pkfix-helper man page

pkfix-helper — preprocess dvips-produced PostScript documents before passing them to pkfix

Synopsis

pkfix-helper [--help] [--verbose] [--force=name=fontspec] [--ps=filename.ps] [--tex=filename.tex] [--cache=filename] [--include=fontspec] [--exclude=regexp] [--keep=fontspec] [--quiet] [--no-repeats] [--spp=number] [input.ps [output.ps]]

Description

Motivation

PostScript documents created with old versions of dvips almost invariably utilize bitmapped (PostScript Type 3) fonts. The problem with bitmapped fonts is that they target a specific device resolution; a PostScript file produced using 300 DPI fonts will look grainy on a 600 DPI printer. Even worse, all bitmapped fonts look grainy when zoomed in on screen. The solution is to use vector (PostScript Type 1) fonts, which are resolution-independent and appear crisp at any size or scale.

While it is no longer difficult to configure dvips to use vector fonts, it is not always possible to rerun dvips on an old .dvi file. The .dvi file and document source may have been lost; or, the source may no longer compile because packages it depends upon may no longer be available.

Heiko Oberdiek's pkfix script replaces bitmapped fonts in dvips-produced PostScript files with the corresponding vector fonts. It works by parsing the PostScript comments with which dvips surrounds bitmapped-font definitions. For example, a font definition beginning with the comment "%DVIPSBitmapFont: Fi cmss10 11 28" and ending with a matching %EndDVIPSBitmapFont is known to define font "Fi" as "cmss10" (Computer Modern Sans Serif at a design size of 10 points) scaled to 11 points. Only the 28 characters actually used by the document are defined. pkfix then replaces the font definition with one that defines "Fi" using the same set of characters but taken from the cmss10.pfb vector font file.

Unfortunately, pkfix works only with versions of dvips newer than v5.58 (ca. 1996). Naturally, the older a PostScript document, the less likely its sources still exist and can still be recompiled. Older versions of dvips lack %DVIPSBitmapFont comments and various other PostScript comments on which pkfix relies. Without PostScript comments to guide it, pkfix is unable to determine which vector fonts correspond with which bitmapped fonts.

Overview

The pkfix-helper script is a preprocessor for pkfix that attempts to determine the association between each document-font name (e.g., "Fi") in a PostScript file and the original font (e.g., "cmss10") and fonts size (e.g., 11 points). It then fabricates the PostScript comments that pkfix expects to see so that pkfix can do its job.

pkfix-helper works by comparing every document font against every .tfm font file it knows about (assuming that each such font has a corresponding .pfb vector version) and selecting the best matching .tfm file for every document font. pkfix-helper has access only to the widths of characters and only to those characters actually used in the document. Also, the program recognizes only a limited set of the most popular .tfm files and scaling factors. Consequently, the comparison is imperfect and pkfix-helper may attribute an incorrect font to a given name. Fonts comprising only one or two characters actually used in a document are particularly problematic for pkfix-helper because many fonts may be near-enough matches to fool the problem.

pkfix-helper is designed so that a user can guide the font-selection process by manually designating matching fonts. With a modicum of diligence and patience a user can correct any mismatched fonts and help the program provide proper input to pkfix.

Options

pkfix-helper accepts on the command line the filename of a PostScript document to process (with the default being the standard input device) and the filename of a modified PostScript document to create (with the default being the standard output device). The program also accepts the following command-line options:

