pgn-extract:
A Portable Game Notation (PGN) Manipulator for Chess Games
Version 22-11 by David J. Barnes

Textual Tag Matching with -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. The -T flag has fewer options and tag-matching syntax does not fit well with command-line usage.

For most situations, use -t with a file of the criteria to be matched:

pgn-extract -t tags games.pgn

where tags is an arbitrary file name - you can use whatever name you wish. In the file are listed tag name and value pairs corresponding to the extraction criteria you wish to use. For textual matching, 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.

Tag matching may be either textual or numeric. Here we cover textual matching. For matching of numeric tags see Date and Elo Matching, Time Control Matching and Numeric Tag Matching.

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"]

To match anywhere within a tag include the --tagsubstr flag in the program's arguments. In this case, a match for "Karpov" would now match

[White "Anatoli Karpov"]

If you wish to limit the year in which those games were played you might list:

White "Tal"
Date "1962"

This is a purely textual prefix match rather than a numeric one.

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 whose Date tags start with the three listed years. Note from this example that comments may be included in the tag file by using an initial '%' character on a line..

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, the pseudo-tag Player may be used

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.

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.

Date and Elo Matching 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. For date matches:

Date < "1962"

would only match games played before 1962. Month and day values may also be included for additional granularity; e.g.:

Date > "2020.03.31"

Elo matches use a similar syntax:

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).

TimeControl Matching with -t

The -t option also supports very limited relational matching of two formats of the TimeControl tag: time+increment and moves+time. In both cases it is the time value of the control that is compared against a numerical value. For instance,

TimeControl >= "3600"

would match games with a TimeControl tag in which the time value is greater-than or equal to 3600 seconds. For instance,

[TimeControl "4500+60"]

and

[TimeControl "40/9000"]

The operators allowed are >, >=, <, <=, =, and <> (not equal to).

Only the first descriptor in a tag value is compared.

Numeric Tag Matching with -t

Matching of tags with numeric values may be performed with -t and a file of tag criteria. All of the relational operators are available: <, <=, >, >=, = and <> (not equal to). In the tag file associated with -t flag the value to be compared against a tag value is inserted between the name of the tag and the value against which a game's value is to be compared. For instance, given a pseudo-tag Difficulty, the following would find all games in which the value of the Difficulty tag was greater-than or equal-to 3:

Difficulty >= "3"

Note that quotes are required around the numeric value.

Multiple relational matches of a single tag are and-ed rather than or-ed. That means that games may be selected based on a range for a particular tag. For instance:

Difficulty >= "3"
Difficulty <= "6"

would find only games with Difficulty tag values of 3, 4, 5 or 6.

Regular Expression Tag Matching with -t

The operator =~ may be used to compare a tag value with a regular expression (details to be added). For instance:

White =~ "^F.*r$"

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.

The --btm and --wtm flags can be used to constrain the match by whether it is 'black to move' or 'white to move'. Otherwise, the player to move information is ignored.

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 and -Tf 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.

The pattern may be optionally followed by arbitrary text to serve as a pattern identification. If used with the --addlabeltag program argument then a MatchLabel tag will be added to the game containing the corresponding label. For instance:

FENPattern "?????rk?/?????pbp/*p?/*/*/8/PPP*/??KR????" castled

would add the tag pair:

[MatchLabel "castled"]

for a match against Black kingside castled behind a fianchetto and White queenside castled.

FENPatternI is a variation of FENPattern that will attempt two possible matches of a pattern: as written and inverted to swap the colours. This is useful if you are interested in finding games with structural properties for either side. If the pattern has an associated label and --addlabeltag is used then a match of the inverted version of the pattern will add the suffix 'I' to the label. For instance:

FENPatternI "?????rk?/?????pbp/*p?/*/*/8/PPP*/??KR????" castled

would add the tag pair:

[MatchLabel "castledI"]

if the inverted match finds a game in which White has castled kingside behind a fianchetto and Black has castled queenside.

See also -Tf for a simpler command-line version of FEN pattern matching.

--fenpattern and --fenpatterni: Match games reaching a position matching the associated FEN pattern

The --fenpattern and --fenpatterni command-line arguments are followed by a single string containing a FEN pattern to be matched. These arguments may appear multiple times. See -t with FENPattern for details of the patterns.

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.
• -TfFENPattern - Extract games matching the given FEN pattern.
• -ThHashCode - Extract games with HashCode designation HashCode.
• -TpPlayer - Extract games where Player has either colour.
• -TrResult - Extract games with result Result (1-0, 0-1 or 1/2).
• -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 Matching with -T
  

  A simple form of relational date matching is available. A date 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 and the day or month/day values are optional.

  So,

  pgn-extract -Tdb1962.01 file.pgn
  

  will look for games played before 1st January 1962. A fuller capability is available in tag files with the -t flag.

