append to files A.pgn, B.pgn, etc. Level 2 uses the initial letter and first digit, producing A0.pgn, B3.pgn, etc.

In fact, values greater than 3 may be used to produce separation of even finer granularity if more than two digits have been used in the classification of a game.

Soundex matching (-S)

There is a simple soundex algorithm available that attempts soundex matches on White, Black, Site, Event, and Annotator tags if the -S flag is used in combination with either the -t flag or the -T flag. The -S flag should precede all -t and -T arguments. It should be noted that the soundex matching does produce false matches.

Output line length (-w or --linelength)

The -w flag allows an approximate line length to be set for output. Normally games are output with lines up to a maximum of 75 characters. Use the -w flag if you want longer output lines. For instance, you might want all the moves of a game to appear on a single line. You would get this effect by specifying -w1000 (say):

pgn-extract -w1000 file.pgn

If some games are more than 1000 characters long then just increase the value.

Output format and language (-W)

By default, pgn-extract rewrites the game score into English Standard Algebraic Notation (SAN) because it is reasonably flexible about the input form that it will accept. To prevent it from rewriting the original form of the moves it reads, use the -W flag.

  • By itself, -W outputs the moves using the input text.
  • Using -Whalg writes the moves in hyphenated long algebraic (e.g., e2-e4).
  • Using -Wlalg writes the moves in long algebraic form (e.g., e2e4).
  • Using -Welalg writes the moves in enhanced long algebraic form (e.g., Ne2e4, e5d6ep). The purpose of enhanced long algebraic form is to reduce the amount of chess-specific knowledge that a post-procesing program might need in order to interpret a chess game. For instance, in order to provide a visualisation.
  • Using -Wxlalg writes the moves in enhanced hyphenated long algebraic form with capture information (e.g., Ng1-f3, Nf6xd5, e5xd6ep). The purpose of enhanced long algebraic form with hyphens and x's is to further reduce the amount of chess-specific knowledge that a post-processing program might need in order to interpret a chess game.
  • Using -Wuci causes the moves of the game to be output in a format that should close to being suitable for input to a UCI-compatible engine. The output format is the same as with -Wlalg but all comments, NAGs, variations, move numbers and checks removed. In addition the whole game is output on a single line.

Output using non-English piece letters is possible using a variation of the -Wsan flag. This flag may have a six-letter suffix indicating the letters to be used in representing pawn, knight, bishop, rook, queen and king in game scores and diagrams. So:

pgn-extract -WsanPNBRQK ...

would output in the (default) English notation, and

pgn-extract -WsanBSLTDK ...

would output in German. Note that the letter for a pawn is required because board positions are sometimes output when an error is detected in a game score.

-Wepd outputs in EPD (Extended Position Description). A game is output as a sequence of EPD descriptions of the position at the start of the game, and following each move. Each EPD line contains the FEN board description, the active colour, castling availability and en passant target square. A c0 comment contains a synopsis of the player, event, site and date tags from the game's header.

-Wuci outputs in long-algebraic notation (-Wlalg) but also strips the game of everything apart from its moves, tags and result. It provides the equivalent of using the following multiple arguments:

-Wlalg -C -N -V -w5000 --nomovenumbers --nochecks

Use the --noresults and --notags options if tags and results are also to be removed.

-Wcm is an obsolete legacy flag and outputs the moves in what I believe to be (or used to be) ChessMaster format.

Forsyth-Edwards Notation (FEN) descriptions (-F)

The -F flag provides a way to generate a suitable FEN description of an arbitrary position. The -F flag, with no additional text, causes pgn-extract to output a FEN description of the final position reached in a game, within the text of a comment. For instance, suppose you were interested in finding games that reach the position after the following moves.

d4 Nf6 c4 e6 Nf3 b6 Nc3 Bb7 e3 Bb4 Bd3 O-O O-O Bxc3 bxc3 c5 *

Storing these moves in the file fen.pgn and running

pgn-extract -F fen.pgn

would generate the score:

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]

1. d4 Nf6 2. c4 e6 3. Nf3 b6 4. Nc3 Bb7 5. e3 Bb4 6. Bd3 O-O 7. O-O Bxc3 8.
bxc3 c5 
{ "rn1q1rk1/pb1p1ppp/1p2pn2/2p5/2PP4/2PBPN2/P4PPP/R1BQ1RK1/ w - c6 0 9" } *