Frequently Used Options

-h, --help
Display usage information and exit. The --verbose and --quiet options can be used to increase and decrease the amount of information presented.
-v, --verbose
Increase the amount of status information that pkfix-helper displays as it runs. Additional instances of --verbose on the command line further increase the program's verbosity. By default, only major operations are displayed. A single --verbose additionally displays information about individual font comparisons. A second --verbose additionally displays details about some of the program's internal operations.
-f name=fontspec, --force=name=fontspec
Force pkfix-helper to associate a specific font with a given font name appearing the document. name is a two-character dvips font name such as "Fa". fontspec is a font specification such as "cmmi8" or "cmsy10 @ 1.1X". An asterisk used in the name of the base font (e.g., "cmti*") will automatically try all integral test font sizes from 5 to 17 points ("cmti5", "cmti6", ..., "cmti17"). An asterisk used as a scale value (e.g., "cmsy10 @ *") will be replaced by the scale value that gives the best match to the original font's metrics. The --force option can be specified repeatedly on the command line.
-p filename.ps, --ps=filename.ps
Create a PostScript file called filename.ps that shows the dvips name and a font sample of every font used by the input document.
-t filename.tex, --tex=filename.tex
Create a Plain TeX file called filename.tex that shows the dvips name and a font sample of every font that pkfix-helper used in the output document.

Infrequently Used Options

-C filename, --cache=filename
Speed up TFM file processing by caching character metrics into file filename. On some systems it takes a long time to read a TFM file, spawn tftopl to convert it to PL format, and extract from the PL data the metrics for each character. The first time --cache is specified, pkfix-helper proceeds as normal then writes all of the extracted character metrics to filename. On subsequent runs in which --cache=filename is specified, pkfix-helper reads the previously extracted metrics from filename, going through the tftopl-based process only for TFM files that were not previously encountered.
-q, --quiet
Instruct pkfix-helper to produce no output during its run except for fatal error messages.
-1, --no-repeats
Prevent pkfix-helper from associating the same fontspec with more than one dvips font name.
-i fontspec, --include=fontspec
Add fontspec to the list of font specifications against which pkfix-helper compares every document font. (In contrast, --force designates a font specification to use only for a specific document font.) The --include option can be specified repeatedly on the command line.
-x regexp, --exclude=regexp
Remove all font specifications matching regular expression regexp from pkfix-helper's list of known fonts. The --exclude option can be specified repeatedly on the command line.
-k fontspec, --keep=fontspec
Do not substitute a vector font for bitmapped font fontspec ("Fa", "Fb", etc.). This is useful when converting documents that use obscure bitmapped fonts for which there is no vector equivalent. For example, it was somewhat common in the past to include graphics such as university or corporate logos into a document by converting the bitmapped image into a single-character font and using that font in LaTeX. --keep prevents such fonts from being replaced. The --keep option can be specified repeatedly on the command line.
-a, --any-scale
Allow any value to be used to scale a font when ""*"" is specified as the scale factor. Normally, pkfix-helper considers only integer multiples of 0.1 that are greater than or equal to 1.0 (i.e., fontname@1X, "@1.1X", "@1.2X", "@1.3X", etc.).
-s, --spp
Specify the number of font samples per page to print to the files indicated using the --ps and --tex options. The default value, 25, should work well in most circumstances.

Diagnostics

"Best match for name is rather poor"
The best font pkfix-helper found for dvips font name name has a mismatch value greater than or equal to 1.0. (The mismatch value is the sum of the squares of the difference between the character widths of a document font and a potential replacement font.) Use the --force option to designate an alternative replacement font or scaling amount.

Examples

For the purpose of the following examples, assume that oldfile.ps is the name of a PostScript file produced by an old version of dvips and utilizing at least one bitmapped font. It's always worth verifying that pkfix can't convert the file on its own:

$ pkfix oldfile.ps newfile.ps
PKFIX 1.3, 2005/02/25 - Copyright (c) 2001, 2005 by Heiko Oberdiek.
==> no fonts converted

(Alternatively pkfix may issue an error message such as "!!! Error: Parse error (@start parameters)!".) Only when pkfix can't replace bitmapped fonts with vector fonts is pkfix-helper needed. In its simplest form, pkfix-helper takes the name of an input file (oldfile.ps in this example) and the name of an output file (pkfix-oldfile.ps), which will have the same contents as the input file but serve as suitable input for pkfix:

$ pkfix-helper oldfile.ps pkfix-oldfile.ps
Reading oldfile.ps ... done.
Number of Type 3 fonts encountered: 10
Bitmapped fonts are typeset at 600 DPI.
Finding character widths ... done.
Reading TFM files ... done (103 TFMs in 193 scaling variations).
Matching fonts:
    Processing Fi ... done (cmr10 @ 1X, mismatch=0.11683).
    Processing Fa ... done (cmti10 @ 1X, mismatch=0.08892).
    Processing Fb ... done (cmr8 @ 1X, mismatch=0.07133).
    Processing Ff ... done (cmbx12 @ 1.2X, mismatch=0.02948).
    Processing Fh ... done (cmtt10 @ 1X, mismatch=0.06895).
    Processing Fd ... done (cmmi10 @ 1X, mismatch=0.03966).
    Processing Fj ... done (cmbx12 @ 1X, mismatch=0.03972).
    Processing Fe ... done (cmbx10 @ 1X, mismatch=0.00762).
    Processing Fg ... done (cmsy10 @ 1X, mismatch=0.00875).
    Processing Fc ... done (cmr6 @ 1X, mismatch=0.00284).
$ pkfix pkfix-oldfile.ps newfile.ps
PKFIX 1.3, 2005/02/25 - Copyright (c) 2001, 2005 by Heiko Oberdiek.
*** Font conversion: `cmti10' -> `CMTI10'.
*** Font conversion: `cmr8' -> `CMR8'.
*** Font conversion: `cmr6' -> `CMR6'.
*** Font conversion: `cmmi10' -> `CMMI10'.
*** Font conversion: `cmbx10' -> `CMBX10'.
*** Font conversion: `cmbx12' -> `CMBX12'.
*** Font conversion: `cmsy10' -> `CMSY10'.
*** Font conversion: `cmtt10' -> `CMTT10'.
*** Font conversion: `cmr10' -> `CMR10'.
*** Font conversion: `cmbx12' -> `CMBX12'.
*** Merging font `CMBX12' (2).
==> 10 converted fonts.
==> 1 merged font.

Although pkfix-helper tries to automate as much as possible the font-detection process, some fonts will invariably be incorrectly identified. The program outputs a warning message if it knows a match is bad but the lack of a warning message does not necessarily indicate that pkfix-helper did a good job. It is therefore strongly recommended that the user produce "before" and "after" font sheets:

$ pkfix-helper -q oldfile.ps pkfix-oldfile.ps \
  --ps=oldfonts.ps --tex=newfonts.tex
$ tex newfonts.tex
This is TeX, Version 3.14159 (Web2C 7.4.5)
(./newfonts.tex [1] )
Output written on newfonts.dvi (1 page, 1292 bytes).
Transcript written on newfonts.log.
$ dvips newfonts.dvi -o newfonts.ps
This is dvips(k) 5.92b Copyright 2002 Radical Eye Software (www.radicaleye.com)
' TeX output 2006.06.11:1636' -> newfonts.ps
<texc.pro><8r.enc><texps.pro>. <cmr6.pfb><cmsy10.pfb><cmbx10.pfb><cmbx12.pfb>
<cmmi10.pfb><cmtt10.pfb><cmr8.pfb><cmti10.pfb><cmr10.pfb>[1]

After running the preceding commands, oldfonts.ps shows samples of the fonts in oldfile.ps and newfonts.ps shows samples of the replacement fonts that pkfix-helper used to produce pkfix-oldfile.ps. Print oldfonts.ps and newfonts.ps and compare them carefully for incorrect fonts and sizes.