• FEN Pattern matches with -T
  

  The -Tf flag allows a form of position matching similar to that available with -t with FENPattern. A FEN board position will be used as a position to be matched. Note that castling rights and who is to move should not be included. The syntax of the pattern allows for wildcard character, documented below. For instance:

  pgn-extract -Tf"*/*/*/*/???PP???/*/*/*"
  

  would match any position on which White occupies both e4 and d4 with pawns, regardless of the position of anything else. Note that the use of wildcard characters will almost certainly require the use of double-quote characters around the FEN pattern to escape them from interpretation by the operating system's command-line interpreter.

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.
:-TpTal
% Where to output the matched games.
:-otalgames.pgn

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.
:-TpTal
% Where to output the matched games.
:-otalgames.pgn
% The game files to be read.
:caro.pgn

and the command invoked as simply:

pgn-extract -Aargs

The -t, -v, -x, -y, -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:

:-ttags
:-vvariations
:-xpositions
:-zendings
:-Rroster

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:

:-TpTal

we could say:

:-t
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:

:-t
White "Tal"
Result "1-0"
:-otalwins.pgn
:-x
e4 c6 d4 d5 exd5 cxd5
% Which game files to process.
:caro.pgn

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.

Suppressing games with inconsistent results

By default, games in which the Result tag conflicts with the terminating result indication are retained, with an error message being output. Some obvious corrections are possible with --fixresulttags but games in which the inconsistency cannot be resolved may be suppressed with --nobadresults.

Outputting selected matched games (--selectonly)

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

1:10,15,89:94

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:

1:10,15,89:94

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.

Output each variation as a separate game (--splitvariants)

The --splitvariants flag will output each variation of a game as a separate game. The headers of the containing game are reproduced for each variation, except for the Result tag, which is replaced with "*" to indicate that it is not necessarily a complete game.

The flag takes an optional positive integer to limit the depth of variants output as separate games. For instance: --splitvariants 1 will only output separate variant games for top-level variants. Others are suppressed from the output. A value of 0 is used to output all variants and may be omitted.

The --splitvariants flag cannot be used with the -V flag.

Start matching from game number N (default 1) (--firstgame)

The --firstgame flag takes a single numerical argument N (N >= 1) to request that matching only starts from game N, which is 1 by default. Games before game N are skipped for matching purposes.

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 any tags (--notags)

The tags for a game will not be output.

Suppress a single tag (--detag)

All occurrences of the named tag will be omitted from the output of games. For instance:

pgn-extract --detag Annotator --nocomments games.pgn

would remove all Annotator tags and comments from the given games.

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. The -V flag cannot be used with the --splitvariants flag.

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.)

Drop opening ply before a matching comment string (--dropbefore)

1. e4 e5 { drop } 2. Nf3 Nc6 3. Bb5 *

pgn-extract --dropbefore drop

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "rnbqkbnr/pppp1ppp/8/4p3/4P3/8/PPPP1PPP/RNBQKBNR w KQkq e6 0 1"]

2. Nf3 Nc6 2. Bb5 *

pgn-extract --dropply 2 file.pgn

pgn-extract --dropply -1 file.pgn

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, -p, --minply, --maxply, --minmoves, --maxmoves)

The --minply, --maxply, --minmoves and --maxmoves are each followed by a number specifying the minimum (maximum) number of ply (moves) required in a game to match. The minimum and maximum options may be combined for imposing both lower and upper bounds. These long-form arguments perform a similar role to the alternative older single-character arguments described immediately below.

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

Note that the initial number of ply will be set from a FEN tag, if present, and therefore might not necessarily match the number of moves present in a particular game.

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 or seventy-five moves with no capture or pawn move (--fifty, --seventyfive, --50, --75)

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

The --seventyfive flag does the save for seventy-five moves. Shorthand alternatives are --50 and --75.

See --markmatches for a way to mark the positions at which matches occur.

Matching only games that include three-fold or five-fold repetition (--repetition, --repetition5)

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

pgn-extract --repetition file.pgn

The --repetition5 flag requests only games that include five-fold repetition.

In both cases, 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

Match 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

  -emy_eco_codes.pgn
  

  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. unless a second number is given (see below). Any existing contents of these files are always overwritten on each run of pgn-extract.

pgn-extract -#250 file.pgn

will 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.

If the number of games per file is followed by a comma and a second unsigned integer, then the second number is used as the name of the first output file. For instance,

pgn-extract -#1,100 file.pgn

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. Note that Chess960 games must have the Variant tag set to "chess 960" to be recognised as such. This affects the encoding of castling moves when output in long-algebraic format.

• 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.
• -Wxolalg is a variation on -Wxlalg that outputs O-O and O-O-O for the castling moves.
• 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.
  This format is compatible with my UCI-analyser and could be used as part of a process to annotate games with engine analysis.

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. A c1 comment contains the game's result.

-Wfen outputs the game as a sequence of FEN positions along with the game tags but no result. The tags may be suppressed with the --notags option. Note that the output from this option is not valid PGN and it has been provided solely as a convenient way to obtain the FEN sequence for a game.

-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.

Output each comment on a separate line

The --commentlines flag will break game output at the start and end of a comment so that comments appear on separate lines from the game text.