The FEN string thus generated might then be used with the -t flag to match games reaching that position. The FEN string above would be cut and pasted to an argument file and used with the -t flag to supply matches:

FEN "rn1q1rk1/pb1p1ppp/1p2pn2/2p5/2PP4/2PBPN2/P4PPP/R1BQ1RK1/ w - c6 0 9"

A variation of -F, when immediately followed by a text string, allows a FEN position of an arbitrary point in a game to be output as a comment. For instance, the following:

pgn-extract -FputFENhere file.pgn

would replace all comments of the form {putFENhere} encountered in games in file.pgn with a comment containing the FEN position at that point. For instance, if file.pgn contained:

d4 Nf6 c4 e6 Nf3 b6 {putFENhere} Nc3 Bb7 e3 Bb4 Bd3 O-O O-O {putFENhere}
Bxc3 bxc3 c5 *

the output would be:

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]

1. d4 Nf6 2. c4 e6 3. Nf3 b6 {
rnbqkb1r/p1pp1ppp/1p2pn2/8/2PP4/5N2/PP2PPPP/RNBQKB1R w KQkq - 0 4 } 4. Nc3
Bb7 5. e3 Bb4 6. Bd3 O-O 7. O-O {
rn1q1rk1/pbpp1ppp/1p2pn2/8/1bPP4/2NBPN2/PP3PPP/R1BQ1RK1 b - - 4 7 } 7...
Bxc3 8. bxc3 c5 *

The text immediately following -F is arbitrary but an exact, complete match is required to output a FEN string.

See --fencomments for the option to add a FEN comment after every move, including the final one.

Material matches (-z)

The -z flag takes a filename of material balances for which you wish to search in games. The basic structure of the file is one or more lines of the form

pieces1 pieces2

Pieces1 and pieces2 are lists of English piece letters for the material for the two sides that you wish to look for in a game. For instance:

rp nb

looks for an game in which a lone Rook and Pawn for one side are competing against a lone Knight and Bishop for the other.

Text may be added after the piece lists as a form of comment.

A comment line may be placed in a material balance file by using a '%' as the first character of the line.

The --markmatches flag may be used to add a comment at the point that the match is found.

The case of the letters is immaterial, there is no need to include Kings in the description, and the order of the pieces does not matter. Apart from Kings, if a piece letter is not listed for a side then that piece is not present within that side's material. A match will be tested for from both White and Black's point of view, so the example above matches the same games as:

nb rp

Some notation may be added after any piece letter, typically to indicate something about the number of occurrences of that piece on one side.

The following are valid for each piece:

  • * (zero or more of that piece).
  • + (one or more of that piece).
  • d (exactly d occurrences of that piece, where d is a digit).
  • d+ (d or more occurrences of that piece).
  • d- (d or fewer occurrences of that piece).



is the starting material position, and QR+B*N*P7- represents material in which we require at least one pawn to be missing from one side and they should have a Queen and Rook, but we don't care about the minor pieces.

In addition, some extra notation is available to specify material relative to the opponent's. These are placed after the piece letter to which they refer.

  • = (the number of these pieces must be the same as the opponent's).
  • # (the number of these pieces must be different the opponent's).
  • > (the number of these pieces more than the opponent has).
  • < (the number of these pieces less than the opponent has).


R+P+ R=P#

looks for Rook and Pawn games with an equal number of Rooks but unbalanced pawns.

In addition > and < may be preceded by a digit:

  • d> (the number of these pieces must be at least d more than the opponent's).
  • d< (the number of these pieces must be at least d less than the opponent's).

Two more notations, >=, <= may be preceded by an optional digit (the default is 1). The meaning of this may not be intuitively obvious and, to an extent, they represent a notational compromise.

  • d>= (the number of these pieces must be exactly d more than the opponent's).
  • d<= (the number of these pieces must be exactly d less than the opponent's).

In this example, both sides have a pair or Rooks but one has exactly one pawn more than the other:

r2p* r=p1>=

Here is an example where one side has sacrificed a Rook and Pawn for Knight and Bishop and we don't care whether Queens are on or off the board, so long as they are balanced:

q*r+n*b*p+ q=r<n>b>p1<

This example represents some of the imprecision that can occur with matches. The meaning of 'r<' is such that this could match positions in which one side as 2 Rooks and the other none. This can be corrected with:

q*r+n*b*p+ q=r1<=n>b>p1<

enforcing strictly one Rook less. We ought also to correct the same problem with the minor pieces:

q*r+n*b*p+ q=r1<=n1>=b1>=p1<

In practice, we probably want to allow general matching of minor pieces so the letter 'L' may be used to stand for a minor piece (Bishop or Knight). This example represents a similar sacrifice of Rook and Pawn for two minor pieces.

q*r+l*p+ q=r1<=l2>=p1<

I would advise against mixing the minor piece letter with Knight and Bishop letters in the piece set for a single side, however, as I am not convinced that it will produce exact results.