Suppose that the choice of "cmbx12 @ 1.2X" for font "Ff" looks wrong; say the characters look taller in oldfonts.ps than in newfonts.ps. This is where the trial-and-error stage begins. Let's hypothesize that "cmb12" is a better match than "cmbx12" but we don't know how much to scale the font. Fortunately, pkfix-helper allows "*" to be used as a scaling factor to tell the program to automatically detect an optimal scaling factor, even if doing so means choosing a nonstandard font size:

$ pkfix-helper oldfile.ps pkfix-oldfile.ps --force="Ff=cmb12 @ *"
Reading oldfile.ps ... done.
Number of Type 3 fonts encountered: 10
Bitmapped fonts are typeset at 600 DPI.
Finding character widths ... done.
Reading TFM files ... failed.
pkfix-helper: Unable to process user-specified TFM file "cmb12"

Oops, it looks like we don't have a cmb12.tfm file on our system. Let's try scaling up cmb10.tfm instead:

$ pkfix-helper oldfile.ps pkfix-oldfile.ps --force="Ff=cmb10 @ *"
Reading oldfile.ps ... done.
Number of Type 3 fonts encountered: 10
Bitmapped fonts are typeset at 600 DPI.
Finding character widths ... done.
Reading TFM files ... done (103 TFMs in 193 scaling variations).
Matching fonts:
    Processing Fi ... done (cmr10 @ 1X, mismatch=0.11683).
    Processing Fa ... done (cmti10 @ 1X, mismatch=0.08892).
    Processing Fb ... done (cmr8 @ 1X, mismatch=0.07133).
    Processing Ff ... done (cmb10 @ 1.5X, mismatch=0.00035).
    Processing Fh ... done (cmtt10 @ 1X, mismatch=0.06895).
    Processing Fd ... done (cmmi10 @ 1X, mismatch=0.03966).
    Processing Fj ... done (cmbx12 @ 1X, mismatch=0.03972).
    Processing Fe ... done (cmbx10 @ 1X, mismatch=0.00762).
    Processing Fg ... done (cmsy10 @ 1X, mismatch=0.00875).
    Processing Fc ... done (cmr6 @ 1X, mismatch=0.00284).

The match has definitely improved, although 15 pt. is certainly an odd size for a font. Then again, many documents do use nonstandard sizes so this may in fact be correct. The best way to verify is once again to produce, print, and compare a pair of font samples and iterate until all of the fonts look correct. Use one instance of --force for each font you want to alter.

Environment

pkfix-helper honors the following environment variables:

GS
The name of the Ghostscript interpreter (default: gs)
TFTOPL
The name of a utility for converting .tfm files to .pl files (default: tftopl)

Bugs

Even when pkfix-helper finds a perfect match (i.e., the correct font in the correct size) the mismatch value is still typically nonzero. The same error is probably what causes pkfix-helper sometimes to consider the wrong font as being a better match than the correct font. Suggestions for fixing these bugs are welcome.

Restrictions

pkfix-helper works only with PostScript files produced by dvips, not with arbitrary PostScript files. The program has not been tested with output from versions of dvips older than v5.490 (ca. 1992); output from older versions may or may not work. Only bitmapped fonts loaded by dvips can be analyzed, not bitmapped fonts loaded by embedded graphics.

pkfix-helper works by comparing character widths, not the actual glyphs. Consequently, it is misled by sets of fonts with similar character widths (at least for those characters used by a given document). As an extreme example, all Computer Modern Teletype fonts of a given design size (e.g., "cmtt10", "cmsltt10", and "cmitt10") use exactly the same widths for all characters. Human assistance is generally needed to guide pkfix-helper's font-matching procedures.

Notes

Files produced using the --tex option are Plain TeX files and therefore must be compiled with tex (or a variation such as pdftex, luatex, etc.), not with latex.

See Also

pkfix(1), dvips(1), tex(1), gs(1)

PostScript Language Reference, Third Edition. Published by Addison-Wesley, ISBN 0-201-37922-8, <http://www.adobe.com/products/postscrip…>.

Author

Scott Pakin, scott+pkfh@pakin.org

Info

2011-06-11 v1.4