Include a comment with a game's line numbers from the input file

The --linenumbers argument is followed by a marker string and the result is that a comment is added to each matched game between the tags and the moves. The comment contains the marker string and the start and end line numbers of where the game originated from in its input file. For instance:

pgn-extract --linenumbers marker games.pgn

would include comments such as { marker:1:20 } and { marker:22:35 }.

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:

:-t
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 (-y, -z, --materialy, --materialz and --addmatchtag)

The -y and -z flags take a filename of material balances for which you wish to search in games. The --materialy and --materialz arguments offer command-line alternatives to using a file.

The basic structure of the file is one or more lines describing a material balance 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 a game in which a lone Rook and Pawn are competing against a lone Knight and Bishop for the other.

The case of the letters is immaterial and the order of the pieces does not matter. There is no need to include Kings in the description except for the special case where pieces1 would otherwise be empty. For instance, to find matches where Black has a single pawn and White only has a King the description would be

k p

Apart from Kings, if a piece letter is not listed for a side then that piece is not present within that side's material. With the -z and --materialz flags, a match will be tested for from both White and Black's point of view, so with -z the example above matches the same games as:

bn pr

However, if this is used with the -y or --materialy flag then only games in which Black has the Rook will be searched for.

Further 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).

So:

QR2B2N2P8 QR2B2N2P8

is the starting material position.

Further 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 from the opponent's).
• > (the number of these pieces more than the opponent has).
• < (the number of these pieces less than the opponent has).

So,

R+P+ R=P#

looks for Rook and Pawn games with a non-zero 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.

Given the following pattern:

q*r*b2n0p* q=r=b0n2p=

the -y flag would find all games where White had a Bishop pair and Black had two Knights, whereas the -z flag would find all BB vs NN games for both sides. With both --materialz and --materialy the full pattern must be supplied as a single command-line string; for instance:

pgn-extract --materialz "q*r*b2n0p* q=r=b0n2p=" games.pgn

The --addmatchtag argument can be used with -z or --materialz to add a MaterialMatch tag pair to a game. The associated string will be "White" if pieces1 matches the White pieces or "Black" if pieces1 matches the black pieces. If this argument is used with -y or --materialy then the associated string will always be "White" so there is no point combining --addmatchtag with -y.

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.

• 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 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.
Black
Date
Event
Result
Round
Site
White

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.

See --xroster for how to suppress tags not included via the -R option.

Don't output tags not included with the -R option (--xroster)

When combined with the -R option, --xroster will suppress from games all tags that are not included in the roster ordering file. For instance, if the file roster contains:

White
Black
Result

then

pgn-extract -Rroster --xroster file.pgn

would only output the White, Black and Result tags (in that order) for each game.

Beware that potentially significant tags such as FEN and SetUp will be suppressed unless explicitly included in the roster file.

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.

However, an alternative approach to annotating games with engine analysis would be to consider using something like my UCI-analyser that can pass suitably formatted PGN files to a UCI-compatible engine. See, for instance, the section on obtaining annotated output in PGN format.

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.

Include a polyglot hashcode comment after each move (--hashcomments)

After every move a 16-character hexadecimal polyglot hashcode is output in a comment. The idea is to make it easy to identify the hashcode associated with a position, for instance for use with hashcode searches (see -H).

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.

If the text FEN is used then this is interpreted to mean that the FEN encoding of the match position should be output in the comment, rather than the literal text FEN.

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.

Add potentially missing castling rights to FEN tags (--addfencastling)

If no castling rights are indicated in the FEN tag of a game (-) then liberally interpret the initial positions of Kings and Rooks to infer the rights. NB: This is not implemented for Chess 960 positions.

Correct Result tags that conflict with the game outcome or terminating result (--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 and don't report the inconsistency.

Inconsistencies between the result tag and terminating result are resolved in favour of the terminating result when either is "*".

Attempt to correct tag strings that are not properly terminated (--fixtagstrings).

Tag strings sometimes contain extra, unescaped quote characters within them. This option attempts to fix those errors.

Add a PlyCount tag (--plycount)

Add the tag PlyCount 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.

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

Add the tag TotalPlyCount to the tags. This contains a count of the total 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.

Allow nested comments (--nestedcomments)

Nested comments are not usually allowed, but some game sources include them. Use --nestedcomments to avoid mismatched closing comment symbols.

Single-line comments

Single line comments introduced via a semicolon (;) character are recognised. They are converted in the output to multi-line comments with curly braces. Curly braces within them are converted to square brackets.

Move lichess comments (--lichesscommentfix)

Move comments that occur immediately after the opening parenthesis of a variation (i.e., before the variation's first move) to after the first move. This was introduced to get around a feature of lichess studies.

Contacting the author

I don't run a 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

Limitations

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.

Portability

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

Acknowledgements

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.

License

pgn-extract: a Portable Game Notation (PGN) extractor.
This file is part of pgn-extract: a Portable Game Notation (PGN) extractor. Copyright (C) 1994-2022 David J. Barnes

pgn-extract 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 3 of the License, or (at your option) any later version.

pgn-extract 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 pgn-extract. If not, see http://www.gnu.org/licenses/.

David J. 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

Historical log of major changes.