  • Position Stability with -z

    The piece sets may be preceded by an optional number indicating the required stability of the position. Normally, if you are looking for a position with a particular set of material characteristics then you probably want that position to last for a reasonable number of moves in order to study its characteristics. The number before the piece sets is how many half-moves you wish that material balance to last. By default, this has a value of 2 so that fleeting positions in the middle of pairs of exchanges do not produce unwanted matches. This example looks for double-Rook and pawn games that last at least 10 half-moves:

    10 R2P+ R=P*

The Seven Tag Roster (-7 or --seven)

This flag discards tag pairs that are not part of the Seven Tag Roster:

Event, Site, Date, Round, White, Black and Result.

However, if the original game included a FEN tag, this is included in the output, as the moves will make no sense otherwise. In addition, if the -e flag has been used for ECO classification, any ECO, Opening, Variation and SubVariation tags are also output.

User-defined tag roster ordering (-R)

The -R flag makes it possible for to define the order in which tags for a game are listed in the output. The flag should be immediately followed by the name of a file that contains a list of tag names, one per line, for instance:

pgn-extract -Rroster file.pgn

where roster might contain:

% Output the tags of the seven tag roster alphabetically.

The '%' character may be used to include comments in the file. Tags not listed in such a file will appear after the required tags have been output.

Include a position evaluation after each move (--evaluation)

The --evaluation argument causes a comment to be appended to every move, which contains an evaluation of the position immediately following that move. The default evaluation is a simplified version of Shannon's board evaluation. In this case, the evaluation is the difference between the value of White's position and Black's, where the value of a position is a weighted sum of the pieces plus a multiplier (0.1) applied to the number of available moves for that player.

I see this primarily as being a hook for people who wish to embed their own evaluations in the output. See the evaluate function in apply.c if you wish to write your own.

Include a comment with a FEN string for the position after each move (--fencomments)

The --fencomments argument causes a comment to be appended to every move, which contains a FEN string for the position immediately following that move. See -F for adding a comment after just the final move.

Don't output ep squares in FEN when the capture is not possible (--nofauxep)

FEN descriptions include the square for a possible en passant capture regardless of whether there is actually an opposing pawn in position to make the capture. For instance, if there is no opposing pawn, or the capture would leave the capturing side in check. The --nofauxep flag suppresses output of the square when a capture is not possible. This makes it easier to compare identical FEN positions resulting from transpositions.

Add a game comment on positional, material, repetition and fifty-move rule matches (--markmatches)

Add a game comment with the text immediately following --markmatches after the move which causes a positional, material, repetition or fifty-move rule match. For instance:

pgn-extract -xvars --markmatches MATCH file.pgn

would add the comment { MATCH } after every move that caused a match from the positional matches specified in the vars file.

See -x for positional matches with moves, -t for positional matches with FEN patterns, and -z for material matches. See also --fifty and --repetition.

Add a Tag containing a hashcode for the game (--addhashcode)

Add the tag HashCode to the tags. This contains a hashcode value generated from the moves of the game. Identical move sequences will produce the same hash code.

Correct Result tags that conflict with the game outcome (checkmate or stalemate) (--fixresulttags)

Where the existing Result tag clearly conflicts with a game's outcome - i.e., checkmate or stalemate - replace the value in the Result tag to match the game outcome.

Add a Tag containing the total ply count (--totalplycount)

Add the tag TotalPlyCount to the tags. This contains a count of the number of ply present in the game being output. Unless variations have been suppressed this will include all moves in variations as well as the main line.

Contacting the author

I don't run a proper mailing list but if you find the program useful and would like or to offer suggestions that you think others might be interested in, then drop me a line at d.j.barnes@kent.ac.uk


The moves, variations, and commentary of each game are held internally and reformatted when a game is extracted, rather than reproducing the original text of the game source.

Lower-case 'b' as the first character of a move is taken to be a move of the b-pawn if one to match the move can be found. Otherwise, Bishop moves are tried as an alternative. There is no back-up on failure if picking a valid pawn move was the wrong choice.

Lower-case 'b' as the first character of a Bishop move is not acceptable in the variations files.

Duplicate detection is not guaranteed to be exact. The -Z flag has slightly more potential to avoid false duplicates as it compares separate values for the end position and move sequence, whereas these are XORed to save space when -Z is not used. However, this will only make a difference and avoid false matches if two different games at the same hashtable index also produce identical XORed values.

The results of the -x, -v, and -t/-T search criteria are AND-ed together. There might be occasions when you wanted to search for games that matched either positional variations or textual variations at the same time, for instance. This requires multiple runs of pgn-extract.

The -Wsan variation that allows selection of the output language is tied to single-character piece descriptions. This does not support Russian usage, for instance, in which the King is described as a character pair.

The files

The sources include a Makefile for the GNU make program, gmake. I also use this with the Minimalist GNU for Windows compiler to produce a Windows command-line executable (see Portability).

The distribution comes with the following files.

COPYINGGNU General Public License
MakefileA build file suitable for use with the GNU make utility.
Windows users might like to use the mingw - Minimalist GNU for Windows version.
apply.[ch]functions concerned with applying moves to a board.
argsfile.[ch]functions concerned with command line argument processing.
bool.hBoolean type definition.
decode.[ch]functions for decoding the text of a move.
defs.hdefinitions relating to boards.
eco.[ch]functions for looking up ECO classifications.
eco.pgnPGN file of ECO classifications.
end.[ch]functions for looking for matching endgames.
fenmatcher.[ch]pattern matching for the FENPattern pseudo tag.
grammar.[ch]the parser.
hashing.[ch]duplicate detection hash tables.
help.htmlThis file.
lex.[ch]the lexical analyser.
lines.[ch]functions for reading lines.
lists.[ch]functions for holding the extraction criteria.
main.cthe program starting point.
map.[ch]functions for implementing move semantics.
moves.[ch]functions for collecting moves and variations.
mymalloc.[ch]functions for memory allocation.
output.[ch]functions concerned with outputting the games.
taglines.[ch]functions for processing files of tag search criteria.
taglist.hconstants for tag and pseudo-tag names
tokens.htype definition for lexical tokens.
typedef.htype definitions.
zobrist.[ch[code for generating Zobrist hash values for board positions.


pgn-extract is regularly used under Windows/DOS (using Minimalist GNU for Windows), and various versions of Linux and Mac OS X.


I would like to thank all those who used the program and made suggestions for things to add. In particular, thanks to Michael Kerry whose help led to better determination of game boundaries in earlier versions, and V. Armando Sole whose own filter program was the inspiration for adding textual variation permutations. John Brogan suggested adding the ! notation to the variation file and provided the spur for duplicate detection. He also supplied the original code for soundex matching (-S).

Jaroslav Poriz, Ron Leamon, Ed Leonard, Charles Frohman, and Robert Wilhelm helped with testing at various times. Bernhard Maerz was instrumental in encouraging the inclusion of ECO classification and material balance matches. He and Peter Otterstaetter suggested the relational operators in tag files, with Peter also providing the spur to make duplicate detection work with bigger game files (-Z) and doing some very useful testing for me.

Kayvan Sylvan requested long algebraic output and identified an error in ECO classification. Cameron Hayne suggested matching on the number of moves in a game. Owen D. Lyne suggested extension of the -E flag, and both tested and provided diagnostic data to help refine the ECO classification aspects of the program. Karl-Martin Skontorp provided the incentive and testing help that enabled me to add the -Wepd option.

FEN pattern matching is based on pattern matching code by Rob Pike. Taken from: http://www.cs.princeton.edu/courses/archive/spr09/cos333/beautiful.html and ideas from Kernighan and Plauger's "Software Tools".

Finally, thanks, of course, to Steven Edwards for his work on developing the PGN standard.


pgn-extract: a Portable Game Notation (PGN) extractor.
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 1, or (at your option) any later version.

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.

David Barnes may be contacted as d.j.barnes @ kent.ac.uk, @kentdjb on Twitter, or via https://www.cs.kent.ac.uk/people/staff/djb/

Change history

  • 19th February 2017: Extended --selectonly and added --skipmatching after a suggestion by Folkert van Heusden.
  • 16th February 2017: Fixed runtime failure when -x and -n were used together. Added missing SetUp tag when a FEN tag is present. Correct "1/2" in a Result tag to "1/2-1/2". Added explicit Q promotion when unspecified. Fixed lack of output with -W (i.e., output as source).
  • 29th January 2017: Added --addhashcomments to add a comment containing a polyglot hash code after each move. Added -H for Zobrist polyglot hash matches.
  • Added text after -F for FEN-position output at arbitrary positions and not simply at the end of a game. Suggested by Josip Salai.
  • 22nd January 2017: Added --stopafter.
  • 20th January 2017: Added --nosetuptags and --onlysetuptags following a discussion with Mike Crockett. Added -p[elu] for finer control of game length matching than provided by -b.
  • 14th January 2017: Fixed --fuzzydepth.
  • 9th January 2017: Improved the efficiency of (lack of) stalemate detection.
  • 1st January 2017: Added --fixresulttags for conflicts between the game termination and the result tag. Source code rewrites of function names for more consistent naming.
  • 8th November 2016: Allows 'b' as a trailing promotion character, previously disallowed.
  • 3rd September 2016: Added --quiescent for position quiescence after a suggestion from Mike Crockett.
  • 17th April 2016: Added --tagsubstr for substring matches on tags.
  • 4th March 2016: Added --quiet to suppress the number of games processed, as an extension of -s. (Suggested by Norm Pollock.)
  • 21st November 2015: Added recognition of Chess960 FEN encodings and application of its castling rules.
  • 16th October 2015: Added --fifty, --repetition, and warnings about inconsistent results in tags on checkmate and stalemate after suggestions by Norm Pollock.
  • 6th May 2015: Fixed errors in the half-move clock on castling and pawn promotion, thanks to Brandon RichardWebster.
  • 23rd Mar 2015: Added --nofauxep after a suggestion by Norm Pollock.
  • 21st Mar 2015: Fixed off-by-one in move number output with -F when white-to-move.
  • 20th Mar 2015: Added -Wxlalg at the suggestion of Bruce Ramsey.
  • 8th Jan 2015: Fixed bug in game counting with -#. Suppressed games with null moves (--) in the main line.
  • 28th October 2014. Added --selectonly after a suggestion by Francis Steen.
  • 2nd September 2014. Corrected an error in the generation of hashcodes when a promotion is made.
  • 31st May 2014. Added --addhashcode.
  • 25th May 2014. Added --totalplycount for Erich Körber.
  • 5th March 2014. Added --keepbroken to allow broken games to be output. Added at the request of Mark Crowther primarily to deal with the problem of live recording where the kings are moved to the centre of the board at the end of a game and erroneously included in the score.
  • 6th September 2013. Corrected failure to 'or' together multiple dates with -T and -t.
  • 16th May 2013. Corrected an error in the whole-move number in FEN output, thanks to Vincent Fleuranceau.
  • 14th May 2013. Null move notation (--) in variations recognised.
  • 16th April 2013. Added --fuzzydepth. This is due to Owen D. Lyne who requested this functionality years ago - sorry for taking so long, Owen!
  • 11th April 2013. Added -Wuci.
  • 29th March 2013. Added --version.
  • 26th March 2013. Fixed crash when a string to be output is longer than the output line length.
  • 12th March 2013. Added long-form versions of -a, -c, -d and -o: --append, --checkfile, --duplicates and --output.
  • 9th February 2013. Added pattern matching based on FEN descriptions and --markmatches for JS.
  • 23rd December 2012. Added --fencomments for Tyler Eaves.
  • 2nd December 2012. Allowed 0 for --plylimit.
  • 22nd September 2008. Added --stalemate for Wieland Belka.
  • 15th September 2008. Added --nochecks and fixed -A so that it handles long-form arguments properly.
  • 22nd December 2007. Added --notags, --plylimit, --nomovenumbers and --noresults after a suggestion by Wieland Belka to be able to create opening books.
    Added --evaluation for Folkert van Heusden.
    Added --stalemate for Norm Pollock.
    Added calculation of the half-move clock to FEN strings.
    Most of the arguments taking filenames can now be separated from the filename with a space.
    Gradually adding long-form alternatives for arguments, e.g. --seven, --notags, etc.
  • 24th April 2007. Fixed a bug with mate annotation. Added the -M flag for checkmate matches, which is due to Richard Jones.
  • 19th October 2005. Added language-specific letters to -Welalg following a suggestion from Folkert van Heusden.
  • 1st May 2004: Fixed an error with ECO classification that was causing the file list to be out of sync.
  • 29th April 2004: Buffered game text before outputting it, so that trailing spaces on lines (which violate the PGN spec) can be deleted.
    Games with zero moves are now acceptable.
  • 26th April 2004: Slight modification to one of the hashing values made in order to try to avoid clashes in ECO matches. ECO matches now have a discretion of up to 6 half moves.
  • 13th February 2002: Added -Welalg as an output format following a suggestion from Rafal Furdzik.
  • 27th March 2001
    • Added output of EPD via -Wepd.
    • Fixed a long standing error in FEN castling rights. These were not being withdrawn if a Rook was captured on its home square. Pointed out by Karl-Martin Skontorp, who also provided the incentive to add -Wepd.
  • 26th April 2000 Added the -R flag for tag ordering.
  • 22nd April 2000 Completed implementation of -A to work with all flags.
  • 21st April 2000
    • Added the -F flag.
    • Added support for reading Russian source files.
  • 11th April 2000
    • Added the -A flag.
    • Extended usage of -Wsan to support output in different languages.
    • Usage of -e with -7 retains an ECO tag in matched games.
    • FEN tags with the -t flag are used as positional matches (equivalent to -x matches).
    • Non-standard tags are now retained in game output.
  • 12th January 2000 C compiler with Red Hat Linux 6 was no longer happy with static initialisations involving stdin, stdout and stderr. Changes made to lex.c and main.c to work around this. Pointed out by Mladen Bestvina.
  • 18th October 1999 Numbers greater than 3 allowed with -E, at the request of Owen Lyne.
  • 15th December 1997 Treat \r as WHITESPACE (for DOS files).
  • 8th June 1997 Added -b flag to set bounds on the number of moves in a game to be matched.
  • 2nd May 1997 Corrected small error when strings were not terminated properly. In tags, this resulted in the corrected tag ending in ]"] instead of "].
  • 17th February 1997 Added a little more error recovery.
  • 15th November 1996 Added -Z.
  • 23rd Sep 1996 It is no longer necessary to omit move numbers from the variations files (-v and -x). This makes it easier to cut and paste games of interest into these files.
  • 28th Jun 1996 It is no longer necessary to terminate the tag file (-t). Relational operators added in the tag file (-t). Added -E flag.
  • 7th May 1996 Corrected failure to make ECO classification when combined with -x. Added lalg and halg as long algebraic output formats.
  • 9th Oct 1995 Add -#
  • 25th Sep 1995: Default to reading stdin if no file arguments are provided.
  • 24th Jul 1995: Added setup from FEN tags.
  • 18th Jul 1995:
    • Added material balance matches with -z.
    • Added 'L' as a minor piece letter in ending files.
  • 14th Jul 1995: Made the order of arguments immaterial.
  • 5th Jul 1995:
    • Added ECO classification with -e.
    • Fixed false partial substring matches with -v, e.g. textual variation move Nc6 is now no longer matched by game move c6.
  • 22nd Mar 1995: Made permutation matching with -v the default and added -P to suppress it.
  • Jan 1995: Added -n and -L.
  • 17th Nov 1994: Liberated the program from using YACC and Lex.
  • 13th Oct 1994: Released test version with ChessMaster output.
  • 20th Sep 1994: Added move rewriting and -W flag.
  • 7th Sep 1994: Added -D flag.
  • 6th Sep 1994: Added -C and -V flags and soundex matching.
  • 5th Sep 1994:
    • Integrated the positional variation code from a separately developed program.
    • Added -N flag.
    • Added ! to the textual variation syntax.
    • Removed the writing to extract.pgn that was present in an earlier unreleased version.
    • Added -d flag.
  • 8th Jul 1994:
    • Added -o flag.
    • Discarded writing to standard output in DOS version because of extensive problems trying to make this work with redirected output. Instead, output is written to the file extract.pgn.
  • 6th Jul 1994: Added -7 flag.
  • 9th May 1994: Added -p flag for variation permutations.
  • 6th May 1994: Added * as a don't-care move in variations files.
  • 26th Apr 1994: Added the -t flag for files of extraction criteria.
  • 25th Apr 1994: Added the -T flag for extraction criteria.
  • 22nd Apr 1994: Added the -f flag for handling lists of PGN files.
  • 13th Apr 1994:
    • Cleaned up the game-length determination by reading/writing files in binary-mode.
    • Added -a flag for appending to existing .pgn files.
    • Added multiple input files.
    • Made verbose output the default behaviour.